rust/compiler/rustc_ast/src/mut_visit.rs

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

1780 lines
60 KiB
Rust
Raw Normal View History

//! A `MutVisitor` represents an AST modification; it accepts an AST piece and
2021-12-11 06:55:13 +00:00
//! mutates it in place. So, for instance, macro expansion is a `MutVisitor`
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
//! that walks over an AST and modifies it.
//!
//! Note: using a `MutVisitor` (other than the `MacroExpander` `MutVisitor`) on
//! an AST before macro expansion is probably a bad idea. For instance,
//! a `MutVisitor` renaming item names in a module will miss all of those
//! that are created by the expansion of a macro.
use crate::ast::*;
2019-02-06 17:33:01 +00:00
use crate::ptr::P;
2019-10-11 10:46:32 +00:00
use crate::token::{self, Token};
2019-02-06 17:33:01 +00:00
use crate::tokenstream::*;
use rustc_data_structures::flat_map_in_place::FlatMapInPlace;
use rustc_data_structures::stack::ensure_sufficient_stack;
use rustc_data_structures::sync::Lrc;
use rustc_span::source_map::Spanned;
2020-04-19 11:00:18 +00:00
use rustc_span::symbol::Ident;
use rustc_span::Span;
use smallvec::{smallvec, Array, SmallVec};
2019-02-06 17:33:01 +00:00
use std::ops::DerefMut;
use std::{panic, ptr};
2022-11-22 05:23:25 +00:00
use thin_vec::ThinVec;
2018-08-30 09:42:16 +00:00
pub trait ExpectOne<A: Array> {
fn expect_one(self, err: &'static str) -> A::Item;
}
impl<A: Array> ExpectOne<A> for SmallVec<A> {
fn expect_one(self, err: &'static str) -> A::Item {
assert!(self.len() == 1, "{}", err);
2018-08-30 09:42:16 +00:00
self.into_iter().next().unwrap()
}
}
pub trait NoopVisitItemKind {
fn noop_visit(&mut self, visitor: &mut impl MutVisitor);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
pub trait MutVisitor: Sized {
/// Mutable token visiting only exists for the `macro_rules` token marker and should not be
/// used otherwise. Token visitor would be entirely separate from the regular visitor if
/// the marker didn't have to visit AST fragments in nonterminal tokens.
const VISIT_TOKENS: bool = false;
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
// Methods in this trait have one of three forms:
//
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
// fn visit_t(&mut self, t: &mut T); // common
// fn flat_map_t(&mut self, t: T) -> SmallVec<[T; 1]>; // rare
// fn filter_map_t(&mut self, t: T) -> Option<T>; // rarest
//
// Any additions to this trait should happen in form of a call to a public
// `noop_*` function that only calls out to the visitor again, not other
// `noop_*` functions. This is a necessary API workaround to the problem of
// not being able to call out to the super default method in an overridden
// default method.
//
// When writing these methods, it is better to use destructuring like this:
//
// fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) {
// visit_a(a);
// visit_b(b);
// }
//
// than to use field access like this:
//
// fn visit_abc(&mut self, abc: &mut ABC) {
// visit_a(&mut abc.a);
// visit_b(&mut abc.b);
// // ignore abc.c
// }
//
// As well as being more concise, the former is explicit about which fields
// are skipped. Furthermore, if a new field is added, the destructuring
// version will cause a compile error, which is good. In comparison, the
// field access version will continue working and it would be easy to
// forget to add handling for it.
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_crate(&mut self, c: &mut Crate) {
noop_visit_crate(c, self)
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_meta_list_item(&mut self, list_item: &mut NestedMetaItem) {
noop_visit_meta_list_item(list_item, self);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_meta_item(&mut self, meta_item: &mut MetaItem) {
noop_visit_meta_item(meta_item, self);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_use_tree(&mut self, use_tree: &mut UseTree) {
noop_visit_use_tree(use_tree, self);
}
fn flat_map_foreign_item(&mut self, ni: P<ForeignItem>) -> SmallVec<[P<ForeignItem>; 1]> {
noop_flat_map_item(ni, self)
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn flat_map_item(&mut self, i: P<Item>) -> SmallVec<[P<Item>; 1]> {
noop_flat_map_item(i, self)
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_fn_header(&mut self, header: &mut FnHeader) {
noop_visit_fn_header(header, self);
2018-06-06 22:50:59 +00:00
}
fn flat_map_field_def(&mut self, fd: FieldDef) -> SmallVec<[FieldDef; 1]> {
noop_flat_map_field_def(fd, self)
}
fn flat_map_trait_item(&mut self, i: P<AssocItem>) -> SmallVec<[P<AssocItem>; 1]> {
noop_flat_map_item(i, self)
}
fn flat_map_impl_item(&mut self, i: P<AssocItem>) -> SmallVec<[P<AssocItem>; 1]> {
noop_flat_map_item(i, self)
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_fn_decl(&mut self, d: &mut P<FnDecl>) {
noop_visit_fn_decl(d, self);
2014-01-06 12:00:46 +00:00
}
2023-12-05 21:39:36 +00:00
fn visit_coroutine_kind(&mut self, a: &mut CoroutineKind) {
noop_visit_coroutine_kind(a, self);
2023-11-28 18:15:39 +00:00
}
fn visit_closure_binder(&mut self, b: &mut ClosureBinder) {
noop_visit_closure_binder(b, self);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_block(&mut self, b: &mut P<Block>) {
noop_visit_block(b, self);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn flat_map_stmt(&mut self, s: Stmt) -> SmallVec<[Stmt; 1]> {
noop_flat_map_stmt(s, self)
}
fn flat_map_arm(&mut self, arm: Arm) -> SmallVec<[Arm; 1]> {
noop_flat_map_arm(arm, self)
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_pat(&mut self, p: &mut P<Pat>) {
noop_visit_pat(p, self);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_anon_const(&mut self, c: &mut AnonConst) {
noop_visit_anon_const(c, self);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_expr(&mut self, e: &mut P<Expr>) {
noop_visit_expr(e, self);
}
/// This method is a hack to workaround unstable of `stmt_expr_attributes`.
/// It can be removed once that feature is stabilized.
fn visit_method_receiver_expr(&mut self, ex: &mut P<Expr>) {
self.visit_expr(ex)
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn filter_map_expr(&mut self, e: P<Expr>) -> Option<P<Expr>> {
noop_filter_map_expr(e, self)
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_generic_arg(&mut self, arg: &mut GenericArg) {
noop_visit_generic_arg(arg, self);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_ty(&mut self, t: &mut P<Ty>) {
noop_visit_ty(t, self);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_lifetime(&mut self, l: &mut Lifetime) {
noop_visit_lifetime(l, self);
}
fn visit_assoc_item_constraint(&mut self, c: &mut AssocItemConstraint) {
noop_visit_assoc_item_constraint(c, self);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_foreign_mod(&mut self, nm: &mut ForeignMod) {
noop_visit_foreign_mod(nm, self);
}
fn flat_map_variant(&mut self, v: Variant) -> SmallVec<[Variant; 1]> {
noop_flat_map_variant(v, self)
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_ident(&mut self, i: &mut Ident) {
noop_visit_ident(i, self);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_path(&mut self, p: &mut Path) {
noop_visit_path(p, self);
}
fn visit_qself(&mut self, qs: &mut Option<P<QSelf>>) {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
noop_visit_qself(qs, self);
2018-06-11 23:26:53 +00:00
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_generic_args(&mut self, p: &mut GenericArgs) {
noop_visit_generic_args(p, self);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_angle_bracketed_parameter_data(&mut self, p: &mut AngleBracketedArgs) {
noop_visit_angle_bracketed_parameter_data(p, self);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_parenthesized_parameter_data(&mut self, p: &mut ParenthesizedArgs) {
noop_visit_parenthesized_parameter_data(p, self);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_local(&mut self, l: &mut P<Local>) {
noop_visit_local(l, self);
}
fn visit_mac_call(&mut self, mac: &mut MacCall) {
noop_visit_mac(mac, self);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_macro_def(&mut self, def: &mut MacroDef) {
noop_visit_macro_def(def, self);
2017-03-17 21:58:48 +00:00
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_label(&mut self, label: &mut Label) {
noop_visit_label(label, self);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_attribute(&mut self, at: &mut Attribute) {
noop_visit_attribute(at, self);
}
fn flat_map_param(&mut self, param: Param) -> SmallVec<[Param; 1]> {
noop_flat_map_param(param, self)
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_generics(&mut self, generics: &mut Generics) {
noop_visit_generics(generics, self);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_trait_ref(&mut self, tr: &mut TraitRef) {
noop_visit_trait_ref(tr, self);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_poly_trait_ref(&mut self, p: &mut PolyTraitRef) {
noop_visit_poly_trait_ref(p, self);
2014-11-07 11:53:45 +00:00
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_variant_data(&mut self, vdata: &mut VariantData) {
noop_visit_variant_data(vdata, self);
}
fn flat_map_generic_param(&mut self, param: GenericParam) -> SmallVec<[GenericParam; 1]> {
noop_flat_map_generic_param(param, self)
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_param_bound(&mut self, tpb: &mut GenericBound) {
noop_visit_param_bound(tpb, self);
}
fn visit_precise_capturing_arg(&mut self, arg: &mut PreciseCapturingArg) {
noop_visit_precise_capturing_arg(arg, self);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_mt(&mut self, mt: &mut MutTy) {
noop_visit_mt(mt, self);
}
fn flat_map_expr_field(&mut self, f: ExprField) -> SmallVec<[ExprField; 1]> {
noop_flat_map_expr_field(f, self)
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_where_clause(&mut self, where_clause: &mut WhereClause) {
noop_visit_where_clause(where_clause, self);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_where_predicate(&mut self, where_predicate: &mut WherePredicate) {
noop_visit_where_predicate(where_predicate, self);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_vis(&mut self, vis: &mut Visibility) {
noop_visit_vis(vis, self);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_id(&mut self, _id: &mut NodeId) {
// Do nothing.
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn visit_span(&mut self, _sp: &mut Span) {
// Do nothing.
}
fn flat_map_pat_field(&mut self, fp: PatField) -> SmallVec<[PatField; 1]> {
noop_flat_map_pat_field(fp, self)
}
fn visit_inline_asm(&mut self, asm: &mut InlineAsm) {
noop_visit_inline_asm(asm, self)
}
fn visit_inline_asm_sym(&mut self, sym: &mut InlineAsmSym) {
noop_visit_inline_asm_sym(sym, self)
}
fn visit_format_args(&mut self, fmt: &mut FormatArgs) {
noop_visit_format_args(fmt, self)
}
fn visit_capture_by(&mut self, capture_by: &mut CaptureBy) {
noop_visit_capture_by(capture_by, self)
}
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
/// Use a map-style function (`FnOnce(T) -> T`) to overwrite a `&mut T`. Useful
/// when using a `flat_map_*` or `filter_map_*` method within a `visit_`
/// method.
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
//
// No `noop_` prefix because there isn't a corresponding method in `MutVisitor`.
pub fn visit_clobber<T: DummyAstNode>(t: &mut T, f: impl FnOnce(T) -> T) {
2019-02-22 02:36:04 +00:00
unsafe {
// Safe because `t` is used in a read-only fashion by `read()` before
// being overwritten by `write()`.
let old_t = ptr::read(t);
let new_t =
panic::catch_unwind(panic::AssertUnwindSafe(|| f(old_t))).unwrap_or_else(|err| {
// Set `t` to some valid but possible meaningless value,
// and pass the fatal error further.
ptr::write(t, T::dummy());
panic::resume_unwind(err);
});
2019-02-22 02:36:04 +00:00
ptr::write(t, new_t);
}
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
// No `noop_` prefix because there isn't a corresponding method in `MutVisitor`.
#[inline]
fn visit_vec<T, F>(elems: &mut Vec<T>, mut visit_elem: F)
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
where
F: FnMut(&mut T),
{
for elem in elems {
visit_elem(elem);
}
}
2022-11-22 05:23:25 +00:00
// No `noop_` prefix because there isn't a corresponding method in `MutVisitor`.
#[inline]
fn visit_thin_vec<T, F>(elems: &mut ThinVec<T>, mut visit_elem: F)
2022-11-22 05:23:25 +00:00
where
F: FnMut(&mut T),
{
for elem in elems {
visit_elem(elem);
}
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
// No `noop_` prefix because there isn't a corresponding method in `MutVisitor`.
#[inline]
fn visit_opt<T, F>(opt: &mut Option<T>, mut visit_elem: F)
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
where
F: FnMut(&mut T),
{
if let Some(elem) = opt {
visit_elem(elem);
}
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
// No `noop_` prefix because there isn't a corresponding method in `MutVisitor`.
fn visit_attrs<T: MutVisitor>(attrs: &mut AttrVec, vis: &mut T) {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
for attr in attrs.iter_mut() {
vis.visit_attribute(attr);
}
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
// No `noop_` prefix because there isn't a corresponding method in `MutVisitor`.
#[allow(unused)]
fn visit_exprs<T: MutVisitor>(exprs: &mut Vec<P<Expr>>, vis: &mut T) {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
exprs.flat_map_in_place(|expr| vis.filter_map_expr(expr))
}
// No `noop_` prefix because there isn't a corresponding method in `MutVisitor`.
fn visit_thin_exprs<T: MutVisitor>(exprs: &mut ThinVec<P<Expr>>, vis: &mut T) {
exprs.flat_map_in_place(|expr| vis.filter_map_expr(expr))
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
// No `noop_` prefix because there isn't a corresponding method in `MutVisitor`.
fn visit_bounds<T: MutVisitor>(bounds: &mut GenericBounds, vis: &mut T) {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
visit_vec(bounds, |bound| vis.visit_param_bound(bound));
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
// No `noop_` prefix because there isn't a corresponding method in `MutVisitor`.
fn visit_fn_sig<T: MutVisitor>(FnSig { header, decl, span }: &mut FnSig, vis: &mut T) {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_fn_header(header);
vis.visit_fn_decl(decl);
Use smaller def span for functions Currently, the def span of a funtion encompasses the entire function signature and body. However, this is usually unnecessarily verbose - when we are pointing at an entire function in a diagnostic, we almost always want to point at the signature. The actual contents of the body tends to be irrelevant to the diagnostic we are emitting, and just takes up additional screen space. This commit changes the `def_span` of all function items (freestanding functions, `impl`-block methods, and `trait`-block methods) to be the span of the signature. For example, the function ```rust pub fn foo<T>(val: T) -> T { val } ``` now has a `def_span` corresponding to `pub fn foo<T>(val: T) -> T` (everything before the opening curly brace). Trait methods without a body have a `def_span` which includes the trailing semicolon. For example: ```rust trait Foo { fn bar(); }``` the function definition `Foo::bar` has a `def_span` of `fn bar();` This makes our diagnostic output much shorter, and emphasizes information that is relevant to whatever diagnostic we are reporting. We continue to use the full span (including the body) in a few of places: * MIR building uses the full span when building source scopes. * 'Outlives suggestions' use the full span to sort the diagnostics being emitted. * The `#[rustc_on_unimplemented(enclosing_scope="in this scope")]` attribute points the entire scope body. * The 'unconditional recursion' lint uses the full span to show additional context for the recursive call. All of these cases work only with local items, so we don't need to add anything extra to crate metadata.
2020-08-12 21:02:14 +00:00
vis.visit_span(span);
2018-08-30 04:18:11 +00:00
}
// No `noop_` prefix because there isn't a corresponding method in `MutVisitor`.
fn visit_attr_args<T: MutVisitor>(args: &mut AttrArgs, vis: &mut T) {
match args {
AttrArgs::Empty => {}
AttrArgs::Delimited(args) => visit_delim_args(args, vis),
AttrArgs::Eq(eq_span, AttrArgsEq::Ast(expr)) => {
vis.visit_expr(expr);
vis.visit_span(eq_span);
}
AttrArgs::Eq(_eq_span, AttrArgsEq::Hir(lit)) => {
unreachable!("in literal form when visiting mac args eq: {:?}", lit)
}
}
}
// No `noop_` prefix because there isn't a corresponding method in `MutVisitor`.
fn visit_delim_args<T: MutVisitor>(args: &mut DelimArgs, vis: &mut T) {
let DelimArgs { dspan, delim: _, tokens } = args;
visit_tts(tokens, vis);
visit_delim_span(dspan, vis);
}
pub fn visit_delim_span<T: MutVisitor>(DelimSpan { open, close }: &mut DelimSpan, vis: &mut T) {
vis.visit_span(open);
vis.visit_span(close);
}
pub fn noop_flat_map_pat_field<T: MutVisitor>(
mut fp: PatField,
vis: &mut T,
) -> SmallVec<[PatField; 1]> {
let PatField { attrs, id, ident, is_placeholder: _, is_shorthand: _, pat, span } = &mut fp;
vis.visit_id(id);
visit_attrs(attrs, vis);
vis.visit_ident(ident);
vis.visit_pat(pat);
vis.visit_span(span);
smallvec![fp]
}
fn noop_visit_use_tree<T: MutVisitor>(use_tree: &mut UseTree, vis: &mut T) {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
let UseTree { prefix, kind, span } = use_tree;
vis.visit_path(prefix);
match kind {
UseTreeKind::Simple(rename) => visit_opt(rename, |rename| vis.visit_ident(rename)),
UseTreeKind::Nested { items, span } => {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
for (tree, id) in items {
vis.visit_id(id);
vis.visit_use_tree(tree);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
vis.visit_span(span);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
UseTreeKind::Glob => {}
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_span(span);
}
pub fn noop_flat_map_arm<T: MutVisitor>(mut arm: Arm, vis: &mut T) -> SmallVec<[Arm; 1]> {
let Arm { attrs, pat, guard, body, span, id, is_placeholder: _ } = &mut arm;
2019-08-14 01:22:51 +00:00
vis.visit_id(id);
visit_attrs(attrs, vis);
vis.visit_pat(pat);
2019-06-23 09:32:16 +00:00
visit_opt(guard, |guard| vis.visit_expr(guard));
2023-11-27 02:15:56 +00:00
visit_opt(body, |body| vis.visit_expr(body));
2019-03-30 22:54:29 +00:00
vis.visit_span(span);
smallvec![arm]
}
fn noop_visit_assoc_item_constraint<T: MutVisitor>(
AssocItemConstraint { id, ident, gen_args, kind, span }: &mut AssocItemConstraint,
vis: &mut T,
) {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_id(id);
vis.visit_ident(ident);
if let Some(gen_args) = gen_args {
2020-11-19 17:28:38 +00:00
vis.visit_generic_args(gen_args);
}
match kind {
AssocItemConstraintKind::Equality { term } => match term {
Term::Ty(ty) => vis.visit_ty(ty),
Term::Const(c) => vis.visit_anon_const(c),
},
AssocItemConstraintKind::Bound { bounds } => visit_bounds(bounds, vis),
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_span(span);
}
pub fn noop_visit_ty<T: MutVisitor>(ty: &mut P<Ty>, vis: &mut T) {
let Ty { id, kind, span, tokens } = ty.deref_mut();
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_id(id);
2019-09-26 16:25:31 +00:00
match kind {
TyKind::Err(_guar) => {}
TyKind::Infer | TyKind::ImplicitSelf | TyKind::Dummy | TyKind::Never | TyKind::CVarArgs => {
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
TyKind::Slice(ty) => vis.visit_ty(ty),
TyKind::Ptr(mt) => vis.visit_mt(mt),
TyKind::Ref(lt, mt) => {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
visit_opt(lt, |lt| noop_visit_lifetime(lt, vis));
vis.visit_mt(mt);
}
TyKind::BareFn(bft) => {
2024-05-17 17:17:48 +00:00
let BareFnTy { safety, ext: _, generic_params, decl, decl_span } = bft.deref_mut();
visit_safety(safety, vis);
generic_params.flat_map_in_place(|param| vis.flat_map_generic_param(param));
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_fn_decl(decl);
vis.visit_span(decl_span);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
TyKind::Tup(tys) => visit_thin_vec(tys, |ty| vis.visit_ty(ty)),
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
TyKind::Paren(ty) => vis.visit_ty(ty),
2023-01-30 09:50:16 +00:00
TyKind::Pat(ty, pat) => {
vis.visit_ty(ty);
vis.visit_pat(pat);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
TyKind::Path(qself, path) => {
vis.visit_qself(qself);
vis.visit_path(path);
}
TyKind::Array(ty, length) => {
vis.visit_ty(ty);
vis.visit_anon_const(length);
}
TyKind::Typeof(expr) => vis.visit_anon_const(expr),
TyKind::TraitObject(bounds, _syntax) => {
visit_vec(bounds, |bound| vis.visit_param_bound(bound))
2019-12-22 22:42:04 +00:00
}
2024-06-05 20:18:52 +00:00
TyKind::ImplTrait(id, bounds) => {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_id(id);
visit_vec(bounds, |bound| vis.visit_param_bound(bound));
}
TyKind::MacCall(mac) => vis.visit_mac_call(mac),
TyKind::AnonStruct(id, fields) | TyKind::AnonUnion(id, fields) => {
vis.visit_id(id);
fields.flat_map_in_place(|field| vis.flat_map_field_def(field));
}
}
visit_lazy_tts(tokens, vis);
vis.visit_span(span);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
fn noop_visit_foreign_mod<T: MutVisitor>(foreign_mod: &mut ForeignMod, vis: &mut T) {
2024-05-17 17:17:48 +00:00
let ForeignMod { safety, abi: _, items } = foreign_mod;
visit_safety(safety, vis);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
items.flat_map_in_place(|item| vis.flat_map_foreign_item(item));
}
pub fn noop_flat_map_variant<T: MutVisitor>(
mut variant: Variant,
visitor: &mut T,
) -> SmallVec<[Variant; 1]> {
let Variant { ident, vis, attrs, id, data, disr_expr, span, is_placeholder: _ } = &mut variant;
visitor.visit_id(id);
visit_attrs(attrs, visitor);
visitor.visit_vis(vis);
visitor.visit_ident(ident);
visitor.visit_variant_data(data);
visit_opt(disr_expr, |disr_expr| visitor.visit_anon_const(disr_expr));
visitor.visit_span(span);
smallvec![variant]
}
fn noop_visit_ident<T: MutVisitor>(Ident { name: _, span }: &mut Ident, vis: &mut T) {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_span(span);
}
fn noop_visit_path<T: MutVisitor>(Path { segments, span, tokens }: &mut Path, vis: &mut T) {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
for PathSegment { ident, id, args } in segments {
vis.visit_id(id);
vis.visit_ident(ident);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
visit_opt(args, |args| vis.visit_generic_args(args));
}
visit_lazy_tts(tokens, vis);
vis.visit_span(span);
}
fn noop_visit_qself<T: MutVisitor>(qself: &mut Option<P<QSelf>>, vis: &mut T) {
visit_opt(qself, |qself| {
let QSelf { ty, path_span, position: _ } = &mut **qself;
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_ty(ty);
vis.visit_span(path_span);
})
2018-06-11 23:26:53 +00:00
}
fn noop_visit_generic_args<T: MutVisitor>(generic_args: &mut GenericArgs, vis: &mut T) {
match generic_args {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
GenericArgs::AngleBracketed(data) => vis.visit_angle_bracketed_parameter_data(data),
GenericArgs::Parenthesized(data) => vis.visit_parenthesized_parameter_data(data),
}
}
fn noop_visit_generic_arg<T: MutVisitor>(arg: &mut GenericArg, vis: &mut T) {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
match arg {
GenericArg::Lifetime(lt) => vis.visit_lifetime(lt),
GenericArg::Type(ty) => vis.visit_ty(ty),
GenericArg::Const(ct) => vis.visit_anon_const(ct),
}
}
fn noop_visit_angle_bracketed_parameter_data<T: MutVisitor>(
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
data: &mut AngleBracketedArgs,
vis: &mut T,
) {
let AngleBracketedArgs { args, span } = data;
visit_thin_vec(args, |arg| match arg {
AngleBracketedArg::Arg(arg) => vis.visit_generic_arg(arg),
AngleBracketedArg::Constraint(constraint) => vis.visit_assoc_item_constraint(constraint),
});
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_span(span);
}
fn noop_visit_parenthesized_parameter_data<T: MutVisitor>(
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
args: &mut ParenthesizedArgs,
vis: &mut T,
) {
let ParenthesizedArgs { inputs, output, span, inputs_span } = args;
visit_thin_vec(inputs, |input| vis.visit_ty(input));
2019-12-01 15:00:08 +00:00
noop_visit_fn_ret_ty(output, vis);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_span(span);
vis.visit_span(inputs_span);
}
fn noop_visit_local<T: MutVisitor>(local: &mut P<Local>, vis: &mut T) {
let Local { id, pat, ty, kind, span, colon_sp, attrs, tokens } = local.deref_mut();
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_id(id);
visit_attrs(attrs, vis);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_pat(pat);
visit_opt(ty, |ty| vis.visit_ty(ty));
2021-06-22 18:00:58 +00:00
match kind {
LocalKind::Decl => {}
LocalKind::Init(init) => {
vis.visit_expr(init);
}
LocalKind::InitElse(init, els) => {
vis.visit_expr(init);
vis.visit_block(els);
}
}
visit_lazy_tts(tokens, vis);
visit_opt(colon_sp, |sp| vis.visit_span(sp));
vis.visit_span(span);
}
fn noop_visit_attribute<T: MutVisitor>(attr: &mut Attribute, vis: &mut T) {
2020-11-05 17:27:48 +00:00
let Attribute { kind, id: _, style: _, span } = attr;
match kind {
2022-08-11 11:06:11 +00:00
AttrKind::Normal(normal) => {
2024-04-21 04:54:50 +00:00
let NormalAttr {
item: AttrItem { unsafety: _, path, args, tokens },
tokens: attr_tokens,
} = &mut **normal;
vis.visit_path(path);
visit_attr_args(args, vis);
visit_lazy_tts(tokens, vis);
2020-11-05 17:27:48 +00:00
visit_lazy_tts(attr_tokens, vis);
}
AttrKind::DocComment(_kind, _sym) => {}
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_span(span);
2017-03-17 21:58:48 +00:00
}
fn noop_visit_mac<T: MutVisitor>(mac: &mut MacCall, vis: &mut T) {
let MacCall { path, args } = mac;
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_path(path);
visit_delim_args(args, vis);
}
fn noop_visit_macro_def<T: MutVisitor>(macro_def: &mut MacroDef, vis: &mut T) {
let MacroDef { body, macro_rules: _ } = macro_def;
visit_delim_args(body, vis);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
fn noop_visit_meta_list_item<T: MutVisitor>(li: &mut NestedMetaItem, vis: &mut T) {
match li {
NestedMetaItem::MetaItem(mi) => vis.visit_meta_item(mi),
NestedMetaItem::Lit(_lit) => {}
}
}
fn noop_visit_meta_item<T: MutVisitor>(mi: &mut MetaItem, vis: &mut T) {
2024-04-21 04:54:50 +00:00
let MetaItem { unsafety: _, path: _, kind, span } = mi;
match kind {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
MetaItemKind::Word => {}
MetaItemKind::List(mis) => visit_thin_vec(mis, |mi| vis.visit_meta_list_item(mi)),
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
MetaItemKind::NameValue(_s) => {}
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_span(span);
}
pub fn noop_flat_map_param<T: MutVisitor>(mut param: Param, vis: &mut T) -> SmallVec<[Param; 1]> {
let Param { attrs, id, pat, span, ty, is_placeholder: _ } = &mut param;
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_id(id);
visit_attrs(attrs, vis);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_pat(pat);
vis.visit_ty(ty);
vis.visit_span(span);
smallvec![param]
}
2011-06-21 00:25:49 +00:00
Implement token-based handling of attributes during expansion This PR modifies the macro expansion infrastructure to handle attributes in a fully token-based manner. As a result: * Derives macros no longer lose spans when their input is modified by eager cfg-expansion. This is accomplished by performing eager cfg-expansion on the token stream that we pass to the derive proc-macro * Inner attributes now preserve spans in all cases, including when we have multiple inner attributes in a row. This is accomplished through the following changes: * New structs `AttrAnnotatedTokenStream` and `AttrAnnotatedTokenTree` are introduced. These are very similar to a normal `TokenTree`, but they also track the position of attributes and attribute targets within the stream. They are built when we collect tokens during parsing. An `AttrAnnotatedTokenStream` is converted to a regular `TokenStream` when we invoke a macro. * Token capturing and `LazyTokenStream` are modified to work with `AttrAnnotatedTokenStream`. A new `ReplaceRange` type is introduced, which is created during the parsing of a nested AST node to make the 'outer' AST node aware of the attributes and attribute target stored deeper in the token stream. * When we need to perform eager cfg-expansion (either due to `#[derive]` or `#[cfg_eval]`), we tokenize and reparse our target, capturing additional information about the locations of `#[cfg]` and `#[cfg_attr]` attributes at any depth within the target. This is a performance optimization, allowing us to perform less work in the typical case where captured tokens never have eager cfg-expansion run.
2020-11-28 23:33:17 +00:00
// No `noop_` prefix because there isn't a corresponding method in `MutVisitor`.
fn visit_attr_tt<T: MutVisitor>(tt: &mut AttrTokenTree, vis: &mut T) {
Implement token-based handling of attributes during expansion This PR modifies the macro expansion infrastructure to handle attributes in a fully token-based manner. As a result: * Derives macros no longer lose spans when their input is modified by eager cfg-expansion. This is accomplished by performing eager cfg-expansion on the token stream that we pass to the derive proc-macro * Inner attributes now preserve spans in all cases, including when we have multiple inner attributes in a row. This is accomplished through the following changes: * New structs `AttrAnnotatedTokenStream` and `AttrAnnotatedTokenTree` are introduced. These are very similar to a normal `TokenTree`, but they also track the position of attributes and attribute targets within the stream. They are built when we collect tokens during parsing. An `AttrAnnotatedTokenStream` is converted to a regular `TokenStream` when we invoke a macro. * Token capturing and `LazyTokenStream` are modified to work with `AttrAnnotatedTokenStream`. A new `ReplaceRange` type is introduced, which is created during the parsing of a nested AST node to make the 'outer' AST node aware of the attributes and attribute target stored deeper in the token stream. * When we need to perform eager cfg-expansion (either due to `#[derive]` or `#[cfg_eval]`), we tokenize and reparse our target, capturing additional information about the locations of `#[cfg]` and `#[cfg_attr]` attributes at any depth within the target. This is a performance optimization, allowing us to perform less work in the typical case where captured tokens never have eager cfg-expansion run.
2020-11-28 23:33:17 +00:00
match tt {
AttrTokenTree::Token(token, _spacing) => {
Implement token-based handling of attributes during expansion This PR modifies the macro expansion infrastructure to handle attributes in a fully token-based manner. As a result: * Derives macros no longer lose spans when their input is modified by eager cfg-expansion. This is accomplished by performing eager cfg-expansion on the token stream that we pass to the derive proc-macro * Inner attributes now preserve spans in all cases, including when we have multiple inner attributes in a row. This is accomplished through the following changes: * New structs `AttrAnnotatedTokenStream` and `AttrAnnotatedTokenTree` are introduced. These are very similar to a normal `TokenTree`, but they also track the position of attributes and attribute targets within the stream. They are built when we collect tokens during parsing. An `AttrAnnotatedTokenStream` is converted to a regular `TokenStream` when we invoke a macro. * Token capturing and `LazyTokenStream` are modified to work with `AttrAnnotatedTokenStream`. A new `ReplaceRange` type is introduced, which is created during the parsing of a nested AST node to make the 'outer' AST node aware of the attributes and attribute target stored deeper in the token stream. * When we need to perform eager cfg-expansion (either due to `#[derive]` or `#[cfg_eval]`), we tokenize and reparse our target, capturing additional information about the locations of `#[cfg]` and `#[cfg_attr]` attributes at any depth within the target. This is a performance optimization, allowing us to perform less work in the typical case where captured tokens never have eager cfg-expansion run.
2020-11-28 23:33:17 +00:00
visit_token(token, vis);
}
AttrTokenTree::Delimited(dspan, _spacing, _delim, tts) => {
visit_attr_tts(tts, vis);
visit_delim_span(dspan, vis);
Implement token-based handling of attributes during expansion This PR modifies the macro expansion infrastructure to handle attributes in a fully token-based manner. As a result: * Derives macros no longer lose spans when their input is modified by eager cfg-expansion. This is accomplished by performing eager cfg-expansion on the token stream that we pass to the derive proc-macro * Inner attributes now preserve spans in all cases, including when we have multiple inner attributes in a row. This is accomplished through the following changes: * New structs `AttrAnnotatedTokenStream` and `AttrAnnotatedTokenTree` are introduced. These are very similar to a normal `TokenTree`, but they also track the position of attributes and attribute targets within the stream. They are built when we collect tokens during parsing. An `AttrAnnotatedTokenStream` is converted to a regular `TokenStream` when we invoke a macro. * Token capturing and `LazyTokenStream` are modified to work with `AttrAnnotatedTokenStream`. A new `ReplaceRange` type is introduced, which is created during the parsing of a nested AST node to make the 'outer' AST node aware of the attributes and attribute target stored deeper in the token stream. * When we need to perform eager cfg-expansion (either due to `#[derive]` or `#[cfg_eval]`), we tokenize and reparse our target, capturing additional information about the locations of `#[cfg]` and `#[cfg_attr]` attributes at any depth within the target. This is a performance optimization, allowing us to perform less work in the typical case where captured tokens never have eager cfg-expansion run.
2020-11-28 23:33:17 +00:00
}
AttrTokenTree::Attributes(AttributesData { attrs, tokens }) => {
visit_attrs(attrs, vis);
visit_lazy_tts_opt_mut(Some(tokens), vis);
Implement token-based handling of attributes during expansion This PR modifies the macro expansion infrastructure to handle attributes in a fully token-based manner. As a result: * Derives macros no longer lose spans when their input is modified by eager cfg-expansion. This is accomplished by performing eager cfg-expansion on the token stream that we pass to the derive proc-macro * Inner attributes now preserve spans in all cases, including when we have multiple inner attributes in a row. This is accomplished through the following changes: * New structs `AttrAnnotatedTokenStream` and `AttrAnnotatedTokenTree` are introduced. These are very similar to a normal `TokenTree`, but they also track the position of attributes and attribute targets within the stream. They are built when we collect tokens during parsing. An `AttrAnnotatedTokenStream` is converted to a regular `TokenStream` when we invoke a macro. * Token capturing and `LazyTokenStream` are modified to work with `AttrAnnotatedTokenStream`. A new `ReplaceRange` type is introduced, which is created during the parsing of a nested AST node to make the 'outer' AST node aware of the attributes and attribute target stored deeper in the token stream. * When we need to perform eager cfg-expansion (either due to `#[derive]` or `#[cfg_eval]`), we tokenize and reparse our target, capturing additional information about the locations of `#[cfg]` and `#[cfg_attr]` attributes at any depth within the target. This is a performance optimization, allowing us to perform less work in the typical case where captured tokens never have eager cfg-expansion run.
2020-11-28 23:33:17 +00:00
}
}
}
// No `noop_` prefix because there isn't a corresponding method in `MutVisitor`.
fn visit_tt<T: MutVisitor>(tt: &mut TokenTree, vis: &mut T) {
2017-02-21 05:05:59 +00:00
match tt {
TokenTree::Token(token, _spacing) => {
visit_token(token, vis);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
TokenTree::Delimited(dspan, _spacing, _delim, tts) => {
visit_tts(tts, vis);
visit_delim_span(dspan, vis);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
}
2013-06-06 21:17:18 +00:00
}
// No `noop_` prefix because there isn't a corresponding method in `MutVisitor`.
fn visit_tts<T: MutVisitor>(TokenStream(tts): &mut TokenStream, vis: &mut T) {
if T::VISIT_TOKENS && !tts.is_empty() {
let tts = Lrc::make_mut(tts);
visit_vec(tts, |tree| visit_tt(tree, vis));
}
}
fn visit_attr_tts<T: MutVisitor>(AttrTokenStream(tts): &mut AttrTokenStream, vis: &mut T) {
if T::VISIT_TOKENS && !tts.is_empty() {
Implement token-based handling of attributes during expansion This PR modifies the macro expansion infrastructure to handle attributes in a fully token-based manner. As a result: * Derives macros no longer lose spans when their input is modified by eager cfg-expansion. This is accomplished by performing eager cfg-expansion on the token stream that we pass to the derive proc-macro * Inner attributes now preserve spans in all cases, including when we have multiple inner attributes in a row. This is accomplished through the following changes: * New structs `AttrAnnotatedTokenStream` and `AttrAnnotatedTokenTree` are introduced. These are very similar to a normal `TokenTree`, but they also track the position of attributes and attribute targets within the stream. They are built when we collect tokens during parsing. An `AttrAnnotatedTokenStream` is converted to a regular `TokenStream` when we invoke a macro. * Token capturing and `LazyTokenStream` are modified to work with `AttrAnnotatedTokenStream`. A new `ReplaceRange` type is introduced, which is created during the parsing of a nested AST node to make the 'outer' AST node aware of the attributes and attribute target stored deeper in the token stream. * When we need to perform eager cfg-expansion (either due to `#[derive]` or `#[cfg_eval]`), we tokenize and reparse our target, capturing additional information about the locations of `#[cfg]` and `#[cfg_attr]` attributes at any depth within the target. This is a performance optimization, allowing us to perform less work in the typical case where captured tokens never have eager cfg-expansion run.
2020-11-28 23:33:17 +00:00
let tts = Lrc::make_mut(tts);
visit_vec(tts, |tree| visit_attr_tt(tree, vis));
Implement token-based handling of attributes during expansion This PR modifies the macro expansion infrastructure to handle attributes in a fully token-based manner. As a result: * Derives macros no longer lose spans when their input is modified by eager cfg-expansion. This is accomplished by performing eager cfg-expansion on the token stream that we pass to the derive proc-macro * Inner attributes now preserve spans in all cases, including when we have multiple inner attributes in a row. This is accomplished through the following changes: * New structs `AttrAnnotatedTokenStream` and `AttrAnnotatedTokenTree` are introduced. These are very similar to a normal `TokenTree`, but they also track the position of attributes and attribute targets within the stream. They are built when we collect tokens during parsing. An `AttrAnnotatedTokenStream` is converted to a regular `TokenStream` when we invoke a macro. * Token capturing and `LazyTokenStream` are modified to work with `AttrAnnotatedTokenStream`. A new `ReplaceRange` type is introduced, which is created during the parsing of a nested AST node to make the 'outer' AST node aware of the attributes and attribute target stored deeper in the token stream. * When we need to perform eager cfg-expansion (either due to `#[derive]` or `#[cfg_eval]`), we tokenize and reparse our target, capturing additional information about the locations of `#[cfg]` and `#[cfg_attr]` attributes at any depth within the target. This is a performance optimization, allowing us to perform less work in the typical case where captured tokens never have eager cfg-expansion run.
2020-11-28 23:33:17 +00:00
}
}
fn visit_lazy_tts_opt_mut<T: MutVisitor>(lazy_tts: Option<&mut LazyAttrTokenStream>, vis: &mut T) {
if T::VISIT_TOKENS {
Implement token-based handling of attributes during expansion This PR modifies the macro expansion infrastructure to handle attributes in a fully token-based manner. As a result: * Derives macros no longer lose spans when their input is modified by eager cfg-expansion. This is accomplished by performing eager cfg-expansion on the token stream that we pass to the derive proc-macro * Inner attributes now preserve spans in all cases, including when we have multiple inner attributes in a row. This is accomplished through the following changes: * New structs `AttrAnnotatedTokenStream` and `AttrAnnotatedTokenTree` are introduced. These are very similar to a normal `TokenTree`, but they also track the position of attributes and attribute targets within the stream. They are built when we collect tokens during parsing. An `AttrAnnotatedTokenStream` is converted to a regular `TokenStream` when we invoke a macro. * Token capturing and `LazyTokenStream` are modified to work with `AttrAnnotatedTokenStream`. A new `ReplaceRange` type is introduced, which is created during the parsing of a nested AST node to make the 'outer' AST node aware of the attributes and attribute target stored deeper in the token stream. * When we need to perform eager cfg-expansion (either due to `#[derive]` or `#[cfg_eval]`), we tokenize and reparse our target, capturing additional information about the locations of `#[cfg]` and `#[cfg_attr]` attributes at any depth within the target. This is a performance optimization, allowing us to perform less work in the typical case where captured tokens never have eager cfg-expansion run.
2020-11-28 23:33:17 +00:00
if let Some(lazy_tts) = lazy_tts {
let mut tts = lazy_tts.to_attr_token_stream();
visit_attr_tts(&mut tts, vis);
*lazy_tts = LazyAttrTokenStream::new(tts);
Implement token-based handling of attributes during expansion This PR modifies the macro expansion infrastructure to handle attributes in a fully token-based manner. As a result: * Derives macros no longer lose spans when their input is modified by eager cfg-expansion. This is accomplished by performing eager cfg-expansion on the token stream that we pass to the derive proc-macro * Inner attributes now preserve spans in all cases, including when we have multiple inner attributes in a row. This is accomplished through the following changes: * New structs `AttrAnnotatedTokenStream` and `AttrAnnotatedTokenTree` are introduced. These are very similar to a normal `TokenTree`, but they also track the position of attributes and attribute targets within the stream. They are built when we collect tokens during parsing. An `AttrAnnotatedTokenStream` is converted to a regular `TokenStream` when we invoke a macro. * Token capturing and `LazyTokenStream` are modified to work with `AttrAnnotatedTokenStream`. A new `ReplaceRange` type is introduced, which is created during the parsing of a nested AST node to make the 'outer' AST node aware of the attributes and attribute target stored deeper in the token stream. * When we need to perform eager cfg-expansion (either due to `#[derive]` or `#[cfg_eval]`), we tokenize and reparse our target, capturing additional information about the locations of `#[cfg]` and `#[cfg_attr]` attributes at any depth within the target. This is a performance optimization, allowing us to perform less work in the typical case where captured tokens never have eager cfg-expansion run.
2020-11-28 23:33:17 +00:00
}
}
}
fn visit_lazy_tts<T: MutVisitor>(lazy_tts: &mut Option<LazyAttrTokenStream>, vis: &mut T) {
Implement token-based handling of attributes during expansion This PR modifies the macro expansion infrastructure to handle attributes in a fully token-based manner. As a result: * Derives macros no longer lose spans when their input is modified by eager cfg-expansion. This is accomplished by performing eager cfg-expansion on the token stream that we pass to the derive proc-macro * Inner attributes now preserve spans in all cases, including when we have multiple inner attributes in a row. This is accomplished through the following changes: * New structs `AttrAnnotatedTokenStream` and `AttrAnnotatedTokenTree` are introduced. These are very similar to a normal `TokenTree`, but they also track the position of attributes and attribute targets within the stream. They are built when we collect tokens during parsing. An `AttrAnnotatedTokenStream` is converted to a regular `TokenStream` when we invoke a macro. * Token capturing and `LazyTokenStream` are modified to work with `AttrAnnotatedTokenStream`. A new `ReplaceRange` type is introduced, which is created during the parsing of a nested AST node to make the 'outer' AST node aware of the attributes and attribute target stored deeper in the token stream. * When we need to perform eager cfg-expansion (either due to `#[derive]` or `#[cfg_eval]`), we tokenize and reparse our target, capturing additional information about the locations of `#[cfg]` and `#[cfg_attr]` attributes at any depth within the target. This is a performance optimization, allowing us to perform less work in the typical case where captured tokens never have eager cfg-expansion run.
2020-11-28 23:33:17 +00:00
visit_lazy_tts_opt_mut(lazy_tts.as_mut(), vis);
}
/// Applies ident visitor if it's an ident; applies other visits to interpolated nodes.
/// In practice the ident part is not actually used by specific visitors right now,
/// but there's a test below checking that it works.
// No `noop_` prefix because there isn't a corresponding method in `MutVisitor`.
pub fn visit_token<T: MutVisitor>(t: &mut Token, vis: &mut T) {
let Token { kind, span } = t;
match kind {
token::Ident(name, _ /*raw*/) | token::Lifetime(name) => {
let mut ident = Ident::new(*name, *span);
vis.visit_ident(&mut ident);
*name = ident.name;
2019-06-05 19:04:52 +00:00
*span = ident.span;
return; // Avoid visiting the span for the second time.
}
token::NtIdent(ident, _is_raw) => {
vis.visit_ident(ident);
}
token::NtLifetime(ident) => {
vis.visit_ident(ident);
}
2020-07-01 10:16:49 +00:00
token::Interpolated(nt) => {
let nt = Lrc::make_mut(nt);
visit_nonterminal(nt, vis);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
_ => {}
}
vis.visit_span(span);
}
// No `noop_` prefix because there isn't a corresponding method in `MutVisitor`.
/// Applies the visitor to elements of interpolated nodes.
//
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
// N.B., this can occur only when applying a visitor to partially expanded
// code, where parsed pieces have gotten implanted ito *other* macro
// invocations. This is relevant for macro hygiene, but possibly not elsewhere.
//
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
// One problem here occurs because the types for flat_map_item, flat_map_stmt,
// etc., allow the visitor to return *multiple* items; this is a problem for the
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
// nodes here, because they insist on having exactly one piece. One solution
// would be to mangle the MutVisitor trait to include one-to-many and
// one-to-one versions of these entry points, but that would probably confuse a
// lot of people and help very few. Instead, I'm just going to put in dynamic
// checks. I think the performance impact of this will be pretty much
// nonexistent. The danger is that someone will apply a `MutVisitor` to a
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
// partially expanded node, and will be confused by the fact that their
// `flat_map_item` or `flat_map_stmt` isn't getting called on `NtItem` or `NtStmt`
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
// nodes. Hopefully they'll wind up reading this comment, and doing something
// appropriate.
//
// BTW, design choice: I considered just changing the type of, e.g., `NtItem` to
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
// contain multiple items, but decided against it when I looked at
// `parse_item_or_view_item` and tried to figure out what I would do with
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
// multiple items there....
fn visit_nonterminal<T: MutVisitor>(nt: &mut token::Nonterminal, vis: &mut T) {
match nt {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
token::NtItem(item) => visit_clobber(item, |item| {
// This is probably okay, because the only visitors likely to
// peek inside interpolated nodes will be renamings/markings,
// which map single items to single items.
vis.flat_map_item(item).expect_one("expected visitor to produce exactly one item")
}),
token::NtBlock(block) => vis.visit_block(block),
token::NtStmt(stmt) => visit_clobber(stmt, |stmt| {
// See reasoning above.
stmt.map(|stmt| {
vis.flat_map_stmt(stmt).expect_one("expected visitor to produce exactly one item")
})
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}),
token::NtPat(pat) => vis.visit_pat(pat),
token::NtExpr(expr) => vis.visit_expr(expr),
token::NtTy(ty) => vis.visit_ty(ty),
token::NtLiteral(expr) => vis.visit_expr(expr),
token::NtMeta(item) => {
2024-04-21 04:54:50 +00:00
let AttrItem { unsafety: _, path, args, tokens } = item.deref_mut();
vis.visit_path(path);
visit_attr_args(args, vis);
visit_lazy_tts(tokens, vis);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
token::NtPath(path) => vis.visit_path(path),
token::NtVis(visib) => vis.visit_vis(visib),
}
}
// No `noop_` prefix because there isn't a corresponding method in `MutVisitor`.
fn visit_defaultness<T: MutVisitor>(defaultness: &mut Defaultness, vis: &mut T) {
match defaultness {
Defaultness::Default(span) => vis.visit_span(span),
Defaultness::Final => {}
}
}
// No `noop_` prefix because there isn't a corresponding method in `MutVisitor`.
2024-05-17 17:17:48 +00:00
fn visit_safety<T: MutVisitor>(safety: &mut Safety, vis: &mut T) {
match safety {
Safety::Unsafe(span) => vis.visit_span(span),
Safety::Safe(span) => vis.visit_span(span),
2024-05-17 17:17:48 +00:00
Safety::Default => {}
}
}
// No `noop_` prefix because there isn't a corresponding method in `MutVisitor`.
fn visit_polarity<T: MutVisitor>(polarity: &mut ImplPolarity, vis: &mut T) {
match polarity {
ImplPolarity::Positive => {}
ImplPolarity::Negative(span) => vis.visit_span(span),
}
}
// No `noop_` prefix because there isn't a corresponding method in `MutVisitor`.
fn visit_constness<T: MutVisitor>(constness: &mut Const, vis: &mut T) {
match constness {
Const::Yes(span) => vis.visit_span(span),
Const::No => {}
}
}
fn noop_visit_closure_binder<T: MutVisitor>(binder: &mut ClosureBinder, vis: &mut T) {
match binder {
ClosureBinder::NotPresent => {}
ClosureBinder::For { span: _, generic_params } => {
generic_params.flat_map_in_place(|param| vis.flat_map_generic_param(param));
}
}
}
fn noop_visit_coroutine_kind<T: MutVisitor>(coroutine_kind: &mut CoroutineKind, vis: &mut T) {
2023-12-05 21:39:36 +00:00
match coroutine_kind {
2023-12-04 21:43:38 +00:00
CoroutineKind::Async { span, closure_id, return_impl_trait_id }
2023-12-05 21:45:01 +00:00
| CoroutineKind::Gen { span, closure_id, return_impl_trait_id }
| CoroutineKind::AsyncGen { span, closure_id, return_impl_trait_id } => {
2023-11-28 18:15:39 +00:00
vis.visit_id(closure_id);
vis.visit_id(return_impl_trait_id);
vis.visit_span(span);
2023-11-28 18:15:39 +00:00
}
}
}
fn noop_visit_fn_decl<T: MutVisitor>(decl: &mut P<FnDecl>, vis: &mut T) {
let FnDecl { inputs, output } = decl.deref_mut();
inputs.flat_map_in_place(|param| vis.flat_map_param(param));
2019-12-01 15:00:08 +00:00
noop_visit_fn_ret_ty(output, vis);
}
fn noop_visit_fn_ret_ty<T: MutVisitor>(fn_ret_ty: &mut FnRetTy, vis: &mut T) {
2019-12-01 15:00:08 +00:00
match fn_ret_ty {
2020-02-15 03:10:59 +00:00
FnRetTy::Default(span) => vis.visit_span(span),
FnRetTy::Ty(ty) => vis.visit_ty(ty),
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
}
2011-06-21 00:25:49 +00:00
fn noop_visit_param_bound<T: MutVisitor>(pb: &mut GenericBound, vis: &mut T) {
match pb {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
GenericBound::Trait(ty, _modifier) => vis.visit_poly_trait_ref(ty),
GenericBound::Outlives(lifetime) => noop_visit_lifetime(lifetime, vis),
GenericBound::Use(args, span) => {
2024-06-05 20:18:52 +00:00
for arg in args {
vis.visit_precise_capturing_arg(arg);
}
vis.visit_span(span);
2024-06-05 20:18:52 +00:00
}
}
}
fn noop_visit_precise_capturing_arg<T: MutVisitor>(arg: &mut PreciseCapturingArg, vis: &mut T) {
match arg {
PreciseCapturingArg::Lifetime(lt) => {
vis.visit_lifetime(lt);
}
PreciseCapturingArg::Arg(path, id) => {
vis.visit_id(id);
vis.visit_path(path);
}
}
}
pub fn noop_flat_map_generic_param<T: MutVisitor>(
mut param: GenericParam,
vis: &mut T,
) -> SmallVec<[GenericParam; 1]> {
2022-04-28 19:59:41 +00:00
let GenericParam { id, ident, attrs, bounds, kind, colon_span, is_placeholder: _ } = &mut param;
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_id(id);
visit_attrs(attrs, vis);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_ident(ident);
visit_vec(bounds, |bound| noop_visit_param_bound(bound, vis));
match kind {
GenericParamKind::Lifetime => {}
GenericParamKind::Type { default } => {
visit_opt(default, |default| vis.visit_ty(default));
}
GenericParamKind::Const { ty, kw_span: _, default } => {
vis.visit_ty(ty);
visit_opt(default, |default| vis.visit_anon_const(default));
}
}
if let Some(colon_span) = colon_span {
vis.visit_span(colon_span);
}
smallvec![param]
}
fn noop_visit_label<T: MutVisitor>(Label { ident }: &mut Label, vis: &mut T) {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_ident(ident);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
fn noop_visit_lifetime<T: MutVisitor>(Lifetime { id, ident }: &mut Lifetime, vis: &mut T) {
vis.visit_id(id);
vis.visit_ident(ident);
}
fn noop_visit_generics<T: MutVisitor>(generics: &mut Generics, vis: &mut T) {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
let Generics { params, where_clause, span } = generics;
params.flat_map_in_place(|param| vis.flat_map_generic_param(param));
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_where_clause(where_clause);
vis.visit_span(span);
}
fn noop_visit_ty_alias_where_clauses<T: MutVisitor>(tawcs: &mut TyAliasWhereClauses, vis: &mut T) {
let TyAliasWhereClauses { before, after, split: _ } = tawcs;
let TyAliasWhereClause { has_where_token: _, span: span_before } = before;
let TyAliasWhereClause { has_where_token: _, span: span_after } = after;
vis.visit_span(span_before);
vis.visit_span(span_after);
}
fn noop_visit_where_clause<T: MutVisitor>(wc: &mut WhereClause, vis: &mut T) {
let WhereClause { has_where_token: _, predicates, span } = wc;
2022-11-22 05:23:25 +00:00
visit_thin_vec(predicates, |predicate| vis.visit_where_predicate(predicate));
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_span(span);
}
fn noop_visit_where_predicate<T: MutVisitor>(pred: &mut WherePredicate, vis: &mut T) {
match pred {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
WherePredicate::BoundPredicate(bp) => {
let WhereBoundPredicate { span, bound_generic_params, bounded_ty, bounds } = bp;
bound_generic_params.flat_map_in_place(|param| vis.flat_map_generic_param(param));
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_ty(bounded_ty);
visit_vec(bounds, |bound| vis.visit_param_bound(bound));
vis.visit_span(span);
2015-10-10 00:28:40 +00:00
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
WherePredicate::RegionPredicate(rp) => {
let WhereRegionPredicate { span, lifetime, bounds } = rp;
noop_visit_lifetime(lifetime, vis);
visit_vec(bounds, |bound| noop_visit_param_bound(bound, vis));
vis.visit_span(span);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
WherePredicate::EqPredicate(ep) => {
let WhereEqPredicate { span, lhs_ty, rhs_ty } = ep;
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_ty(lhs_ty);
vis.visit_ty(rhs_ty);
vis.visit_span(span);
}
}
2011-06-21 00:25:49 +00:00
}
fn noop_visit_variant_data<T: MutVisitor>(vdata: &mut VariantData, vis: &mut T) {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
match vdata {
VariantData::Struct { fields, recovered: _ } => {
fields.flat_map_in_place(|field| vis.flat_map_field_def(field));
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
VariantData::Tuple(fields, id) => {
vis.visit_id(id);
fields.flat_map_in_place(|field| vis.flat_map_field_def(field));
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
VariantData::Unit(id) => vis.visit_id(id),
2014-11-07 11:53:45 +00:00
}
}
fn noop_visit_trait_ref<T: MutVisitor>(TraitRef { path, ref_id }: &mut TraitRef, vis: &mut T) {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_id(ref_id);
vis.visit_path(path);
2011-06-21 00:25:49 +00:00
}
fn noop_visit_poly_trait_ref<T: MutVisitor>(p: &mut PolyTraitRef, vis: &mut T) {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
let PolyTraitRef { bound_generic_params, trait_ref, span } = p;
bound_generic_params.flat_map_in_place(|param| vis.flat_map_generic_param(param));
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_trait_ref(trait_ref);
vis.visit_span(span);
}
pub fn noop_flat_map_field_def<T: MutVisitor>(
mut fd: FieldDef,
visitor: &mut T,
) -> SmallVec<[FieldDef; 1]> {
let FieldDef { span, ident, vis, id, ty, attrs, is_placeholder: _ } = &mut fd;
visitor.visit_id(id);
visit_attrs(attrs, visitor);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
visitor.visit_vis(vis);
visit_opt(ident, |ident| visitor.visit_ident(ident));
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
visitor.visit_ty(ty);
visitor.visit_span(span);
smallvec![fd]
}
pub fn noop_flat_map_expr_field<T: MutVisitor>(
mut f: ExprField,
vis: &mut T,
) -> SmallVec<[ExprField; 1]> {
let ExprField { ident, expr, span, is_shorthand: _, attrs, id, is_placeholder: _ } = &mut f;
vis.visit_id(id);
visit_attrs(attrs, vis);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_ident(ident);
vis.visit_expr(expr);
vis.visit_span(span);
smallvec![f]
}
fn noop_visit_mt<T: MutVisitor>(MutTy { ty, mutbl: _ }: &mut MutTy, vis: &mut T) {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_ty(ty);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
pub fn noop_visit_block<T: MutVisitor>(block: &mut P<Block>, vis: &mut T) {
let Block { id, stmts, rules: _, span, tokens, could_be_bare_literal: _ } = block.deref_mut();
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_id(id);
stmts.flat_map_in_place(|stmt| vis.flat_map_stmt(stmt));
visit_lazy_tts(tokens, vis);
vis.visit_span(span);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
pub fn noop_visit_item_kind(kind: &mut impl NoopVisitItemKind, vis: &mut impl MutVisitor) {
kind.noop_visit(vis)
}
impl NoopVisitItemKind for ItemKind {
fn noop_visit(&mut self, vis: &mut impl MutVisitor) {
match self {
ItemKind::ExternCrate(_orig_name) => {}
ItemKind::Use(use_tree) => vis.visit_use_tree(use_tree),
ItemKind::Static(box StaticItem { ty, safety: _, mutability: _, expr }) => {
vis.visit_ty(ty);
visit_opt(expr, |expr| vis.visit_expr(expr));
}
ItemKind::Const(item) => {
visit_const_item(item, vis);
}
ItemKind::Fn(box Fn { defaultness, generics, sig, body }) => {
visit_defaultness(defaultness, vis);
vis.visit_generics(generics);
visit_fn_sig(sig, vis);
visit_opt(body, |body| vis.visit_block(body));
}
2024-05-17 17:17:48 +00:00
ItemKind::Mod(safety, mod_kind) => {
visit_safety(safety, vis);
match mod_kind {
ModKind::Loaded(items, _inline, ModSpans { inner_span, inject_use_span }) => {
items.flat_map_in_place(|item| vis.flat_map_item(item));
vis.visit_span(inner_span);
vis.visit_span(inject_use_span);
}
ModKind::Unloaded => {}
}
}
ItemKind::ForeignMod(nm) => vis.visit_foreign_mod(nm),
ItemKind::GlobalAsm(asm) => vis.visit_inline_asm(asm),
ItemKind::TyAlias(box TyAlias { defaultness, generics, where_clauses, bounds, ty }) => {
visit_defaultness(defaultness, vis);
vis.visit_generics(generics);
visit_bounds(bounds, vis);
visit_opt(ty, |ty| vis.visit_ty(ty));
noop_visit_ty_alias_where_clauses(where_clauses, vis);
}
ItemKind::Enum(EnumDef { variants }, generics) => {
vis.visit_generics(generics);
variants.flat_map_in_place(|variant| vis.flat_map_variant(variant));
}
ItemKind::Struct(variant_data, generics) | ItemKind::Union(variant_data, generics) => {
vis.visit_generics(generics);
vis.visit_variant_data(variant_data);
2024-03-14 08:41:38 +00:00
}
ItemKind::Impl(box Impl {
defaultness,
2024-05-17 17:17:48 +00:00
safety,
generics,
constness,
polarity,
of_trait,
self_ty,
items,
}) => {
visit_defaultness(defaultness, vis);
2024-05-17 17:17:48 +00:00
visit_safety(safety, vis);
vis.visit_generics(generics);
visit_constness(constness, vis);
visit_polarity(polarity, vis);
visit_opt(of_trait, |trait_ref| vis.visit_trait_ref(trait_ref));
vis.visit_ty(self_ty);
items.flat_map_in_place(|item| vis.flat_map_impl_item(item));
}
2024-05-17 17:17:48 +00:00
ItemKind::Trait(box Trait { safety, is_auto: _, generics, bounds, items }) => {
visit_safety(safety, vis);
vis.visit_generics(generics);
visit_bounds(bounds, vis);
items.flat_map_in_place(|item| vis.flat_map_trait_item(item));
}
ItemKind::TraitAlias(generics, bounds) => {
vis.visit_generics(generics);
visit_bounds(bounds, vis);
}
ItemKind::MacCall(m) => vis.visit_mac_call(m),
ItemKind::MacroDef(def) => vis.visit_macro_def(def),
2024-03-15 11:21:03 +00:00
ItemKind::Delegation(box Delegation {
id,
qself,
path,
rename,
body,
from_glob: _,
}) => {
vis.visit_id(id);
vis.visit_qself(qself);
vis.visit_path(path);
if let Some(rename) = rename {
vis.visit_ident(rename);
}
if let Some(body) = body {
vis.visit_block(body);
}
2023-11-26 12:57:31 +00:00
}
ItemKind::DelegationMac(box DelegationMac { qself, prefix, suffixes, body }) => {
vis.visit_qself(qself);
vis.visit_path(prefix);
2024-03-15 11:21:03 +00:00
if let Some(suffixes) = suffixes {
for (ident, rename) in suffixes {
vis.visit_ident(ident);
if let Some(rename) = rename {
vis.visit_ident(rename);
}
}
}
if let Some(body) = body {
vis.visit_block(body);
}
}
2023-11-26 12:57:31 +00:00
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
2018-06-06 22:50:59 +00:00
}
impl NoopVisitItemKind for AssocItemKind {
fn noop_visit(&mut self, visitor: &mut impl MutVisitor) {
match self {
AssocItemKind::Const(item) => {
visit_const_item(item, visitor);
}
AssocItemKind::Fn(box Fn { defaultness, generics, sig, body }) => {
visit_defaultness(defaultness, visitor);
visitor.visit_generics(generics);
visit_fn_sig(sig, visitor);
visit_opt(body, |body| visitor.visit_block(body));
2024-03-14 08:41:38 +00:00
}
AssocItemKind::Type(box TyAlias {
defaultness,
generics,
where_clauses,
bounds,
ty,
}) => {
visit_defaultness(defaultness, visitor);
visitor.visit_generics(generics);
visit_bounds(bounds, visitor);
visit_opt(ty, |ty| visitor.visit_ty(ty));
noop_visit_ty_alias_where_clauses(where_clauses, visitor);
}
AssocItemKind::MacCall(mac) => visitor.visit_mac_call(mac),
2024-03-15 11:21:03 +00:00
AssocItemKind::Delegation(box Delegation {
id,
qself,
path,
rename,
body,
from_glob: _,
}) => {
visitor.visit_id(id);
visitor.visit_qself(qself);
visitor.visit_path(path);
if let Some(rename) = rename {
visitor.visit_ident(rename);
}
if let Some(body) = body {
visitor.visit_block(body);
}
2023-11-26 12:57:31 +00:00
}
AssocItemKind::DelegationMac(box DelegationMac { qself, prefix, suffixes, body }) => {
visitor.visit_qself(qself);
visitor.visit_path(prefix);
2024-03-15 11:21:03 +00:00
if let Some(suffixes) = suffixes {
for (ident, rename) in suffixes {
visitor.visit_ident(ident);
if let Some(rename) = rename {
visitor.visit_ident(rename);
}
}
}
if let Some(body) = body {
visitor.visit_block(body);
}
}
2023-11-26 12:57:31 +00:00
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
}
fn visit_const_item<T: MutVisitor>(
2023-05-04 14:08:33 +00:00
ConstItem { defaultness, generics, ty, expr }: &mut ConstItem,
visitor: &mut T,
) {
visit_defaultness(defaultness, visitor);
2023-05-04 14:08:33 +00:00
visitor.visit_generics(generics);
visitor.visit_ty(ty);
visit_opt(expr, |expr| visitor.visit_expr(expr));
}
fn noop_visit_fn_header<T: MutVisitor>(header: &mut FnHeader, vis: &mut T) {
2024-05-17 17:17:48 +00:00
let FnHeader { safety, coroutine_kind, constness, ext: _ } = header;
visit_constness(constness, vis);
2023-12-05 21:39:36 +00:00
coroutine_kind.as_mut().map(|coroutine_kind| vis.visit_coroutine_kind(coroutine_kind));
2024-05-17 17:17:48 +00:00
visit_safety(safety, vis);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
pub fn noop_visit_crate<T: MutVisitor>(krate: &mut Crate, vis: &mut T) {
2022-03-03 23:45:25 +00:00
let Crate { attrs, items, spans, id, is_placeholder: _ } = krate;
vis.visit_id(id);
visit_attrs(attrs, vis);
items.flat_map_in_place(|item| vis.flat_map_item(item));
let ModSpans { inner_span, inject_use_span } = spans;
2022-03-03 23:45:25 +00:00
vis.visit_span(inner_span);
vis.visit_span(inject_use_span);
2011-06-21 00:25:49 +00:00
}
// Mutates one item into possibly many items.
pub fn noop_flat_map_item<K: NoopVisitItemKind>(
mut item: P<Item<K>>,
visitor: &mut impl MutVisitor,
) -> SmallVec<[P<Item<K>>; 1]> {
let Item { ident, attrs, id, kind, vis, span, tokens } = item.deref_mut();
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
visitor.visit_id(id);
visit_attrs(attrs, visitor);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
visitor.visit_vis(vis);
visitor.visit_ident(ident);
kind.noop_visit(visitor);
visit_lazy_tts(tokens, visitor);
visitor.visit_span(span);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
smallvec![item]
}
impl NoopVisitItemKind for ForeignItemKind {
fn noop_visit(&mut self, visitor: &mut impl MutVisitor) {
match self {
ForeignItemKind::Static(box StaticItem { ty, mutability: _, expr, safety: _ }) => {
visitor.visit_ty(ty);
visit_opt(expr, |expr| visitor.visit_expr(expr));
}
ForeignItemKind::Fn(box Fn { defaultness, generics, sig, body }) => {
visit_defaultness(defaultness, visitor);
visitor.visit_generics(generics);
visit_fn_sig(sig, visitor);
visit_opt(body, |body| visitor.visit_block(body));
}
ForeignItemKind::TyAlias(box TyAlias {
defaultness,
generics,
where_clauses,
bounds,
ty,
}) => {
visit_defaultness(defaultness, visitor);
visitor.visit_generics(generics);
visit_bounds(bounds, visitor);
visit_opt(ty, |ty| visitor.visit_ty(ty));
noop_visit_ty_alias_where_clauses(where_clauses, visitor);
}
ForeignItemKind::MacCall(mac) => visitor.visit_mac_call(mac),
}
}
2014-01-06 12:00:46 +00:00
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
pub fn noop_visit_pat<T: MutVisitor>(pat: &mut P<Pat>, vis: &mut T) {
let Pat { id, kind, span, tokens } = pat.deref_mut();
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_id(id);
2019-09-26 15:18:31 +00:00
match kind {
PatKind::Err(_guar) => {}
PatKind::Wild | PatKind::Rest | PatKind::Never => {}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
PatKind::Ident(_binding_mode, ident, sub) => {
vis.visit_ident(ident);
visit_opt(sub, |sub| vis.visit_pat(sub));
}
PatKind::Lit(e) => vis.visit_expr(e),
PatKind::TupleStruct(qself, path, elems) => {
vis.visit_qself(qself);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_path(path);
visit_thin_vec(elems, |elem| vis.visit_pat(elem));
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
PatKind::Path(qself, path) => {
vis.visit_qself(qself);
vis.visit_path(path);
}
PatKind::Struct(qself, path, fields, _etc) => {
vis.visit_qself(qself);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_path(path);
fields.flat_map_in_place(|field| vis.flat_map_pat_field(field));
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
PatKind::Box(inner) => vis.visit_pat(inner),
PatKind::Deref(inner) => vis.visit_pat(inner),
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
PatKind::Ref(inner, _mutbl) => vis.visit_pat(inner),
PatKind::Range(e1, e2, Spanned { span: _, node: _ }) => {
visit_opt(e1, |e| vis.visit_expr(e));
visit_opt(e2, |e| vis.visit_expr(e));
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_span(span);
}
2018-10-19 14:40:07 +00:00
PatKind::Tuple(elems) | PatKind::Slice(elems) | PatKind::Or(elems) => {
visit_thin_vec(elems, |elem| vis.visit_pat(elem))
2019-12-22 22:42:04 +00:00
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
PatKind::Paren(inner) => vis.visit_pat(inner),
PatKind::MacCall(mac) => vis.visit_mac_call(mac),
}
visit_lazy_tts(tokens, vis);
vis.visit_span(span);
}
fn noop_visit_anon_const<T: MutVisitor>(AnonConst { id, value }: &mut AnonConst, vis: &mut T) {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_id(id);
vis.visit_expr(value);
}
fn noop_visit_inline_asm<T: MutVisitor>(asm: &mut InlineAsm, vis: &mut T) {
// FIXME: Visit spans inside all this currently ignored stuff.
let InlineAsm {
template: _,
template_strs: _,
operands,
clobber_abis: _,
options: _,
line_spans: _,
} = asm;
for (op, span) in operands {
match op {
InlineAsmOperand::In { expr, reg: _ }
| InlineAsmOperand::Out { expr: Some(expr), reg: _, late: _ }
| InlineAsmOperand::InOut { expr, reg: _, late: _ } => vis.visit_expr(expr),
InlineAsmOperand::Out { expr: None, reg: _, late: _ } => {}
InlineAsmOperand::SplitInOut { in_expr, out_expr, reg: _, late: _ } => {
vis.visit_expr(in_expr);
if let Some(out_expr) = out_expr {
vis.visit_expr(out_expr);
}
}
InlineAsmOperand::Const { anon_const } => vis.visit_anon_const(anon_const),
InlineAsmOperand::Sym { sym } => vis.visit_inline_asm_sym(sym),
2023-12-25 20:53:01 +00:00
InlineAsmOperand::Label { block } => vis.visit_block(block),
}
vis.visit_span(span);
}
}
fn noop_visit_inline_asm_sym<T: MutVisitor>(
InlineAsmSym { id, qself, path }: &mut InlineAsmSym,
vis: &mut T,
) {
vis.visit_id(id);
vis.visit_qself(qself);
vis.visit_path(path);
}
fn noop_visit_format_args<T: MutVisitor>(fmt: &mut FormatArgs, vis: &mut T) {
// FIXME: visit the template exhaustively.
let FormatArgs { span, template: _, arguments } = fmt;
for FormatArgument { kind, expr } in arguments.all_args_mut() {
match kind {
FormatArgumentKind::Named(ident) | FormatArgumentKind::Captured(ident) => {
vis.visit_ident(ident)
}
FormatArgumentKind::Normal => {}
}
vis.visit_expr(expr);
}
vis.visit_span(span);
}
2020-05-19 20:56:20 +00:00
pub fn noop_visit_expr<T: MutVisitor>(
Expr { kind, id, span, attrs, tokens }: &mut Expr,
2020-05-19 20:56:20 +00:00
vis: &mut T,
) {
vis.visit_id(id);
visit_attrs(attrs, vis);
match kind {
ExprKind::Array(exprs) => visit_thin_exprs(exprs, vis),
2020-09-21 20:55:58 +00:00
ExprKind::ConstBlock(anon_const) => {
vis.visit_anon_const(anon_const);
2020-09-21 20:55:58 +00:00
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
ExprKind::Repeat(expr, count) => {
vis.visit_expr(expr);
vis.visit_anon_const(count);
}
ExprKind::Tup(exprs) => visit_thin_exprs(exprs, vis),
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
ExprKind::Call(f, args) => {
vis.visit_expr(f);
visit_thin_exprs(args, vis);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
ExprKind::MethodCall(box MethodCall {
seg: PathSegment { ident, id, args: seg_args },
receiver,
args: call_args,
span,
}) => {
vis.visit_method_receiver_expr(receiver);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_id(id);
vis.visit_ident(ident);
visit_opt(seg_args, |args| vis.visit_generic_args(args));
visit_thin_exprs(call_args, vis);
vis.visit_span(span);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
ExprKind::Binary(_binop, lhs, rhs) => {
vis.visit_expr(lhs);
vis.visit_expr(rhs);
}
ExprKind::Unary(_unop, ohs) => vis.visit_expr(ohs),
ExprKind::Cast(expr, ty) => {
vis.visit_expr(expr);
vis.visit_ty(ty);
}
ExprKind::Type(expr, ty) => {
vis.visit_expr(expr);
vis.visit_ty(ty);
}
ExprKind::AddrOf(_kind, _mut, ohs) => vis.visit_expr(ohs),
ExprKind::Let(pat, scrutinee, span, _recovered) => {
vis.visit_pat(pat);
vis.visit_expr(scrutinee);
vis.visit_span(span);
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
ExprKind::If(cond, tr, fl) => {
vis.visit_expr(cond);
vis.visit_block(tr);
visit_opt(fl, |fl| ensure_sufficient_stack(|| vis.visit_expr(fl)));
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
ExprKind::While(cond, body, label) => {
visit_opt(label, |label| vis.visit_label(label));
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_expr(cond);
vis.visit_block(body);
}
2023-12-08 22:51:50 +00:00
ExprKind::ForLoop { pat, iter, body, label, kind: _ } => {
visit_opt(label, |label| vis.visit_label(label));
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_pat(pat);
vis.visit_expr(iter);
vis.visit_block(body);
}
ExprKind::Loop(body, label, span) => {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
visit_opt(label, |label| vis.visit_label(label));
vis.visit_block(body);
vis.visit_span(span);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
ExprKind::Match(expr, arms, _kind) => {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_expr(expr);
arms.flat_map_in_place(|arm| vis.flat_map_arm(arm));
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
ExprKind::Closure(box Closure {
binder,
capture_clause,
2022-12-20 16:15:55 +00:00
constness,
2023-12-05 21:39:36 +00:00
coroutine_kind,
movability: _,
fn_decl,
body,
fn_decl_span,
fn_arg_span,
}) => {
vis.visit_closure_binder(binder);
2022-12-20 16:15:55 +00:00
visit_constness(constness, vis);
2023-12-05 21:39:36 +00:00
coroutine_kind.as_mut().map(|coroutine_kind| vis.visit_coroutine_kind(coroutine_kind));
vis.visit_capture_by(capture_clause);
vis.visit_fn_decl(fn_decl);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_expr(body);
vis.visit_span(fn_decl_span);
vis.visit_span(fn_arg_span);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
ExprKind::Block(blk, label) => {
visit_opt(label, |label| vis.visit_label(label));
vis.visit_block(blk);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
2024-06-27 18:56:57 +00:00
ExprKind::Gen(_capture_by, body, _kind, decl_span) => {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_block(body);
2024-06-27 18:56:57 +00:00
vis.visit_span(decl_span);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
2023-04-25 18:59:16 +00:00
ExprKind::Await(expr, await_kw_span) => {
vis.visit_expr(expr);
vis.visit_span(await_kw_span);
}
ExprKind::Assign(el, er, span) => {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_expr(el);
vis.visit_expr(er);
vis.visit_span(span);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
ExprKind::AssignOp(_op, el, er) => {
vis.visit_expr(el);
vis.visit_expr(er);
}
ExprKind::Field(el, ident) => {
vis.visit_expr(el);
vis.visit_ident(ident);
}
ExprKind::Index(el, er, brackets_span) => {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_expr(el);
vis.visit_expr(er);
vis.visit_span(brackets_span);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
ExprKind::Range(e1, e2, _lim) => {
visit_opt(e1, |e1| vis.visit_expr(e1));
visit_opt(e2, |e2| vis.visit_expr(e2));
}
ExprKind::Underscore => {}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
ExprKind::Path(qself, path) => {
vis.visit_qself(qself);
vis.visit_path(path);
}
ExprKind::Break(label, expr) => {
visit_opt(label, |label| vis.visit_label(label));
visit_opt(expr, |expr| vis.visit_expr(expr));
}
ExprKind::Continue(label) => {
visit_opt(label, |label| vis.visit_label(label));
}
ExprKind::Ret(expr) => {
visit_opt(expr, |expr| vis.visit_expr(expr));
}
ExprKind::Yeet(expr) => {
visit_opt(expr, |expr| vis.visit_expr(expr));
}
ExprKind::Become(expr) => vis.visit_expr(expr),
ExprKind::InlineAsm(asm) => vis.visit_inline_asm(asm),
ExprKind::FormatArgs(fmt) => vis.visit_format_args(fmt),
2022-09-11 07:37:49 +00:00
ExprKind::OffsetOf(container, fields) => {
vis.visit_ty(container);
2023-04-16 21:21:11 +00:00
for field in fields.iter_mut() {
2022-09-11 07:37:49 +00:00
vis.visit_ident(field);
}
}
ExprKind::MacCall(mac) => vis.visit_mac_call(mac),
ExprKind::Struct(se) => {
let StructExpr { qself, path, fields, rest } = se.deref_mut();
vis.visit_qself(qself);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_path(path);
fields.flat_map_in_place(|field| vis.flat_map_expr_field(field));
match rest {
StructRest::Base(expr) => vis.visit_expr(expr),
StructRest::Rest(_span) => {}
StructRest::None => {}
}
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
ExprKind::Paren(expr) => {
vis.visit_expr(expr);
}
ExprKind::Yield(expr) => {
visit_opt(expr, |expr| vis.visit_expr(expr));
}
ExprKind::Try(expr) => vis.visit_expr(expr),
ExprKind::TryBlock(body) => vis.visit_block(body),
ExprKind::Lit(_token) => {}
ExprKind::IncludedBytes(_bytes) => {}
ExprKind::Err(_guar) => {}
ExprKind::Dummy => {}
}
visit_lazy_tts(tokens, vis);
vis.visit_span(span);
2011-06-21 00:25:49 +00:00
}
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
pub fn noop_filter_map_expr<T: MutVisitor>(mut e: P<Expr>, vis: &mut T) -> Option<P<Expr>> {
Some({
vis.visit_expr(&mut e);
e
})
}
2019-09-26 16:34:50 +00:00
pub fn noop_flat_map_stmt<T: MutVisitor>(
Stmt { kind, mut span, mut id }: Stmt,
2019-09-26 16:34:50 +00:00
vis: &mut T,
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
) -> SmallVec<[Stmt; 1]> {
vis.visit_id(&mut id);
let stmts: SmallVec<_> = noop_flat_map_stmt_kind(kind, vis)
.into_iter()
.map(|kind| Stmt { id, kind, span })
.collect();
if stmts.len() > 1 {
panic!(
"cloning statement `NodeId`s is prohibited by default, \
the visitor should implement custom statement visiting"
);
}
vis.visit_span(&mut span);
stmts
}
2016-06-17 02:30:01 +00:00
fn noop_flat_map_stmt_kind<T: MutVisitor>(kind: StmtKind, vis: &mut T) -> SmallVec<[StmtKind; 1]> {
2019-09-26 16:34:50 +00:00
match kind {
StmtKind::Let(mut local) => smallvec![StmtKind::Let({
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_local(&mut local);
local
})],
StmtKind::Item(item) => vis.flat_map_item(item).into_iter().map(StmtKind::Item).collect(),
StmtKind::Expr(expr) => vis.filter_map_expr(expr).into_iter().map(StmtKind::Expr).collect(),
StmtKind::Semi(expr) => vis.filter_map_expr(expr).into_iter().map(StmtKind::Semi).collect(),
StmtKind::Empty => smallvec![StmtKind::Empty],
2020-02-29 16:32:20 +00:00
StmtKind::MacCall(mut mac) => {
let MacCallStmt { mac: mac_, style: _, attrs, tokens } = mac.deref_mut();
visit_attrs(attrs, vis);
vis.visit_mac_call(mac_);
visit_lazy_tts(tokens, vis);
2020-02-29 16:32:20 +00:00
smallvec![StmtKind::MacCall(mac)]
2013-01-15 22:59:39 +00:00
}
}
2011-06-21 00:25:49 +00:00
}
fn noop_visit_vis<T: MutVisitor>(visibility: &mut Visibility, vis: &mut T) {
let Visibility { kind, span, tokens } = visibility;
match kind {
VisibilityKind::Public | VisibilityKind::Inherited => {}
VisibilityKind::Restricted { path, id, shorthand: _ } => {
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
vis.visit_id(id);
vis.visit_path(path);
Overhaul `syntax::fold::Folder`. This commit changes `syntax::fold::Folder` from a functional style (where most methods take a `T` and produce a new `T`) to a more imperative style (where most methods take and modify a `&mut T`), and renames it `syntax::mut_visit::MutVisitor`. The first benefit is speed. The functional style does not require any reallocations, due to the use of `P::map` and `MoveMap::move_{,flat_}map`. However, every field in the AST must be overwritten; even those fields that are unchanged are overwritten with the same value. This causes a lot of unnecessary memory writes. The imperative style reduces instruction counts by 1--3% across a wide range of workloads, particularly incremental workloads. The second benefit is conciseness; the imperative style is usually more concise. E.g. compare the old functional style: ``` fn fold_abc(&mut self, abc: ABC) { ABC { a: fold_a(abc.a), b: fold_b(abc.b), c: abc.c, } } ``` with the imperative style: ``` fn visit_abc(&mut self, ABC { a, b, c: _ }: &mut ABC) { visit_a(a); visit_b(b); } ``` (The reductions get larger in more complex examples.) Overall, the patch removes over 200 lines of code -- even though the new code has more comments -- and a lot of the remaining lines have fewer characters. Some notes: - The old style used methods called `fold_*`. The new style mostly uses methods called `visit_*`, but there are a few methods that map a `T` to something other than a `T`, which are called `flat_map_*` (`T` maps to multiple `T`s) or `filter_map_*` (`T` maps to 0 or 1 `T`s). - `move_map.rs`/`MoveMap`/`move_map`/`move_flat_map` are renamed `map_in_place.rs`/`MapInPlace`/`map_in_place`/`flat_map_in_place` to reflect their slightly changed signatures. - Although this commit renames the `fold` module as `mut_visit`, it keeps it in the `fold.rs` file, so as not to confuse git. The next commit will rename the file.
2019-02-05 04:20:55 +00:00
}
}
visit_lazy_tts(tokens, vis);
vis.visit_span(span);
}
fn noop_visit_capture_by<T: MutVisitor>(capture_by: &mut CaptureBy, vis: &mut T) {
match capture_by {
CaptureBy::Ref => {}
CaptureBy::Value { move_kw } => {
vis.visit_span(move_kw);
}
}
}
/// Some value for the AST node that is valid but possibly meaningless. Similar
/// to `Default` but not intended for wide use. The value will never be used
/// meaningfully, it exists just to support unwinding in `visit_clobber` in the
/// case where its closure panics.
pub trait DummyAstNode {
fn dummy() -> Self;
}
impl<T> DummyAstNode for Option<T> {
fn dummy() -> Self {
Default::default()
}
}
impl<T: DummyAstNode + 'static> DummyAstNode for P<T> {
fn dummy() -> Self {
P(DummyAstNode::dummy())
}
}
impl DummyAstNode for Item {
fn dummy() -> Self {
Item {
attrs: Default::default(),
id: DUMMY_NODE_ID,
span: Default::default(),
vis: Visibility {
kind: VisibilityKind::Public,
span: Default::default(),
tokens: Default::default(),
},
ident: Ident::empty(),
kind: ItemKind::ExternCrate(None),
tokens: Default::default(),
}
}
}
impl DummyAstNode for Expr {
fn dummy() -> Self {
Expr {
id: DUMMY_NODE_ID,
2024-02-25 21:22:09 +00:00
kind: ExprKind::Dummy,
span: Default::default(),
attrs: Default::default(),
tokens: Default::default(),
}
}
}
impl DummyAstNode for Ty {
fn dummy() -> Self {
Ty {
id: DUMMY_NODE_ID,
kind: TyKind::Dummy,
span: Default::default(),
tokens: Default::default(),
}
}
}
impl DummyAstNode for Pat {
fn dummy() -> Self {
Pat {
id: DUMMY_NODE_ID,
kind: PatKind::Wild,
span: Default::default(),
tokens: Default::default(),
}
}
}
impl DummyAstNode for Stmt {
fn dummy() -> Self {
Stmt { id: DUMMY_NODE_ID, kind: StmtKind::Empty, span: Default::default() }
}
}
impl DummyAstNode for Crate {
fn dummy() -> Self {
Crate {
attrs: Default::default(),
items: Default::default(),
2022-03-03 23:45:25 +00:00
spans: Default::default(),
id: DUMMY_NODE_ID,
is_placeholder: Default::default(),
}
}
}
impl<N: DummyAstNode, T: DummyAstNode> DummyAstNode for crate::ast_traits::AstNodeWrapper<N, T> {
fn dummy() -> Self {
crate::ast_traits::AstNodeWrapper::new(N::dummy(), T::dummy())
}
}