Separate dataflow analysis and results
`Analysis` gets put into `Results` with `EntryStates`, by `iterate_to_fixpoint`. This has two problems:
- `Results` is passed various places where only `Analysis` is needed.
- `EntryStates` is passed around mutably everywhere even though it is immutable.
This commit mostly separates `Analysis` from `Results` and fixes these two problems.
r? `@davidtwco`
mir-opt: execute MatchBranchSimplification after GVN
This can provide more opportunities for MatchBranchSimplification.
Currently, rustc does not optimize the following code into a single statement at mir-opt, and this PR fixes the first case.
```rust
pub fn match1(c: bool, v1: i32, v2: i32) -> i32 {
if c { v1 - v2 } else { v1 - v2 }
}
pub fn match2(c: bool, v1: i32) -> i32 {
if c { v1 - 1 } else { v1 - 1 }
}
```
https://rust.godbolt.org/z/Y8xPMjrfM
r? mir-opt
Remove global `next_disambiguator` state and handle it with a `DisambiguatorState` type
This removes `Definitions.next_disambiguator` as it doesn't guarantee deterministic def paths when `create_def` is called in parallel. Instead a new `DisambiguatorState` type is passed as a mutable reference to `create_def` to help create unique def paths. `create_def` calls with distinct `DisambiguatorState` instances must ensure that that the def paths are unique without its help.
Anon associated types did rely on this global state for uniqueness and are changed to use (method they're defined in + their position in the method return type) as the `DefPathData` to ensure uniqueness. This also means that the method they're defined in appears in error messages, which is nicer.
`DefPathData::NestedStatic` is added to use for nested data inside statics instead of reusing `DefPathData::AnonConst` to avoid conflicts with those.
cc `@oli-obk`
Use a closure instead of three chained iterators
Fixes the perf regression from #123948
That PR had chained a third option to the iterator which apparently didn't optimize well
Introduce Arena::try_alloc_from_iter.
`alloc_from_iter` already collects the iterator for reentrancy. So adding an early exit for a fallible iterator integrates naturally into the code. This avoids the other solution to allocate and dump the allocation.
Extend the alignment check to borrows
The current alignment check does not include checks for creating misaligned references from raw pointers, which is now added in this patch.
When inserting the check we need to be careful with references to field projections (e.g. `&(*ptr).a`), in which case the resulting reference must be aligned according to the field type and not the type of the pointer.
r? `@saethlin`
cc `@RalfJung,` after our discussion in #134424
compiletest: Re-land using the new non-libtest executor by default
This PR re-lands #139998, which had the misfortune of triggering download-rustc in its CI jobs, so we didn't get proper test metrics for comparison with the old implementation. So that was PR was reverted in #140233, with the intention of re-landing it alongside a dummy compiler change to inhibit download-rustc.
---
Original PR description for #139998:
>The new executor was implemented in #139660, but required a manual opt-in. This PR activates the new executor by default, but leaves the old libtest-based executor in place (temporarily) to make reverting easier if something unexpectedly goes horribly wrong.
>
>Currently the new executor can be explicitly disabled by passing the `-N` flag to compiletest (e.g. `./x test ui -- -N`), but eventually that flag will be removed, alongside the removal of the libtest dependency. The flag is mostly there to make manual comparative testing easier if something does go wrong.
>
>As before, there *should* be no user-visible difference between the old executor and the new executor.
---
r? jieyouxu
[compiletest] Parallelize test discovery
Certain filesystems are slow to service individual read requests, but can service many in parallel. This change brings down the time to run a single cached test on one of those filesystems from 40s to about 8s.
Rollup of 8 pull requests
Successful merges:
- #137683 (Add a tidy check for GCC submodule version)
- #138968 (Update the index of Result to make the summary more comprehensive)
- #139572 (docs(std): mention const blocks in const keyword doc page)
- #140152 (Unify the format of rustc cli flags)
- #140193 (fix ICE in `#[naked]` attribute validation)
- #140205 (Tidying up UI tests [2/N])
- #140284 (remove expect() in `unnecessary_transmutes`)
- #140290 (rustdoc: fix typo change from equivelent to equivalent)
r? `@ghost`
`@rustbot` modify labels: rollup
The current alignment check does not include checks for creating
misaligned references from raw pointers, which is now added in this
patch.
When inserting the check we need to be careful with references to
field projections (e.g. `&(*ptr).a`), in which case the resulting
reference must be aligned according to the field type and not the
type of the pointer.
Make #![feature(let_chains)] bootstrap conditional in compiler/
Let chains have been stabilized recently in #132833, so we can remove the gating from our uses in the compiler (as the compiler uses edition 2024).
Suggest {to,from}_ne_bytes for transmutations between arrays and integers, etc
implements #136067
Rust has helper methods for many kinds of safe transmutes, for example integer<->bytes. This is a lint against using transmute for these cases.
```rs
fn bytes_at_home(x: [u8; 4]) -> u32 {
transmute(x)
}
// other examples
transmute::<[u8; 2], u16>();
transmute::<[u8; 8], f64>();
transmute::<u32, [u8; 4]>();
transmute::<char, u32>();
transmute::<u32, char>();
```
It would be handy to suggest `u32::from_ne_bytes(x)`.
This is implemented for `[u8; _]` -> `{float int}`
This also implements the cases:
`fXX` <-> `uXX` = `{from_bits, to_bits}`
`uXX` -> `iXX` via `cast_unsigned` and `cast_signed`
{`char` -> `u32`, `bool` -> `n8`} via `from`
`u32` -> `char` via `from_u32_unchecked` (note: notes `from_u32().unwrap()`) (contested)
`u8` -> `bool` via `==` (debatable)
---
try-job: aarch64-gnu
try-job: test-various
mitigate MSVC alignment issue on x86-32
This implements mitigation for https://github.com/rust-lang/rust/issues/112480 by stopping to emit `align` attributes on loads and function arguments when building for a win32 MSVC target. MSVC is known to not properly align `u64` and similar types, and claiming to LLVM that everything is properly aligned increases the chance that this will cause problems.
Of course, the misalignment is still a bug, but we can't fix that bug, only MSVC can.
Also add an errata note to the platform support page warning users about this known problem.
try-job: `i686-msvc*`
`Results` contains and `Analysis` and an `EntryStates`. The unfortunate
thing about this is that the analysis needs to be mutable everywhere
(`&mut Analysis`) which forces the `Results` to be mutable everywhere,
even though `EntryStates` is immutable everywhere.
To fix this, this commit renames `Results` as `AnalysisAndResults`,
renames `EntryStates` as `Results`, and separates the analysis and
results as much as possible. (`AnalysisAndResults` doesn't get much use,
it's mostly there to facilitate method chaining of
`iterate_to_fixpoint`.)
`Results` is immutable everywhere, which:
- is a bit clearer on how the data is used,
- avoids an unnecessary clone of entry states in
`locals_live_across_suspend_points`, and
- moves the results outside the `RefCell` in Formatter.
The commit also reformulates `ResultsHandle` as the generic `CowMut`,
which is simpler than `ResultsHandle` because it doesn't need the
`'tcx` lifetime and the trait bounds. It also which sits nicely
alongside the new use of `Cow` in `ResultsCursor`.
Use `is_lang_item` and `as_lang_item` instead of handrolling their logic
Various cleanups and deduplication. Most notably `if is_lang_item(foo, bar) {} else if is_lang_item...` chains are turned into matches. No behaviour changes intended beyond turning ICEs into fatal "lang item not found" errors
Remove early exits from JumpThreading.
This removes early exits from https://github.com/rust-lang/rust/pull/131203 as I asked during review.
The correctness of the backtracking is `mutated_statement` clearing all relevant conditions. If `process_statement` fails to insert a new condition, for instance by const-eval failure, `mutated_statement` still removes the obsolete conditions from the state.
r? `@compiler-errors`
Do not remove trivial `SwitchInt` in analysis MIR
This PR ensures that we don't prematurely remove trivial `SwitchInt` terminators which affects both the borrow-checking and runtime semantics (i.e. UB) of the code. Previously the `SimplifyCfg` optimization was removing `SwitchInt` terminators when they was "trivial", i.e. when all arms branched to the same basic block, even if that `SwitchInt` terminator had the side-effect of reading an operand which (for example) may not be initialized or may point to an invalid place in memory.
This behavior is unlike all other optimizations, which are only applied after "analysis" (i.e. borrow-checking) is finished, and which Miri disables to make sure the compiler doesn't silently remove UB.
Fixing this code "breaks" (i.e. unmasks) code that used to borrow-check but no longer does, like:
```rust
fn foo() {
let x;
let (0 | _) = x;
}
```
This match expression should perform a read because `_` does not shadow the `0` literal pattern, and the compiler should have to read the match scrutinee to compare it to 0. I've checked that this behavior does not actually manifest in practice via a crater run which came back clean: https://github.com/rust-lang/rust/pull/139042#issuecomment-2767436367
As a side-note, it may be tempting to suggest that this is actually a good thing or that we should preserve this behavior. If we wanted to make this work (i.e. trivially optimize out reads from matches that are redundant like `0 | _`), then we should be enabling this behavior *after* fixing this. However, I think it's kinda unprincipled, and for example other variations of the code don't even work today, e.g.:
```rust
fn foo() {
let x;
let (0.. | _) = x;
}
```
do not emit `OpaqueCast` projections with `-Znext-solver`
We normalize opaque types in their defining scope if the new solver is enabled. This means projections do not contain any 'revealable' opaque types we need to worry about. We either have a type which has been normalized by writeback or we need to normalize it anyways.
r? ```@compiler-errors``` ```@oli-obk```
This will allow us to eagerly translate messages on a top-level
diagnostic, such as a `LintDiagnostic`. As a bonus, we can remove the
awkward closure passed into Subdiagnostic and make better use of
`Into`.
Overhaul `AssocItem`
`AssocItem` has multiple fields that only make sense some of the time. E.g. the `name` can be empty if it's an RPITIT associated type. It's clearer and less error prone if these fields are moved to the relevant `kind` variants.
r? ``@fee1-dead``
To accurately reflect that RPITIT assoc items don't have a name. This
avoids the use of `kw::Empty` to mean "no name", which is error prone.
Helps with #137978.
Visit place in `BackwardIncompatibleDropHint` statement
Remove a weird hack from the `LocalUpdater` where we were manually visiting the place stored in a `StatementKind::BackwardIncompatibleDropHint` because the MIR visitor impls weren't doing so.
Also, clean up `BackwardIncompatibleDropHint`s in `CleanupPostBorrowck`, since they're not needed for runtime MIR.
Rollup of 9 pull requests
Successful merges:
- #138336 (Improve `-Z crate-attr` diagnostics)
- #139636 (Encode dep node edge count as u32 instead of usize)
- #139666 (cleanup `mir_borrowck`)
- #139695 (compiletest: consistently use `camino::{Utf8Path,Utf8PathBuf}` throughout)
- #139699 (Proactively update coroutine drop shim's phase to account for later passes applied during shim query)
- #139718 (enforce unsafe attributes in pre-2024 editions by default)
- #139722 (Move some things to rustc_type_ir)
- #139760 (UI tests: migrate remaining compile time `error-pattern`s to line annotations when possible)
- #139776 (Switch attrs to `diagnostic::on_unimplemented`)
r? `@ghost`
`@rustbot` modify labels: rollup
`hir::AssocItem` currently has a boolean `fn_has_self_parameter` field,
which is misplaced, because it's only relevant for associated fns, not
for associated consts or types. This commit moves it (and renames it) to
the `AssocKind::Fn` variant, where it belongs.
This requires introducing a new C-style enum, `AssocTag`, which is like
`AssocKind` but without the fields. This is because `AssocKind` values
are passed to various functions like `find_by_ident_and_kind` to
indicate what kind of associated item should be searched for, and having
to specify `has_self` isn't relevant there.
New methods:
- Predicates `AssocItem::is_fn` and `AssocItem::is_method`.
- `AssocItem::as_tag` which converts `AssocItem::kind` to `AssocTag`.
Removed `find_by_name_and_kinds`, which is unused.
`AssocItem::descr` can now distinguish between methods and associated
functions, which slightly improves some error messages.
In `rustc_mir_transform`, iterate over index newtypes instead of ints
Just makes code more idiomatic/easier to read, IMHO.
Also, some drive-by simplifications and cleanups.
Micro-optimize `InstSimplify`'s `simplify_primitive_clone`
r? ````@compiler-errors```` , since you already did #139411 and got randomly selected for #139638 (feel free to reassign!)
Another one similar in spirit to #139411, but this time for `simplify_primitive_clone`, which is doing a bit of redundant work. Might not show up in benches, but probably worth micro-optimizing since the transformation is run even for debug builds.
See inline comments for my reasoning for making these changes.
Use delayed bug for normalization errors in drop elaboration
Normalization can fail due to a lot of different earlier errors, so just use span_delayed_bug if normalization failed.
Closes#137287Closes#135668
r? compiler-errors
Allow GVN to produce places and not just locals.
That may be too big of a hammer, as we may introduce new deref projections (possible UB footgun + probably not good for perf).
The second commit opts out of introducing projections that don't have a stable offset, which is probably what we want. Hence no new Deref and no new Index projections.
Fixes https://github.com/rust-lang/rust/issues/138936
cc `@scottmcm` `@dianqk`
gvn: Invalid dereferences for all non-local mutations
Fixes#132353.
This PR removes the computation value by traversing SSA locals through `for_each_assignment_mut`.
Because the `for_each_assignment_mut` traversal skips statements which have side effects, such as dereference assignments, the computation may be unsound. Instead of `for_each_assignment_mut`, we compute values by traversing in reverse postorder.
Because we compute and use the symbolic representation of values on the fly, I invalidate all old values when encountering a dereference assignment. The current approach does not prevent the optimization of a clone to a copy.
In the future, we may add an alias model, or dominance information for dereference assignments, or SSA form to help GVN.
r? cjgillot
cc `@jieyouxu` #132356
cc `@RalfJung` #133474
Misc query tweaks
Remove some redundant work around `cache_on_disk` and `ensure_ok`, since `Result<(), ErrorGuaranteed>` queries don't need to cache or recompute their "value" if they are only used for their result.
`for_each_assignment_mut` can skip assignment statements with side effects,
which can result in some assignment statements retrieving outdated value.
For example, it may skip a dereference assignment statement.
Various local trait item iteration cleanups
Adding a trait impl for `Foo` unconditionally affected all queries that are interested in a completely independent trait `Bar`. Perf has no effect on this. We probably don't have a good perf test for this tho.
r? `@compiler-errors`
I am unsure about 9d05efb66f as it doesn't improve anything wrt incremental, because we still do all the checks for valid `Drop` impls, which subsequently will still invoke many queries and basically keep the depgraph the same.
I want to do
9549077a47/compiler/rustc_middle/src/ty/trait_def.rs (L141)
but would leave that to a follow-up PR, this one changes enough things as it is
coverage: Avoid splitting spans during span extraction/refinement
This PR removes or simplifies some of the steps involved in extracting coverage-relevant spans from MIR, and preparing them for use in coverage instrumentation metadata.
A common theme is that we now try harder to avoid modifying or combining spans in non-trivial ways, because those modifications present the most risk for weird behaviour or ICEs.
The main changes are:
- When extracting spans from MIR call terminators, try to restrict them to just the function name.
- Instead of splitting spans around “holes”, just discard any span that overlaps with a hole.
- Instead of splitting macro-invocation spans into two parts, truncate them to just the macro name and subsequent `!`.
---
This results in a lot of tiny changes to the spans that end up in coverage metadata, and a few changes to coverage reports. Judging by test snapshots, these changes appear to be quite minor in practice.
This is a way to shrink call spans that doesn't involve mixing different spans,
and avoids overlap with argument spans.
This patch also removes some low-value comments that were causing rustfmt to
ignore the match arms.
Remove existing AFIDT implementation
This experiment will need to be reworked differently; I don't think we'll be going with the `dyn* Future` approach that is currently implemented.
r? oli-obk
Fixes#136286Fixes#137706Fixes#137895
Tracking:
* #133119
