Miri: fix error on dangling pointer inbounds offset
We used to claim that the pointer was "dereferenced", but that is just not true.
Can be reviewed commit-by-commit. The first commit is an unrelated rename that didn't seem worth splitting into its own PR.
r? `@oli-obk`
Don't check unnecessarily that impl trait is RPIT
We have this random `return_type_impl_trait` function to detect if a function returns an RPIT which is used in outlives suggestions, but removing it doesn't actually change any diagnostics. Let's just remove it.
Also, suppress a spurious outlives error from a ReError.
Fixes#114274
Operand types are now tracked explicitly, so there is no need to reserve ID 0
for the special always-zero counter.
As part of the renumbering, this change fixes an off-by-one error in the way
counters were counted by the `coverageinfo` query. As a result, functions
should now have exactly the number of counters they actually need, instead of
always having an extra counter that is never used.
Operand types are now tracked explicitly, so there is no need for expression
IDs to avoid counter IDs by descending from `u32::MAX`. Instead they can just
count up from 0, and can be used directly as indices when necessary.
Because the three kinds of operand are now distinguished explicitly, we no
longer need fiddly code to disambiguate counter IDs and expression IDs based on
the total number of counters/expressions in a function.
This does increase the size of operands from 4 bytes to 8 bytes, but that
shouldn't be a big deal since they are mostly stored inside boxed structures,
and the current coverage code is not particularly size-optimized anyway.
The actual motivation here is to prevent `rustfmt` from suddenly reformatting
these enum variants onto a single line, when they become slightly shorter in
the future.
But there's no harm in adding some helpful documentation at the same time.
Map RPITIT's opaque type bounds back from projections to opaques
An RPITIT in a program's AST is eventually translated into both a projection GAT and an opaque. The opaque is used for default trait methods, like:
```
trait Foo {
fn bar() -> impl Sized { 0i32 }
}
```
The item bounds for both the projection and opaque are identical, and both have a *projection* self ty. This is mostly okay, since we can normalize this projection within the default trait method body to the opaque, but it does two things:
1. it leads to bugs in places where we don't normalize item bounds, like `deduce_future_output_from_obligations`
2. it leads to extra match arms that are both suspicious looking and also easy to miss
This PR maps the opaque type bounds of the RPITIT's *opaque* back to the opaque's self type to avoid this quirk. Then we can fix the UI test for #108304 (1.) and also remove a bunch of match arms (2.).
Fixes#108304
r? `@spastorino`
Rollup of 7 pull requests
Successful merges:
- #114099 (privacy: no nominal visibility for assoc fns )
- #114128 (When flushing delayed span bugs, write to the ICE dump file even if it doesn't exist)
- #114138 (Adjust spans correctly for fn -> method suggestion)
- #114146 (Skip reporting item name when checking RPITIT GAT's associated type bounds hold)
- #114147 (Insert RPITITs that were shadowed by missing ADTs that resolve to [type error])
- #114155 (Replace a lazy `RefCell<Option<T>>` with `OnceCell<T>`)
- #114164 (Add regression test for `--cap-lints allow` and trait bounds warning)
r? `@ghost`
`@rustbot` modify labels: rollup
When `staged_api` is enabled, effective visibilities are computed earlier
and this can trigger an ICE in some cases.
In particular, if a impl of a trait method has a visibility then an error
will be reported for that, but when privacy invariants are being checked,
the effective visibility will still be greater than the nominal visbility
and that will trigger a `span_bug!`.
However, this invariant - that effective visibilites are limited to
nominal visibility - doesn't make sense for associated functions.
Signed-off-by: David Wood <david@davidtw.co>
new unstable option: -Zwrite-long-types-to-disk
This option guards the logic of writing long type names in files and instead using short forms in error messages in rustc_middle/ty/error behind a flag. The main motivation for this change is to disable this behaviour when running ui tests.
This logic can be triggered by running tests in a directory that has a long enough path, e.g. /my/very-long-path/where/rust-codebase/exists/
This means ui tests can fail depending on how long the path to their file is.
Some ui tests actually rely on this behaviour for their assertions, so for those we enable the flag manually.
Normalize the RHS of an `Unsize` goal in the new solver
`Unsize` goals are... tricky. Not only do they structurally match on their self type, but they're also structural on their other type parameter. I'm pretty certain that it is both incomplete and also just plain undesirable to not consider normalizing the RHS of an unsize goal. More practically, I'd like for this code to work:
```rust
trait A {}
trait B: A {}
impl A for usize {}
impl B for usize {}
trait Mirror {
type Assoc: ?Sized;
}
impl<T: ?Sized> Mirror for T {
type Assoc = T;
}
fn main() {
// usize: Unsize<dyn B>
let x = Box::new(1usize) as Box<<dyn B as Mirror>::Assoc>;
// dyn A: Unsize<dyn B>
let y = x as Box<<dyn A as Mirror>::Assoc>;
}
```
---
In order to achieve this, we add `EvalCtxt::normalize_non_self_ty` (naming modulo bikeshedding), which *must* be used for all non-self type arguments that are structurally matched in candidate assembly. Currently this is only necessary for `Unsize`'s argument, but I could see future traits requiring this (hopefully rarely) in the future. It uses `repeat_while_none` to limit infinite looping, and normalizes the self type until it is no longer an alias.
Also, we need to fix feature gate detection for `trait_upcasting` and `unsized_tuple_coercion` when HIR typeck has unnormalized types. We can do that by checking the `ImplSource` returned by selection, which necessitates adding a new impl source for tuple upcasting.
interpret: Unify projections for MPlaceTy, PlaceTy, OpTy
For ~forever, we didn't really have proper shared code for handling projections into those three types. This is mostly because `PlaceTy` projections require `&mut self`: they might have to `force_allocate` to be able to represent a project part-way into a local.
This PR finally fixes that, by enhancing `Place::Local` with an `offset` so that such an optimized place can point into a part of a place without having requiring an in-memory representation. If we later write to that place, we will still do `force_allocate` -- for now we don't have an optimized path in `write_immediate` that would avoid allocation for partial overwrites of immediately stored locals. But in `write_immediate` we have `&mut self` so at least this no longer pollutes all our type signatures.
(Ironically, I seem to distantly remember that many years ago, `Place::Local` *did* have an `offset`, and I removed it to simplify things. I guess I didn't realize why it was so useful... I am also not sure if this was actually used to achieve place projection on `&self` back then.)
The `offset` had type `Option<Size>`, where `None` represent "no projection was applied". This is needed because locals *can* be unsized (when they are arguments) but `Place::Local` cannot store metadata: if the offset is `None`, this refers to the entire local, so we can use the metadata of the local itself (which must be indirect); if a projection gets applied, since the local is indirect, it will turn into a `Place::Ptr`. (Note that even for indirect locals we can have `Place::Local`: when the local appears in MIR, we always start with `Place::Local`, and only check `frame.locals` later. We could eagerly normalize to `Place::Ptr` but I don't think that would actually simplify things much.)
Having done all that, we can finally properly abstract projections: we have a new `Projectable` trait that has the basic methods required for projecting, and then all projection methods are implemented for anything that implements that trait. We can even implement it for `ImmTy`! (Not that we need that, but it seems neat.) The visitor can be greatly simplified; it doesn't need its own trait any more but it can use the `Projectable` trait. We also don't need the separate `Mut` visitor any more; that was required only to reflect that projections on `PlaceTy` needed `&mut self`.
It is possible that there are some more `&mut self` that can now become `&self`... I guess we'll notice that over time.
r? `@oli-obk`
This is necessary for closure captures in 2021 edition, as they capture individual fields, not the full mentioned variables. So it may try to capture a field of an opaque (because the hidden type is known to be something with a field).
rustdoc: handle cross-crate RPITITs correctly
Filter out the internal associated types synthesized during the desugaring of RPITITs, they really shouldn't show up in the docs.
This also fixes#113929 since we're no longer invoking `is_impossible_associated_item` (renamed from `is_impossible_method`) which cannot handle them (leading to an ICE). I don't think it makes sense to try to make `is_impossible_associated_item` handle this exotic kind of associated type (CC original author `@compiler-errors).`
@ T-rustdoc reviewers, currently I'm throwing out ITIT assoc tys before cleaning assoc tys at each usage-site. I'm thinking about making `clean_middle_assoc_item` return an `Option<_>` instead and doing the check inside of it to prevent any call sites from forgetting the check for ITITs. Since I wasn't sure if you would like that approach, I didn't go through with it. Let me know what you think.
<details><summary>Explanation on why <code>is_impossible_associated_item(itit_assoc_ty)</code> leads to an ICE</summary>
Given the following code:
```rs
pub trait Trait { fn def<T>() -> impl Default {} }
impl Trait for () {}
```
The generated associated type looks something like (simplified):
```rs
type {opaque#0}<T>: Default = impl Default; // the name is actually `kw::Empty` but this is the `def_path_str` repr
```
The query `is_impossible_associated_item` goes through all predicates of the associated item – in this case `<T as Sized>` – to check if they contain any generic parameters from the (generic) associated type itself. For predicates that don't contain any *own* generics, it does further processing, part of which is instantiating the predicate with the generic arguments of the impl block (which is only correct if they truly don't contain any own generics since they wouldn't get instantiated this way leading to an ICE).
It checks if `parent_def_id(T) == assoc_ty_def_id` to get to know if `T` is owned by the assoc ty. Unfortunately this doesn't work for ITIT assoc tys. In this case, the parent of `T` is `Trait::def` (!) which is the associated function (I'm pretty sure this is very intentional) which is of course not equal to the assoc ty `Trait::{opaque#0}`.
</details>
`@rustbot` label A-cross-crate-reexports
This option guards the logic of writing long type names in files and
instead using short forms in error messages in rustc_middle/ty/error
behind a flag. The main motivation for this change is to disable this
behaviour when running ui tests.
This logic can be triggered by running tests in a directory that has a
long enough path, e.g. /my/very-long-path/where/rust-codebase/exists/
This means ui tests can fail depending on how long the path to their
file is.
Some ui tests actually rely on this behaviour for their assertions,
so for those we enable the flag manually.
Rollup of 4 pull requests
Successful merges:
- #113887 (new solver: add a separate cache for coherence)
- #113910 (Add FnPtr ty to SMIR)
- #113913 (error/E0691: include alignment in error message)
- #113914 (rustc_target: drop duplicate code)
r? `@ghost`
`@rustbot` modify labels: rollup
Still more complexity, but this allows computing exact `NaiveLayout`s
for null-optimized enums, and thus allows calls like
`transmute::<Option<&T>, &U>()` to work in generic contexts.
THis significantly complicates `NaiveLayout` logic, but is necessary to
ensure that bounds like `NonNull<T>: PointerLike` hold in generic
contexts.
Also implement exact layout computation for structs.
Querify unused trait check.
This code transitively loads information for all bodies, and from resolutions. As it does not return a value, it should be beneficial to have it as a query.
Don't translate compiler-internal bug messages
These are not very useful to be translated, as
* translators would get really weird and bad english versions to start out from,
* compiler devs have to do some work for what is supposed to be dead code and just a sanity check,
* the target audience is other compiler devs.
r? `@davidtwco`
Make it clearer that edition functions are `>=`, not `==`
r? `@Nilstrieb`
We could also perhaps derive `Ord` on `Edition` and use comparison operators.
Add the `no-builtins` attribute to functions when `no_builtins` is applied at the crate level.
**When `no_builtins` is applied at the crate level, we should add the `no-builtins` attribute to each function to ensure it takes effect in LTO.**
This is also the reason why no_builtins does not take effect in LTO as mentioned in #35540.
Now, `#![no_builtins]` should be similar to `-fno-builtin` in clang/gcc, see https://clang.godbolt.org/z/z4j6Wsod5.
Next, we should make `#![no_builtins]` participate in LTO again. That makes sense, as LTO also takes into consideration function-level instruction optimizations, such as the MachineOutliner. More importantly, when a user writes a large `#![no_builtins]` crate, they would like this crate to participate in LTO as well.
We should also add a function-level no_builtins attribute to allow users to have more control over it. This is similar to Clang's `__attribute__((no_builtin))` feature, see https://clang.godbolt.org/z/Wod6KK6eq. Before implementing this feature, maybe we should discuss whether to support more fine-grained control, such as `__attribute__((no_builtin("memcpy")))`.
Related discussions:
- #109821
- #35540
Next (a separate pull request?):
- [ ] Revert #35637
- [ ] Add a function-level `no_builtin` attribute?
Safe Transmute: Fix ICE (due to UnevaluatedConst)
This patch updates the code that looks at the `Assume` type when evaluating if transmutation is possible. An ICE was being triggered in the case that the `Assume` parameter contained an unevaluated const (in this test case, due to a function with missing parameter names).
Fixes#110892
This patch updates the code that looks at the `Assume` type when evaluating if
transmutation is possible. An ICE was being triggered in the case that the
`Assume` parameter contained an unevaluated const (in this test case, due to a
function with missing parameter names).
Fixes#110892
Use u64 for incr comp allocation offsets
Fixes https://github.com/rust-lang/rust/issues/76037
Fixes https://github.com/rust-lang/rust/issues/95780
Fixes https://github.com/rust-lang/rust/issues/111613
These issues are all reporting ICEs caused by using `u32` to store offsets to allocations in the incremental compilation cache. This PR aims to lift that limitation by changing the offset type in question to `u64`.
There are two perf runs in this PR. The first reports a regression, and the second does not. The changes are the same in both. I rebased the PR then did the second perf run because I noticed that the primary regression in it was very commonly seen in spurious regression reports.
I do not know what the perf run will report when this is merged. I would not be surprised to see regression or neutral, but the cachegrind diffs for the regression point at `try_mark_previous_green` which is a common source of inexplicable regressions and I don't think should be perturbed by this PR.
I'm not opposed to adding a regression test such as
```rust
fn main() {
println!("{}", [37; 1 << 30].len());
}
```
But that program takes 1 minute to compile and consumes 4.6 GB of memory then writes that much to disk. Is that a concerning amount of resource use for a test?
r? `@nnethercote`
Streamline size estimates (take 2)
This was merged in #113684 but then [something happened](https://github.com/rust-lang/rust/pull/113684#issuecomment-1636811985):
> There has been a bors issue that lead to the merge commit of this PR getting purged from master.
> You'll have to make a new PR to reapply it.
So this is exactly the same changes.
`@bors` r=wesleywiser
Add support for inherent projections in new solver
Not hard to support these, and it cuts out a really big chunk of failing UI tests with `--compare-mode=next-solver`
r? `@lcnr` (feel free to reassign, anyone can review this)
This means we call `MonoItem::size_estimate` (which involves a query)
less often: just once per mono item, and then once more per inline item
placement. After that we can reuse the stored value as necessary. This
means `CodegenUnit::compute_size_estimate` is cheaper.
Resurrect: rustc_target: Add alignment to indirectly-passed by-value types, correcting the alignment of byval on x86 in the process.
Same as #111551, which I [accidentally closed](https://github.com/rust-lang/rust/pull/111551#issuecomment-1571222612) :/
---
This resurrects PR #103830, which has sat idle for a while.
Beyond #103830, this also:
- fixes byval alignment for types containing vectors on Darwin (see `tests/codegen/align-byval-vector.rs`)
- fixes byval alignment for overaligned types on x86 Windows (see `tests/codegen/align-byval.rs`)
- fixes ABI for types with 128bit requested alignment on ARM64 Linux (see `tests/codegen/aarch64-struct-align-128.rs`)
r? `@nikic`
---
`@pcwalton's` original PR description is reproduced below:
Commit 88e4d2c from five years ago removed
support for alignment on indirectly-passed arguments because of problems with
the `i686-pc-windows-msvc` target. Unfortunately, the `memcpy` optimizations I
recently added to LLVM 16 depend on this to forward `memcpy`s. This commit
attempts to fix the problems with `byval` parameters on that target and now
correctly adds the `align` attribute.
The problem is summarized in [this comment] by `@eddyb.` Briefly, 32-bit x86 has
special alignment rules for `byval` parameters: for the most part, their
alignment is forced to 4. This is not well-documented anywhere but in the Clang
source. I looked at the logic in Clang `TargetInfo.cpp` and tried to replicate
it here. The relevant methods in that file are
`X86_32ABIInfo::getIndirectResult()` and
`X86_32ABIInfo::getTypeStackAlignInBytes()`. The `align` parameter attribute
for `byval` parameters in LLVM must match the platform ABI, or miscompilations
will occur. Note that this doesn't use the approach suggested by eddyb, because
I felt it was overkill to store the alignment in `on_stack` when special
handling is really only needed for 32-bit x86.
As a side effect, this should fix#80127, because it will make the `align`
parameter attribute for `byval` parameters match the platform ABI on LLVM
x86-64.
[this comment]: #80822 (comment)
Make it clearer that we're just checking for an RPITIT
Tiny nit to use `is_impl_trait_in_trait` more, to make it clearer that we're just checking whether a def-id is an RPITIT, rather than doing something meaningful with the `opt_rpitit_info`.
r? `@spastorino`
Use maybe_body_owned_by for multiple suggestions
This is a continued work from https://github.com/rust-lang/rust/pull/113567
We have several other suggestions not working for closure, this PR use `maybe_body_owned_by` to fix them and add test cases for them.
Add a cache for `maybe_lint_level_root_bounded`
`maybe_lint_level_root_bounded` is called many times and traces node sub-paths many times. This PR adds a cache that lets many of these tracings be skipped, avoiding lots of calls to functions like `Map::attrs` and `Map::parent_id`.
r? `@cjgillot`
refactor proof tree formatting
mostly:
- handle indentation via a separate formatter
- change nested to use a closure
tested it after rebasing on top of #113536 and everything looks good.
r? `````@BoxyUwU`````
miri: protect Move() function arguments during the call
This gives `Move` operands a meaning specific to function calls:
- for the duration of the call, the place the operand comes from is protected, making all read and write accesses insta-UB.
- the contents of that place are reset to `Uninit`, so looking at them again after the function returns, we cannot observe their contents
Turns out we can replace the existing "retag return place" hack with the exact same sort of protection on the return place, which is nicely symmetric.
Fixes https://github.com/rust-lang/rust/issues/112564
Fixes https://github.com/rust-lang/miri/issues/2927
This starts with a Miri rustc-push, since we'd otherwise conflict with a PR that recently landed in Miri.
(The "miri tree borrows" commit is an unrelated cleanup I noticed while doing the PR. I can remove it if you prefer.)
r? `@oli-obk`
Introduce `trait DebugWithInfcx` to debug format types with universe info
Seeing universes of infer vars is valuable for debugging but currently we have no way of easily debug formatting a type with the universes of all the infer vars shown. In the future I hope to augment the new solver's proof tree output with a `DebugWithInfcx` impl so that it can show universes but I left that out of this PR as it would be non trivial and this is already large and complex enough.
The goal here is to make the various abstractions taking `T: Debug` able to use the codepath for printing out universes, that way we can do `debug!("{:?}", my_x)` and have `my_x` have universes shown, same for the `write!` macro. It's not possible to put the `Infcx: InferCtxtLike<I>` into the formatter argument to `Debug::fmt` so it has to go into the self ty. For this we introduce the type `OptWithInfcx<I: Interner, Infcx: InferCtxtLike<I>, T>` which has the data `T` optionally coupled with the infcx (more on why it's optional later).
Because of coherence/orphan rules it's not possible to write the impl `Debug for OptWithInfcx<..., MyType>` when `OptWithInfcx` is in a upstream crate. This necessitates a blanket impl in the crate defining `OptWithInfcx` like so: `impl<T: DebugWithInfcx> Debug for OptWithInfcx<..., T>`. It is not intended for people to manually call `DebugWithInfcx::fmt`, the `Debug` impl for `OptWithInfcx` should be preferred.
The infcx has to be optional in `OptWithInfcx` as otherwise we would end up with a large amount of code duplication. Almost all types that want to be used with `OptWithInfcx` do not themselves need access to the infcx so if we were to not optional we would end up with large `Debug` and `DebugWithInfcx` impls that were practically identical other than that when formatting their fields we wrap the field in `OptWithInfcx` instead of formatting it alone.
The only types that need access to the infcx themselves are ty/const/region infer vars, everything else is implemented by having the `Debug` impl defer to `OptWithInfcx` with no infcx available. The `DebugWithInfcx` impl is pretty much just the standard `Debug` impl except that instead of recursively formatting fields with `write!(f, "{x:?}")` we must do `write!(f, "{:?}", opt_infcx.wrap(x))`. This is some pretty rough boilerplate but I could not think of an alternative unfortunately.
`OptWithInfcx::wrap` is an eager `Option::map` because 99% of callsites were discarding the existing data in `OptWithInfcx` and did not need lazy evaluation.
A trait `InferCtxtLike` was added instead of using `InferCtxt<'tcx>` as we need to implement `DebugWithInfcx` for types living in `rustc_type_ir` which are generic over an interner and do not have access to `InferCtxt` since it lives in `rustc_infer`. Additionally I suspect that adding universe info to new solver proof tree output will require an implementation of `InferCtxtLike` for something that is not an `InferCtxt` although this is not the primary motivaton.
---
To summarize:
- There is a type `OptWithInfcx` which bundles some data optionally with an infcx with allows us to pass an infcx into a `Debug` impl. It's optional instead of being there unconditionally so that we can share code for `Debug` and `DebugWithInfcx` impls that don't care about whether there is an infcx available but have fields that might care.
- There is a trait `DebugWithInfcx` which allows downstream crates to add impls of the form `Debug for OptWithInfcx<...>` which would normally be forbidden by orphan rules/coherence.
- There is a trait `InferCtxtLike` to allow us to implement `DebugWithInfcx` for types that live in `rustc_type_ir`
This allows debug formatting various `ty::*` structures with universes shown by using the `Debug` impl for `OptWithInfcx::new(ty, infcx)`
---
This PR does not add `DebugWithInfcx` impls to absolutely _everything_ that should realistically have them, for example you cannot use `OptWithInfcx<Obligation<Predicate>>`. I am leaving this to a future PR to do so as it would likely be a lot more work to do.
Rename `adjustment::PointerCast` and variants using it to `PointerCoercion`
It makes it sounds like the `ExprKind` and `Rvalue` are supposed to represent all pointer related casts, when in reality their just used to share a little enum variants. Make it clear there these are only coercions and that people who see this and think "why are so many pointer related casts not in these variants" aren't insane.
This enum was added in #59987. I'm not sure whether the variant sharing is actually worth it, but this at least makes it less confusing.
r? oli-obk
It makes it sound like the `ExprKind` and `Rvalue` are supposed to represent all pointer related
casts, when in reality their just used to share a some enum variants. Make it clear there these
are only coercion to make it clear why only some pointer related "casts" are in the enum.
Move `TyCtxt::mk_x` to `Ty::new_x` where applicable
Part of rust-lang/compiler-team#616
turns out there's a lot of places we construct `Ty` this is a ridiculously huge PR :S
r? `@oli-obk`
Specialize `try_destructure_mir_constant` for its sole user (pretty printing)
We can't remove the query, as we need to invoke it from rustc_middle, but can only implement it in mir interpretation/const eval.
r? `@RalfJung` for a first round.
While we could move all the logic into pretty printing, that would end up duplicating a bit of code with const eval, which doesn't seem great either.
Rollup of 8 pull requests
Successful merges:
- #113010 (rust-installer & rls: remove exclusion from rustfmt & tidy )
- #113317 ( -Ztrait-solver=next: stop depending on old solver)
- #113319 (`TypeParameterDefinition` always require a `DefId`)
- #113320 (Add some extra information to opaque type cycle errors)
- #113321 (Move `ty::ConstKind` to `rustc_type_ir`)
- #113337 (Winnow specialized impls during selection in new solver)
- #113355 (Move most coverage code out of `rustc_codegen_ssa`)
- #113356 (Add support for NetBSD/riscv64 aka. riscv64gc-unknown-netbsd.)
r? `@ghost`
`@rustbot` modify labels: rollup
Winnow specialized impls during selection in new solver
We need to be able to winnow impls that are specialized by more specific impls in order for codegen to be able to proceed.
r? ``@lcnr``
Move `ty::ConstKind` to `rustc_type_ir`
Needed this in another PR for custom debug impls, and this will also be required to move the new solver into a separate crate that does not use `TyCtxt` so that r-a and friends can depend on the trait solver.
Rebased on top of #113325, only the second and third commits needs reviewing
Add some extra information to opaque type cycle errors
Plus a bunch of cleanups.
This should help users debug query cycles due to auto trait checking. We'll probably want to fix cycle errors in most (or all?) cases by looking at the current item's hidden types (new solver does this), and by delaying the auto trait checks to after typeck.
Effects/keyword generics MVP
This adds `feature(effects)`, which adds `const host: bool` to the generics of const functions, const traits and const impls. This will be used to replace the current logic around const traits.
r? `@oli-obk`
