| Age | Commit message (Collapse) | Author | Lines |
|---|
|
|
|
|
|
Miri: better document and fix dynamic const pattern soundness checks
https://github.com/rust-lang/const-eval/issues/42 got me thinking about soundness for consts being used in patterns, and I found a hole in our existing dynamic checks: a const referring to a mutable static *in a different crate* was not caught. This PR fixes that. It also adds some comments that explain which invariants are crucial for soundness of const-patterns.
Curiously, trying to weaponize this soundness hole failed: pattern matching compilation ICEd when encountering the cross-crate static, saying "expected allocation ID alloc0 to point to memory". I don't know why that would happen, statics *should* be entirely normal memory for pattern matching to access.
r? @oli-obk
Cc @rust-lang/wg-const-eval
|
|
rename-unique: Change calls and doc in raw_vec.rs
rename-unique: Change empty() -> dangling() in const-ptr-unique-rpass.rs
|
|
|
|
Const qualification for `StructuralEq`
Furthers #62411. Resolves #62614.
The goal of this PR is to implement the logic in #67088 on the MIR instead of the HIR. It uses the `Qualif` trait to track `StructuralPartialEq`/`StructuralEq` in the final value of a `const`. Then, if we encounter a constant during HAIR lowering whose value may not be structurally matchable, we emit the `indirect_structural_match` lint.
This PR contains all the tests present in #67088 and emits the proper warnings for the corner cases. This PR does not handle #65466, which would require that we be [more aggressive](https://github.com/rust-lang/rust/blob/42abbd8878d3b67238f3611b0587c704ba94f39c/src/librustc_mir_build/hair/pattern/const_to_pat.rs#L126-L130) when checking matched types for `PartialEq`. I think that should be done separately.
Because this works on MIR and uses dataflow, this PR should accept more cases than #67088. Notably, the qualifs in the final value of a const are encoded cross-crate, so matching on a constant whose value is defined in another crate to be `Option::<TyWithCustomEqImpl>::None` should work. Additionally, if a `const` has branching/looping, we will only emit the warning if any possible control flow path could result in a type with a custom `PartialEq` impl ending up as the final value of a `const`. I'm not sure how #67088 handled this.
AFAIK, it's not settled that these are the semantics we actually want: it's just how the `Qualif` framework happens to work. If the cross-crate part is undesirable, it would be quite easy to change the result of `mir_const_qualif().custom_eq` to `true` before encoding it in the crate metadata. This way, other crates would have to assume that all publicly exported constants may not be safe for matching.
r? @pnkfelix
cc @eddyb
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
normalize field projection ty to fix broken MIR issue
Fixes #71344
r? @eddyb
|
|
|
|
|
|
|
|
[breaking change] Disallow statics initializing themselves
fixes #71078
Self-initialization is unsound because it breaks privacy assumptions that unsafe code can make. In
```rust
pub mod foo {
#[derive(Debug, Copy, Clone)]
pub struct Foo {
x: (),
}
}
pub static FOO: foo::Foo = FOO;
```
unsafe could could expect that ony functions inside the `foo` module were able to create a value of type `Foo`.
|
|
|
|
|
|
|
|
|
|
Reading from the return place is fine
Const eval thinks that reading from local `_0` is UB, but it isn't. `_0` is just a normal local like any other, and codegen handles it that way too. The only special thing is that the `Return` terminator will read from it.
I've hit these errors while working on an NRVO pass that can merge other locals with `_0` in https://github.com/rust-lang/rust/pull/71003.
r? @oli-obk
|
|
|
|
Exhaustively match on `{Statement,Terminator}Kind` during const checking
This adds a pre-monomorphization error for inline assembly in a const context as well.
r? @oli-obk
|
|
Detect mistyped associated consts in `Instance::resolve`.
*Based on #71049 to prevent redundant/misleading downstream errors.*
Fixes #70942 by refusing to resolve an associated `const` if it doesn't have the same type in the `impl` that it does in the `trait` (which we assume had errored, and `delay_span_bug` guards against bugs).
|
|
|
|
Copy its value to the `return_place` upon leaving a call frame
|
|
|
|
|
|
Fix some tests failing in `--pass check` mode
r? @RalfJung
|
|
miri-unleash test for llvm_asm
r? @oli-obk @ecstatic-morse
|
|
|
|
|
|
|
|
|
|
|
|
Add `ConstKind::Error` and convert `ErrorHandled::Reported` to it.
By replicating the `ty::Error` approach to encoding "an error has occurred", all of the mechanisms that skip redundant/downstream errors are engaged and help out (see the reduction in test output).
This PR also adds `ErrorHandled::Linted` for the lint case because using `ErrorHandled::Reported` *without* having emitted an error that is *guaranteed* to stop compilation, is incorrect now.
r? @oli-obk cc @rust-lang/wg-const-eval @varkor @yodaldevoid
|
|
Miri error messages: avoid try terminology
r? @oli-obk
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
