ICE: broken mir while building rustc #114375
Comments
matthiaskrgr
added
I-ICE
rustbot
added
the
needs-triage
|
You can also reproduce this in the rustc repo itself where will encounter the error with |
|
Reduced: // crate foo
#![feature(type_alias_impl_trait)]
type Tait = impl Sized;
fn _constrain() -> Tait {}
struct WrapperWithDrop<T>(T);
impl<T> Drop for WrapperWithDrop<T> {
fn drop(&mut self) {}
}
pub struct Foo(WrapperWithDrop<Tait>);// crate bar
pub fn drop_foo(_: foo::Foo) {}cargo rustc -p bar -- -Z validate-mir@rustbot label F-type_alias_impl_trait |
saethlin
removed
the
needs-triage
|
Found another repro I think: use std::backtrace::Backtrace;
const _: fn() = || {
let backtrace: Backtrace = Backtrace::capture();
};this one bisects to #109075 |
pub fn foo() {
std::backtrace::Backtrace::force_capture();
} |
|
@rustbot prioritize |
rustbot
added
the
I-prioritize
|
WG-prioritization assigning priority (Zulip discussion). @rustbot label -I-prioritize +P-medium |
rustbot
added
P-medium
|
Ok, after a bit more back ond forth on Zulip, I'm tentatively bumping this issue to P-critical and raise these questions (echoing the comments from the discussion):
Next week Rust 1.72 will be released, artifacts are going to be build real soon, I think. Thanks for additional comments to help improve my understanding of the context here. @rustbot label -P-medium +P-critical |
|
The underlying issue here is an issue with TAIT. #109075 was an implementation-only change simplifying some backtrace code in I have no compiler experience, but my initial hunch would be that the library change is preferable for now. I'll get a PR ready until tomorrow, unless someone beats me to it. |
wesleywiser
added
requires-nightly
compiler-errors
added
requires-nightly
|
#112777 should not be reverted. All it does is cause us to now eagerly reveal opaques that were being skipped over during normalization, which means that we're just not hitting this code anymore: rust/compiler/rustc_const_eval/src/transform/validate.rs Lines 539 to 541 in ee5cb9e It does not introduce a bug, it just reveals one that already existed. I'm gonna use the minimized example (#114375 (comment)) here. What's happening here is that we're generating a drop shim for which then tries to validate that Before #112777, we'd emit something that looks like: which is a field that is incompletely normalized, but does hit the |
|
#115005 fixes this issue by not considering mir validity on shims until they have been optimized, which uses the reveal-all param-env for that shim. This ICE shouldn't be possible to trigger without rust/compiler/rustc_mir_transform/src/pass_manager.rs Lines 120 to 122 in 6ef7d16 I don't personally think that #114988 needs to be reverted, either. |
Don't do intra-pass validation on MIR shims Fixes rust-lang#114375 In the test that was committed, we end up generating the drop shim for `struct Foo` that looks like: ``` fn std::ptr::drop_in_place(_1: *mut Foo) -> () { let mut _0: (); bb0: { goto -> bb5; } bb1: { return; } bb2 (cleanup): { resume; } bb3: { goto -> bb1; } bb4 (cleanup): { drop(((*_1).0: foo::WrapperWithDrop<()>)) -> [return: bb2, unwind terminate]; } bb5: { drop(((*_1).0: foo::WrapperWithDrop<()>)) -> [return: bb3, unwind: bb2]; } } ``` In `bb4` and `bb5`, we assert that `(*_1).0` has type `WrapperWithDrop<()>`. However, In a user-facing param env, the type is actually `WrapperWithDrop<Tait>`. These types are not equal in a user-facing param-env (and can't be made equal even if we use `DefiningAnchor::Bubble`, since it's a non-local TAIT).
wesleywiser
removed
the
I-compiler-nominated
Code
rustc -Zvalidate-mir file.rsMeta
rustc --version --verbose:Error output
Not posting the entire 1.7K lines backtrace 🙃
Backtrace
The text was updated successfully, but these errors were encountered: