ICE on "computing whether GenFuture<_> is Copy"
#95034
Comments
aliemjay
added
C-bug
rustbot
added
A-async-await
aliemjay
changed the title
Copy"Copy"
|
@rustbot claim |
|
Wow this ICE is super weird. I believe it has to do something with the Generalizer in rustc_infer, the fact that generator interiors are represented with higher-kinded types, the fact that the Gonna keep working on it for a couple of hours, but I think it might be impossible to fix currently. Or at least it requires someone a lot smarter than me. |
Is this a thing in current rust? 🤯 Honestly, I was surprised that rustc accepts trait Tr<A> { type B; type C; }
struct S<A, B, T: Tr<A, B = B>>(<T as Tr<A>>::C);Does this pattern exist in practice? People would use PhantomData and this should avoid this problem (I hope). |
Yep, see https://doc.rust-lang.org/nightly/nightly-rustc/src/rustc_infer/infer/combine.rs.html#506-530. This wf trick just ends up not working for higher-kinded types due to strange caveats... |
|
Well, either a caveat or just a bug I cannot find. |
|
I honestly don't think I understand enough about variance and the Generalizer type relation to fix this issue. I think this would technically be solved by lazy projection normalization, since this is due to a inference variable created by normalizing a projection type. I will at least add a known-bug test for this. Sorry! @rustbot release-assignment |
|
@compiler-errors have you found the exact regressed commit? It would take ages on my potato pc to bisect through commits. And thanks again for your efforts. Very much appreciated. |
|
I am pretty certain this is due to #85499 |
|
Uh this is complicated. Just trying to reduce the MCVE a bit. Here are some failed attempts:
However, I did manage to get a more reduced version: https://play.rust-lang.org/?version=nightly&mode=debug&edition=2021&gist=12b8530b4dc298dcf302f955d82fc0f1 That being said, there is a lot here that has to go perfectly (wrong).
I suspect that this hints at the root of the problem here. We should not be assigning bivariant to parameters (this happens when rustc "gives up" - and should error out, sooner than where we are getting the ice probably). It's also a bit interesting though that the call through I think this is probably just |
|
Assigning priority as discussed in the Zulip thread of the Prioritization Working Group. @rustbot label -I-prioritize +P-medium |
rustbot
added
P-medium
|
I was able to minimize it further so that I hope the new ICE provides further insight. pub trait Object<'a> { // <- can't remove `'a`
type Error; // <- can't remove because then `E`s are no longer constrained below
type Future;
}
impl Object<'static> for u8 {
type Error = ();
type Future = ();
}
impl<'a, E, A: Object<'a, Error = E>> Object<'a> for (A,) {
type Error = ();
type Future = CustomFut<'a, E, A>; // <- Replacing `E` with `A::Error` compiles
}
pub struct CustomFut<'a, E, A: Object<'a, Error = E>> {
ph: std::marker::PhantomData<(A::Future,)>, // <- adding `E` to the tuple compiles
}
pub trait AsyncFn {
fn call_async(&self);
}
impl<F: Fn() -> Fut, Fut> AsyncFn for F {
fn call_async(&self) {}
}
async fn create<T: Object<'static>>() {
let _fut: T::Future = unimplemented!();
std::future::ready(()).await; // <- any await point should work
}
pub fn test() {
let _ = create::<(u8,)>.call_async(); // <- .call_async() is important (can't remove or replace with `(...)()`; so is `(u8,)` (`u8` doesn't work)
}Error outputOutput with backtrace
|
|
Yes, I saw that error in one of my repros above. This is smaller though. |
|
@rustbot label +AsyncAwait-Triaged |
rustbot
added
the
AsyncAwait-Triaged
…jackh726 Add known-bug for rust-lang#95034 Couldn't fix the issue, since I am no type theorist and inference variables in universes above U0 scare me. But I at least wanted to add a known-bug test for it. cc rust-lang#95034 (does not fix)
…jackh726 Add known-bug for rust-lang#95034 Couldn't fix the issue, since I am no type theorist and inference variables in universes above U0 scare me. But I at least wanted to add a known-bug test for it. cc rust-lang#95034 (does not fix)
…jackh726 Add known-bug for rust-lang#95034 Couldn't fix the issue, since I am no type theorist and inference variables in universes above U0 scare me. But I at least wanted to add a known-bug test for it. cc rust-lang#95034 (does not fix)
Rollup of 7 pull requests Successful merges: - rust-lang#95102 (Add known-bug for rust-lang#95034) - rust-lang#95579 (Add `<[[T; N]]>::flatten{_mut}`) - rust-lang#95634 (Mailmap update) - rust-lang#95705 (Promote x86_64-unknown-none target to Tier 2 and distribute build artifacts) - rust-lang#95761 (Kickstart the inner usage of `macro_metavar_expr`) - rust-lang#95782 (Windows: Increase a pipe's buffer capacity to 64kb) - rust-lang#95791 (hide an #[allow] directive from the Arc::new_cyclic doc example) Failed merges: r? `@ghost` `@rustbot` modify labels: rollup
|
Fixed by #100980 |
Code
Compiled with the command
rustc --edition 2021 --crate-type lib repro.rs. Surprisingly, It is not reproducible with--crate-type bin!With any of the following changes, the code compiles fine:
impl<'a, E, A: Object<'a, Error = E>> Object<'a> for (A,) { type Error = (); - type Future = CustomFut<'a, E, A>; + type Future = CustomFut<'a, A::Error, A>; fn create() -> Self::Future { unimplemented!() }or
pub struct CustomFut<'f, E, A: Object<'f, Error = E>> { - ph: PhantomData<(A::Future,)>, + ph: PhantomData<(E, A::Future,)>, }Meta
The code compiles fine on v1.55.0. It was reproduced on v1.56.0 up to the current nightly with a similar error output.
Last good nightly:
1.56.0-nightly (b03ccace5 2021-08-24)Regressed nightly:
1.56.0-nightly (0afc20860 2021-08-25)rustc --version --verbose:Error output
Output with backtrace
@rustbot label A-async-await A-generators regression-from-stable-to-stable
The text was updated successfully, but these errors were encountered: