error: internal compiler error: src/librustc/ty/relate.rs:753: impossible case reached: can't relate: (T,) with Lifetime('_#1r) #57156
Assignees
Labels
I-ICE
Issue: The compiler panicked, giving an Internal Compilation Error (ICE) ❄️
P-high
High priority
regression-from-stable-to-beta
Performance or correctness regression from stable to beta.
T-compiler
Relevant to the compiler team, which will review and decide on the PR/issue.
Comments
KamilaBorowska
changed the title
estebank
added
I-ICE
|
Ping @rust-lang/compiler |
|
Discussed at the T-compiler meeting. P-high to at least bisect and figure out where the regression happened. |
|
Bisected to nightly-2018-11-04, now going to attempt to bisect to a PR |
|
Specifically 3fc70e8 |
|
It's this call to Here is a backtrace with debug symbols, with rustc at 9723a49 |
|
@mikeyhew hmm, that does seem awfully early in the system to invoke the layout code, now that I think about it -- given that this is just an assertion, maybe we can move that logic later in the compiler? Or maybe i'm on the wrong trail here, and the problem is something else. |
|
@nikomatsakis By "moving that logic to later in the compiler", do you mean doing the same assertion but delaying it until later? I don't think that would get rid of the error, but I don't really know, and maybe it's worth a shot. If you meant "later" in a more general sense, there are existing checks during codegen, but @eddyb wanted to also have this check in object-safety as an additional safeguard. |
that is what I meant, but I was second guessing that after I wrote it. At minimum, it'd be worth digging in a bit more deeply to understand the ICE. |
|
Relating a type with a lifetime looks like some sort of very silly error that should be fixed. |
|
Here's what I've unwrapped so far: While determining if The error occurs during |
|
btw, more reduced and less erroneous example: trait CP<'a, T>: 'a + Fn(T) -> bool {
fn cb(&self) -> Box<dyn CP<'a, T, Output=bool>>;
}
impl<'a, T, F> CP<'a, T> for F
where F: 'a + Clone + Fn(T) -> bool
{
fn cb(&self) -> Box<dyn CP<'a, T, Output=bool>> {
Box::new(self.clone())
}
}
impl<'a, T> Clone for Box<dyn CP<'a, T, Output=bool>>
where T: 'a
{
fn clone(&self) -> Self {
self.cb()
}
}
fn main() {
let x: Box<dyn CP<i32, Output=bool>> = Box::new(move |x| x == 0);
let y = x.clone();
println!("{}", x(0));
println!("{}", y(1));
} |
|
This is the obligation: And this is the predicate being considered as a candidate to fulfill the obligation: The |
|
How did the extra lifetime get there? The substs for |
|
Even more reduced: trait Foo<Args> {
type Output;
}
trait Bar<'a, T>: Foo<T, Output=bool> {
fn cb(&self) -> Box<dyn Bar<'a, T, Output=bool>>;
}remove the The |
|
OK, I'm going to pack it in for today. Here's what we know: There's an extra lifetime parameter being added to the substs for the My best guess at this point is maybe there's a bug in the canonicalizer that introduces the new region. That's where I would look next. EDIT: but I'm definitely in way over my head here |
|
You can see that the substs is |
|
@arielb1 but where does the lifetime come from? It's an anonymous late-bound region, which I thought could be be the lifetime of the reference in FYI I'm pretty sure I won't be able to find the bug — I don't have the time, and I'm not sure what I'm looking for anyway. |
|
ok I'll have it |
arielb1
added a commit
to arielb1/rust
that referenced
this issue
Jan 5, 2019
bors
added a commit
that referenced
this issue
Jan 11, 2019
use the correct supertrait substitution in `object_ty_for_trait` beta-nominating because regression. Fixes #57156.
bors
added a commit
that referenced
this issue
Jan 11, 2019
use the correct supertrait substitution in `object_ty_for_trait` beta-nominating because regression. Fixes #57156.
pietroalbini
pushed a commit
to pietroalbini/rust
that referenced
this issue
Jan 11, 2019
|
After fixing this, |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
I tried the following code in editions 2015 and 2018:
I expected it to print:
And this does happen on stable, but not on beta or nightly, where I get this (this particular error is from nightly):
Meta:
rustc --version --verbose:Backtrace:
The text was updated successfully, but these errors were encountered: