ICE: bincode v2 alpha cannot be used with nightly rustc #92470
Comments
kdy1
added
C-bug
|
Still ICE's on |
|
I created a minimal reproduction of the issue. This ICEs on stable as of 1.57 fn main() {
let writer = WriteImpl;
let mut encoder = EncoderImpl::<_>::new(writer);
Item.encode(&mut encoder);
}
pub trait Encode {
fn encode<E: Encoder>(&self, encoder: E);
}
pub trait Encoder {
type W: Writer;
fn writer(&mut self) -> &mut Self::W;
}
pub trait Writer {
fn write(&mut self);
}
struct Item;
impl Encode for Item {
fn encode<E: Encoder>(&self, mut encoder: E) {
encoder.writer().write();
self.encode(&mut encoder);
}
}
struct EncoderImpl<W: Writer> {
writer: W,
}
impl<W: Writer> EncoderImpl<W> {
pub fn new(writer: W) -> EncoderImpl<W> {
EncoderImpl { writer }
}
}
impl<W: Writer> Encoder for EncoderImpl<W> {
type W = W;
fn writer(&mut self) -> &mut Self::W {
&mut self.writer
}
}
struct WriteImpl;
impl Writer for WriteImpl {
fn write(&mut self) {}
}
impl<'a, T> Encoder for &'a mut T
where
T: Encoder,
{
type W = T::W;
fn writer(&mut self) -> &mut Self::W {
T::writer(self)
}
}The expectation is that the type checker should hit the recursion limit. In fact you can see this happen if you comment out the line |
|
@Aaron1011 I was told by @lqd that this should hopefully be fixed by #92210 which doesn't seem to be changing anything. |
|
Minimized a bit further: fn main() {
encode(&mut EncoderImpl);
}
pub trait Encoder {
type W;
fn writer(&self) -> Self::W;
}
fn encode<E: Encoder>(mut encoder: E) {
encoder.writer();
encode(&mut encoder);
}
struct EncoderImpl;
impl Encoder for EncoderImpl {
type W = ();
fn writer(&self) {}
}
impl<'a, T: Encoder> Encoder for &'a mut T {
type W = T::W;
fn writer(&self) -> Self::W {
panic!()
}
} |
|
(That was an incorrect guess on my part: I've seen this query in other incremental issues recently, and this actually does not seem to be related to that at all) |
|
Bisected to 50171c3 with |
Are you sure you bisected correctly? A cargo update should not have introduced it, as the bug is reproducible without any cargo involvement at all. But the ICE does seem to be a regression, as according to experiments on godbolt.org, it doesn't ICE on rustc 1.30.0, but does ICE on rustc 1.31.0 (both Zoey's shared snippet and my example).
|
|
Yeah the bisect is very weird. I tried running some versions manually that were reported as good by the tool and they still ICEd. I think |
|
Another reproducer from fn to_fn_once<F:FnOnce()>(f: F) -> F { f }
fn f<T: std::fmt::Display>(y: T) {
struct Foo<U: std::fmt::Display> {
x: U
};
let foo = Foo{ x: "x" };
let c = to_fn_once(move|| {
println!("{} {}", foo.x, y);
});
c();
c();
//~^ ERROR use of moved value
}
fn main() {
f("S");
} |
|
Triage: Fixed by #100757, I think it has a regression test that should cover this case, closing. |
|
Reopened by #104610 |
matthiaskrgr
added
the
S-bug-has-test
See: kdy1/wasm-perf#1
Code
minimal reproduction.
I'll see if I can reduce further.
Meta
rustc --version --verbose:Error output
Backtrace
The text was updated successfully, but these errors were encountered: