Miri/CTFE fails when const generic is used as value #62790
Comments
|
@eddyb asks what pub(super) fn monomorphize<T: TypeFoldable<'tcx> + Subst<'tcx>>(
&self,
t: T,
) -> InterpResult<'tcx, T> {
match self.stack.last() {
Some(frame) => Ok(self.monomorphize_with_substs(t, frame.instance.substs)?),
None => if t.needs_subst() {
err!(TooGeneric).into()
} else {
Ok(t)
},
}
}
fn monomorphize_with_substs<T: TypeFoldable<'tcx> + Subst<'tcx>>(
&self,
t: T,
substs: SubstsRef<'tcx>
) -> InterpResult<'tcx, T> {
// miri doesn't care about lifetimes, and will choke on some crazy ones
// let's simply get rid of them
let substituted = t.subst(*self.tcx, substs);
if substituted.needs_subst() {
return err!(TooGeneric);
}
Ok(self.tcx.normalize_erasing_regions(ty::ParamEnv::reveal_all(), substituted))
}Note that the ICE is somewhere in |
|
@oli-obk do you know why the |
|
But none of those call the By identity substs I mean the |
It looks like const-prop is doing that, though: rust/src/librustc_mir/transform/const_prop.rs Lines 168 to 176 in fe499a7 |
|
That's super weird then. It's possible the offending situation propagates from earlier or something. |
jonas-schievink
added
A-const-eval
Centril
added
F-const_generics
|
@eddyb last night came to the conclusion that this line is a problem: Specifically, |
|
Fixed by #63497. |
test arrray try_from (interesting const generic usage) Currently fails, see rust-lang/rust#62435 (comment). Blocked on rust-lang/rust#62790.
Miri/CTFE fails to execute code where a const generic parameter is used as a value, such as what
try_fromfor arrays does these days:yields
The "obvious" fix seems to be to call
monomorphizeon the const before evaluating it; i did that in RalfJung@a8754d4d33. However, that ICEs:ICE backtrace
The backtrace shows that it is the first
monomorphizeineval_const_to_opthat fails. The backtrace also involves const-prop -- of course it does, there is hardly any ICE for the Miri engine that does not.@eddyb says
I am not sure what this means.^^ @oli-obk does this help to find the ICE that currently prevents supporting const-generics? It's something in const-prop again. :/
The text was updated successfully, but these errors were encountered: