fix: check uncertain subst when unsized coerce #20608
Conversation
rustbot
added
the
S-waiting-on-review
| @@ -716,7 +717,7 @@ impl<'db> InferenceTable<'db> { | |||
| let goal: Goal = coerce_unsized_tref.cast(Interner); | |||
|
|
|||
| self.commit_if_ok(|table| match table.solve_obligation(goal) { | |||
| Ok(Certainty::Yes) => Ok(()), | |||
| Ok((_, Certainty::Yes) | (HasChanged::Yes, Certainty::Maybe(_))) => Ok(()), | |||
There was a problem hiding this comment.
Sorry for the late reply. Has been quite busy on migrating more things to next-solver 😅
I think we shouldn't accept all ambiguous solutions here. If we success here, we skip for the remaining structural coercion attempts tried after this, so this might regress other type inferences or be false negative on type mismatches.
IMO we have to check extra conditions for the trait solve result like in rustc
or try this even better WIP implementation rust-lang/rust#141926
There was a problem hiding this comment.
I added this simply because what r-a used to do in chalk based trait solver, which accpets incomplete but not identical subst. It's far from perfect though.
There was a problem hiding this comment.
But it seems that rustc current impl isn't too hard to follow.
There was a problem hiding this comment.
Yes, the previous rust-analyzer had been doing so but I hope we could do better 😄
There was a problem hiding this comment.
But r-a doesn't have any proof tree related infra at the moment.
There was a problem hiding this comment.
Yes, we have to implement it, too.
Actually, that's why I worked on #20578 before solving this issue.
There was a problem hiding this comment.
Actually, I setup the proof tree infra in my (to be published) coercion PR.
related: #20422