internal: add_explicit_type respects coercions
#9552
Conversation
| #[test] | ||
| fn add_explicit_type_inserts_coercions() { | ||
| // This should set `*const [u32]` | ||
| check_assist( | ||
| add_explicit_type, | ||
| r#" | ||
| //- minicore: coerce_unsized | ||
| fn f() { | ||
| let $0x: *const [_] = &[3]; | ||
| } | ||
| "#, | ||
| r#" | ||
| fn f() { | ||
| let x: *const [_] = &[3]; | ||
| } | ||
| "#, | ||
| ); | ||
| } |
There was a problem hiding this comment.
So for some reason the coercions here are all slices of type variables instead of a conrete type(even if using a 3u32 literal, forgot to add that to the commit). Is this expected?
There was a problem hiding this comment.
Oh, we need to resolve the types in adjustments after inference, see this function: https://github.com/rust-analyzer/rust-analyzer/blob/6f3a3bdde4e526d880b9d8d9c613d2cef50a5f3b/crates/hir_ty/src/infer.rs#L390
There was a problem hiding this comment.
Oooh ye that totally makes sense. Though out of interest why do we lose the concrete type information for the array/slice there even if it is known before the coercion? I would expect that to persist in this specific case.
There was a problem hiding this comment.
The type we coerce to is *const [_], which contains a type variable. The type variable gets unified with the u32, but to actually see that we have to resolve it in the end.
|
bors r+ |
| @@ -407,6 +407,12 @@ impl<'a> InferenceContext<'a> { | |||
| for (_, subst) in result.method_resolutions.values_mut() { | |||
| *subst = self.table.resolve_completely(subst.clone()); | |||
| } | |||
| for adjustment in result.expr_adjustments.values_mut().flatten() { | |||
There was a problem hiding this comment.
We can avoid the clones here and above using mem::replace, do you think it's worth it?
There was a problem hiding this comment.
No, cloning Chalk types is cheap.
There was a problem hiding this comment.
Yeah, I suppose we'd still have to intern (or at least clonel the error type.
or so I'd like to say but there is one odd case here where it doesn't work(see review)
Fixes #6107