infer: update resolver when descending into block #7627
Conversation
| enum Either { A(bool), B } | ||
| match Either::B { | ||
| //^^^^^^^^^ Missing match arm | ||
| Either::A(true | false) => (), |
There was a problem hiding this comment.
Not sure this is what this test intends to check
There was a problem hiding this comment.
Looks like it checks that there's a "missing match arm" diagnostic because Either::B isn't handled. But the previous comment was indeed a bit strange. Is it actually working now?
crates/hir_ty/src/infer/expr.rs
Outdated
| } else { | ||
| ty | ||
| Expr::Block { statements, tail, label, id: _ } => { | ||
| self.resolver = resolver_for_expr(self.db.upcast(), self.owner, tgt_expr); |
There was a problem hiding this comment.
Can we push directly a new scope onto the Resolver, then pop it at the end of the block, so we reuse it?
This feels a bit iffy -- we replace it, but keep using the new one even after the end of the block. Can't that lead to problems?
There was a problem hiding this comment.
Oh, yeah, I meant to swap the old one back
There was a problem hiding this comment.
Can we push directly a new scope onto the
Resolver, then pop it at the end of the block, so we reuse it?
We can do that, but that sort of feels like mixing concerns. I think type inference shouldn't really know about name resolution scoping rules.
Perhaps a nicer solution would be to have an HIR body visitor that gives you the correct Resolver for every expression?
There was a problem hiding this comment.
Maybe. Or at least an API like self.resolver = resolver_for_expr(self.db.upcast(), self.owner, tgt_expr, self.resolver); /* ... */ self.resolver.pop();?
(I'm pretty sure that won't work because of the borrow checker, but you know what I mean)
|
bors r+ |
No description provided.