Improve error message for ambiguous numeric types in closure parameters #147577
Conversation
rustbot
added
S-waiting-on-review
|
r? @davidtwco rustbot has assigned @davidtwco. Use |
| err.span_suggestion_verbose( | ||
| pat.span.shrink_to_hi(), | ||
| "specify the type in the closure argument list", | ||
| format!(": &{concrete_type}"), |
There was a problem hiding this comment.
This also suggests & for PatKind::Ref(_, Mut) when it should probably suggest &mut
There was a problem hiding this comment.
This doesn't account for nested refs like &&x (should suggest : &&T not : &T)
There was a problem hiding this comment.
Good point! Added a check in dbc7327 (#147577)
(I cannot come up with a mutable test case which is caught by E0689 🤔)
There was a problem hiding this comment.
I cannot come up with a mutable test case which is caught by E0689
[&mut 0].into_iter().map(|&mut x| x.pow(0))There was a problem hiding this comment.
I think it isn't caught by E0689 but other codes like E0282 (I guess other inference errors are stronger here), so the suggestion here wouldn't be showed.
| } | ||
| // For closure parameters with reference patterns (e.g., |&v|), suggest the type annotation | ||
| // on the pattern itself, e.g., |&v: &i32| | ||
| (FileName::Real(_), Node::Pat(pat)) |
There was a problem hiding this comment.
(preexisting) Matching on the file name presumably to suppress this suggestion if this comes from a macro expansion (FileName::MacroExpansion) is wild and simply incorrect, lol. E.g., it means it never suggests anything if the source is STDIN (FileName::Anon) like for printf '/* ... */' | rustc -.
There was a problem hiding this comment.
Makes sense, but addressing it in this PR feels like out-of-scope, so added FIXME: 632e759 (#147577)
Or do you want me to resolve it here?
There was a problem hiding this comment.
Or do you want me to resolve it here?
No, a FIXME is fine 👍
| if let Node::Pat(binding_pat) = self.tcx.hir_node(hir_id) | ||
| && let hir::PatKind::Binding(..) = binding_pat.kind | ||
| && let Node::Pat(parent_pat) = parent_node | ||
| && matches!(parent_pat.kind, hir::PatKind::Ref(..)) => |
There was a problem hiding this comment.
Food for thought (I don't have time to investigate this): If the unsolved inference variable is not an integer/float one ({integer} / {float}) but a normal one, then need_type_info / InferSourceKind is able to figure out the type annotation for various kinds of patterns.
E.g.,
[].into_iter().filter(|&x| x.pow(0)); // suggests `: &T` on master[].into_iter().filter(|(x,)| x.pow(0)); // suggests `: &(T,)` on masterSo I guess ideally we would be modifying needs_type_info to properly deal with integer/float ty vars since that seems to have all the infrastructure in place already (unless they really follow a different code path in which case my suggestion would be nonsensical).
JohnTitor
force-pushed
the
issue-147312
branch
from
eb3ad28 to
632e759
Compare
4 participants
Closes #147312