Fix inference to intersections #30857
Conversation
| inferFromTypes(source, t); | ||
| // we want to infer string for T, not Promise<string> | string. For intersection types | ||
| // we only infer to single naked type variables. | ||
| if (target.flags & TypeFlags.Union ? typeVariableCount !== 0 : typeVariableCount === 1) { |
There was a problem hiding this comment.
The old comment on this behavior was
// Next, if target containings a single naked type variable, make a secondary inference to that type
// variable. This gives meaningful results for union types in co-variant positions and intersection
// types in contra-variant positions (such as callback parameters).should we perhaps be checking the contravariant flag here, rather than (or perhaps alongside) count?
There was a problem hiding this comment.
I don't think the contravariant flag is relevant here (or maybe I'm just not seeing how).
There was a problem hiding this comment.
The original in-code comment on this behavior said that it was inferences to intersections in contravariant positions (see above) that gave good results - it's possible that we could capture that sentiment by inferring to only one intersection parameter in covariant positions, but all of them in contravariant positions (and vice-versa for unions, I imagine?)
I dunno really; this is just speculation based on the text of a comment we removed, and assuming that there was insight in that comment which our tests didn't quite capture - without concrete examples, though, I'm not really sure.
…section Fix inference to intersections # Conflicts: # src/compiler/checker.ts
…30968) Fix inference to intersections
Fixes #30442.