fix: Implement fallback properly #20721
Conversation
|
|
||
| ctx.infer_mut_body(); | ||
|
|
||
| ctx.infer_closures(); |
There was a problem hiding this comment.
Closure analysis should be done after fallback.
025c4d8 to
69544c0
Compare
| self.table.fallback_if_possible(); | ||
| /// Clones `self` and calls `resolve_all()` on it. | ||
| // FIXME: Remove this. | ||
| pub(crate) fn fixme_resolve_all_clone(&self) -> InferenceResult { |
There was a problem hiding this comment.
I see that this clone and resolve all pattern already exists in our consteval modules.
Is this function being introduced to group necessary calls together that should be called in order to handle fallbacks correctly?
There was a problem hiding this comment.
I did not introduce new usages of the pattern. I created the function since I was changing the ordering of things we do on finish, and this pattern requires running only some passes. I also took the chance to give it a scary name 😅
There was a problem hiding this comment.
Looks good to me overall, though this has small conflict with master (Maybe #20717? 😅 )
fallback.rs was ported straight from rustc (minus the lint parts). This fixes the `!` regressions.
69544c0 to
6b133c6
Compare
|
What broke CI? |
|
It's passing now 🤷 Anyway, merging based on approval above. |
fallback.rs was ported straight from rustc (minus the lint parts).
This fixes the
!regressions.rustc uses graphs from rustc-data-structures. I opted to depend on an external dependency (
petgraph), since implementing graph handling ourselves doesn't seem like a good idea.