Harden IDE some more #2932
Harden IDE some more #2932
Conversation
no errors were reported.
We get sometimes stake symbol error because a completer is run in its creation context, which comes from a previous run. This should not happen. Adding an assert to flag the erronenous completion when it happens.
| val effectiveOwner = | ||
| if (curOwner.isTerm && defDenot.symbol.isType) | ||
| // Don't mix NoPrefix and thisType prefixes, since type comparer | ||
| // would not detect types to be compatible. Note: If we replace the |
There was a problem hiding this comment.
What kind of issue did you see before this change? Could this be boiled down to a test case?
There was a problem hiding this comment.
It's only in the applied type branch that this became an issue. We saw the same type parameter TypeRef once with a NoPrefix prefix, the other with a ThisType. They should be the same but subtype checking failed. We could make subtype checking more lenient, but I think it's better not to have variations how we construct these things. I believe the current scheme hides the difference because we follow an alias from TypeParamAccessor to TypeParam, that's why we did not see it so far.
|
Rebased to master. |
Parser could get into an infinite loop because `acceptStatSepUnlessAtEnd`
did not make progress. I believe the concrete example was a string interpolator
where we encounter a `{ ... )`.
|
I collected all IDE improvements here. Will submit the other improvements separately. |
More fixes that prevent crashes and infinite loops observed in the IDE.
Also, some more fundamental changes that I added here to test them out in isolation of the other work on applied types I am doing.