Sema: Respect TVO_CanBindToPack #61777
Conversation
slavapestov
force-pushed
the
respect-the-pack
branch
from
7289baa to
0b8927c
Compare
| // can't merge these two type variables. | ||
| if (rep1->getImpl().canBindToPack() | ||
| != rep2->getImpl().canBindToPack()) | ||
| return getTypeMatchFailure(locator); |
There was a problem hiding this comment.
Instead of failing should this just from an unsolved constraint and wait until one of type variables is resolved so it could be fixed?
There was a problem hiding this comment.
TVO_CanBindToPack should never change. It's set when the type variable is created in CSGen.
There was a problem hiding this comment.
That's reasonable but we cannot diagnose here and at the same time cannot fail right away either I think, we need to waiting into more is know about one of the types or it's marked as a placeholder...
There was a problem hiding this comment.
Right, we can't unconditionally fail. When Pavel says "so it could be fixed", I think he means that this constraint needs to ultimately return SolutionKind::Solved in diagnostic mode with some kind of fix recorded. The solver should never fail solving a constraint in diagnostic mode.
There was a problem hiding this comment.
should never fail solving a constraint in diagnostic mode
Shouldn't fail unless it leads to a worse solution i.e. if the impact of the fix is greater than previous solutions that have been discovered.
| @@ -4307,6 +4307,14 @@ ConstraintSystem::matchTypesBindTypeVar( | |||
| } | |||
| } | |||
|
|
|||
| if (type->is<PackType>() || type->is<PackArchetypeType>()) { | |||
| if (!typeVar->getImpl().canBindToPack()) | |||
| return getTypeMatchFailure(locator); | |||
There was a problem hiding this comment.
I think we need a bi-directional fix here similar to fixReferenceMismatch otherwise constraint solver is just going to fail with "failed to produce a diagnostic" because it wouldn't be able to form a valid solution due to this binding always failing.
There was a problem hiding this comment.
I'll let @hborla handle this one :) I was just rebasing this PR so that she could finish it off.
There was a problem hiding this comment.
Yes, I'm going to pick this up and address the feedback about diagnostics!
|
Subsumed by #64498 |
No description provided.