[ParseableInterfaces] Skip value witnesses of resilient conformances #20419
Conversation
|
@swift-ci Please test |
|
Build failed |
| @@ -2850,6 +2833,23 @@ void ConformanceChecker::checkNonFinalClassWitness(ValueDecl *requirement, | |||
| } | |||
| } | |||
|
|
|||
| ResolveWitnessResult | |||
| ConformanceChecker::resolveWitnessAsOpaque(ValueDecl *requirement) { | |||
| assert(!isa<AssociatedTypeDecl>(requirement) && "Use resolveTypeWitnessVia*"); | |||
There was a problem hiding this comment.
What's the reason this doesn't just take an Answered by looking below.AssociatedTypeDecl?
There was a problem hiding this comment.
The whole conformance-checking logic is split between type witnesses and non-type witnesses, and in practice it makes sense because they're resolved pretty differently. In this case, you can't actually make an associated type opaque, so the assertion makes sense.
|
Build failed |
|
Also depends on #20437. I'll rebase once that goes in. |
checkConformance was pretty long. No functionality change.
No functionality change.
We can't devirtualize through these conformances anyway, so we can get a (probably tiny) speedup by not doing any resolution of non-type witnesses. rdar://problem/43824088
jrose-apple
force-pushed
the
resolution-dissolution
branch
from
97bcb4f to
61dd307
Compare
|
@swift-ci Please test |
|
Build failed |
|
Build failed |
|
This turned out to cause some problems. I'll write up test cases this afternoon, but with #20485 there's at least not a correctness issue. Reverting. |
…rmances" This reverts commit 61dd307, the effective part of swiftlang#20419. rdar://problem/43824088
We can't devirtualize through these conformances anyway, so we can get a (probably tiny) speedup by not doing any resolution of non-type witnesses.
Based on top of #20390. I suggest reviewing commit by commit.
rdar://problem/43824088