Sema: Accept tautological assoc type inference candidates when same-typed. #7875
Conversation
|
@swift-ci Please test |
|
@DougGregor or @slavapestov, does this look OK? Is there a better way to ask a generic environment whether an associated type is same-typed to another one than trawling the requirements? |
lib/Sema/TypeCheckProtocol.cpp
Outdated
| dmt->getAssocType()); | ||
| for (auto &reqt : witness->getDeclContext() | ||
| ->getGenericSignatureOfContext() | ||
| ->getCanonicalSignature() |
There was a problem hiding this comment.
getCanonicalSignature() won't change the signature, it just canonicalizes the types themselves; you don't need that here
|
Build failed |
lib/Sema/TypeCheckProtocol.cpp
Outdated
| // If this associated type is same-typed to another associated type | ||
| // on `Self`, then it may still be an interesting candidate if we find | ||
| // an answer for that other type. | ||
| if (witness->getDeclContext()->getAsProtocolExtensionContext() |
There was a problem hiding this comment.
If you want to check that two specific associated types are equal, you can map both into context and compare archetypes for pointer equality.
To see if an associated type is equal to some other associated type, the only way that's slightly saner than yours is to get the canonical generic signature builder for the signature, and ask it for the equivalence class of the associated type. But I don't know if the right APIs are public or sane.
So what you have here might be your best bet.
lib/Sema/TypeCheckProtocol.cpp
Outdated
| auto selfTy = witness->getDeclContext()->getSelfInterfaceType(); | ||
| auto selfAssocTy = DependentMemberType::get(selfTy, | ||
| dmt->getAssocType()); | ||
| for (auto &reqt : witness->getDeclContext() |
There was a problem hiding this comment.
I feel like factoring out 'witness->getDeclContext()' and 'dc->getGenericSignatureOfContext()' would make this code easier to read.
…yped.
`X := Self.X` is still interesting information if it came from a witness candidate in a context where `X == Y` and we can infer a concrete witness for Y, for example:
```swift
protocol SameTypedDefault {
associatedtype X
associatedtype Y
static var x: X { get }
static var y: Y { get }
}
extension SameTypedDefault where Y == X {
static var x: X {
return y
}
}
struct UsesSameTypedDefault: SameTypedDefault {
static var y: Int {
return 0
}
}
```
Defer checking whether such a witness candidate satisfies associated type requirements until we consider a complete solution system, since `ConformingType.X` isn't a valid type until we've settled on a witness for X, and its abilities depend on what we choose for Y. Fixes SR-4143.
jckarter
force-pushed
the
assoc-type-inference-same-type-candidate
branch
from
4addc8f to
cdc3c1b
Compare
|
@swift-ci Please test |
1 similar comment
|
@swift-ci Please test |
|
@slavapestov Thanks, I updated the patch in response to your feedback. How's this look? Reasonable for 3.1? |
|
LGTM |
X := Self.Xis still interesting information if it came from a witness candidate in a context whereX == Yand we can infer a concrete witness for Y, for example:Defer checking whether such a witness candidate satisfies associated type requirements until we consider a complete solution system, since
ConformingType.Xisn't a valid type until we've settled on a witness for X, and its abilities depend on what we choose for Y. Fixes SR-4143.