Sema: Better prepared overloads #84339
Conversation
slavapestov
force-pushed
the
better-prepared-overloads
branch
from
3a0ab19 to
1ea7382
Compare
slavapestov
requested review from
AnthonyLatsis,
hborla and
xedin
as code owners
|
@swift-ci Please test |
|
@swift-ci Please test source compatibility |
|
@swift-ci Please test Linux |
… look like FuncDecl case
getTypeOfMemberReference() already opens the accessor's thrown error type for us.
slavapestov
force-pushed
the
better-prepared-overloads
branch
from
1ea7382 to
98d0ea5
Compare
slavapestov
requested review from
artemcm,
tshortli and
DougGregor
as code owners
…t Post() versions
|
swiftlang/swift-package-manager#9150 |
|
@swift-ci Please smoke test macOS |
|
@swift-ci Please test Windows |
|
@swift-ci Please test source compatibility |
| ASSERT(type && "Expected non-null type"); | ||
|
|
||
| if (preparedOverload) { | ||
| preparedOverload->recordedNodeType(node, type); |
There was a problem hiding this comment.
Because property wrapper application happens as part of getTypeOfMemberReference() and it calls setType(), so we have to record it and play it back.
| cs.addConstraint(ConstraintKind::Bind, typeVar, subst(gp), locator); | ||
| for (auto [gp, typeVar] : replacements) { | ||
| cs.addConstraint(ConstraintKind::Bind, typeVar, subst(gp), locator, | ||
| /*isFavored=*/false, preparedOverload); |
There was a problem hiding this comment.
I'm not a big fan of the fact that we have to add prepared overloads to all these various APIs, including addConstraint. I think prepared overloads belong in the solver not in the constraint system.
There was a problem hiding this comment.
Once getTypeOf{Member,}Reference() are always called with preparedOverload != nullptr, I'm planning on changing a bunch of APIs that currently take an optional preparedOverload to be methods on the PreparedOverloadBuilder itself, since in those places we'll always be able to call it on a builder that is provided. I agree it is annoying to have to plumb it through to intercept various places that would otherwise record trail changes, but I hope to separate the two code paths completely (edit: well, probably not 100%) at some point.
There was a problem hiding this comment.
Alright, let's see how it turns out.
Also, move some methods from ConstraintSystem.h to ConstraintSystem.cpp.
slavapestov
force-pushed
the
better-prepared-overloads
branch
from
98d0ea5 to
e9070cf
Compare
|
@swift-ci Please smoke test |
|
@swift-ci Please test source compatibility |
|
@swift-ci Please smoke test macOS |
…verloads Sema: Better prepared overloads
This PR is another intermediate step on my quest to simplify getTypeOfReference()/getTypeOfMemberReference().
These functions are used in exactly two places: when applying a disjunction choice inside the solver, and in protocol witness matching. I cleaned up the latter code path a bit to make it look more like what happens in the solver, and this allowed me to simplify the parameter lists of both functions a fair bit; both take an OverloadChoice now.
The Pre()/Post() split is transitional and I'm going to re-organize this a little bit more, but I'd like to land this separately and test source compatibility first.