Reapply: [Sema] Remove some unreachable code from CSApply #84280
Conversation
hamishknight
requested review from
bnbarham,
hborla,
jckarter,
kavon,
rintaro,
slavapestov,
tshortli and
xedin
as code owners
|
@swift-ci please test source compatibility |
|
@swift-ci please SourceKit stress test |
|
Nice! |
|
After this I'd like to try and see if we can replace UnresolvedType with ErrorType |
|
Thanks! I will take a look at this one and others on monday. |
|
Oh fun, TIL we allow PlaceholderTypes in interface types for e.g FuncDecls, specifically so we can diagnose them later in MiscDiagnostics and suggest replacing with the inferred return type. That explains this existing SILGen crasher: protocol a {
func b() -> _
}
do {
func c(d : a) {
let e = d.b
}
}But it also means that the following crashes with this change: func foo() -> _ {}
foo()Looks like we need to at least record a fix to stop the constraint system from attempting to apply the solution here (as well as ensuring we also reject PlaceholderTypes in FuncDecls with no body) |
|
Fixing the above case in #84345 |
Using `*` previously meant we'd invoke undefined behavior.
It's not actually clear these are necessary anymore, and in one case it actually makes a diagnostic worse.
And ensure we don't end up with invalid key paths in CSApply.
I believe these code paths could only be reached by re-typechecking invalid code in the old CSDiag implementation.
Enforce that we don't ever encounter solutions that contain holes or error types.
hamishknight
force-pushed
the
csapply-cleanup
branch
from
bca2d6f to
3484f63
Compare
|
@swift-ci please test |
|
@swift-ci please test Linux |
|
@swift-ci please smoke test Linux |
|
swiftlang/swift-package-manager#9153 @swift-ci please smoke test Linux |
3 participants
Reapply #36028 with an even stricter set of assertions to ensure we never have a solution with errors or holes in CSApply. Depends on #84279.