[Sema] Don’t allow unresolved type variables if LeaveBraceStmtBodyUnchecked is true

[ahoppen]

[Sema] Don’t allow unresolved type variables if LeaveBraceStmtBodyUnchecked is true

According to Pavel, we want to eliminate allowing unresolved variables as much as possible. Removing this flag doesn’t break any test cases (at least not in a meaningful way) and fixes a crasher, so it seems reasonable to remove it.

When doing operator completion, we re-type-check the sequence expression. If we have an unresolve type already applied to ParsedExpr, which is the last element of the sequence, the type checker crashes in validation-test/IDE/crashers_2_fixed/0008-must-conform-to-literal-protocol.swift, because there are still inactive constraints in the constraint system when it finishes solving.

Previously, we would ignore because we allowed free type variables, which we no longer do since the last commit.

Fixes rdar://76686564

[ahoppen]

@swift-ci Please test

[xedin]

Is a solution with holes applied to the AST because this particular completion kind is still using old logic that relies on type-checked AST? If so, I think this fix is okay for now since in the future we’ll just stop doing AST transforms as switch to use typeCheckForCodeCompletion.

[ahoppen]

Yeah, AFAICT conforming method lists hasn’t been ported to typeCheckForCompletion yet (https://github.com/apple/swift/blob/main/lib/IDE/ConformingMethodList.cpp#L71).

[ahoppen]

@xedin Thanks for your feedback. As discussed offline, removing the AllowUnresolvedTypeVariables flag seems to fix the crash and not cause any other regressions, so it seems like a reasonable fix to me (especially since you would like to eliminate allowing unresolved type variables anyway).

[ahoppen]

@swift-ci Please test

[swift-ci]

Build failed
Swift Test Linux Platform
Git Sha - 74ff692

[swift-ci]

Build failed
Swift Test OS X Platform
Git Sha - 74ff692

[ahoppen]

@swift-ci Please test

[xedin]

Makes sense to me!

[swift-ci]

Build failed
Swift Test OS X Platform
Git Sha - 3d1f9b6

[ahoppen]

@swift-ci Please test macOS

[rintaro]

We should remove another usage of TypeCheckExprFlags::AllowUnresolvedTypeVariables in visitReturnStmt. In fact the reproducer in rdar://76686564 is not fixed without removing it (probably because it's in an implicit ReturnStmt).

[xedin]

@ahoppen @rintaro It looks like there are just a couple of usages of TypeCheckExprFlags::AllowUnresolvedTypeVariables in lib/Sema, should we try to remove all of them and see what in code completion breaks?