Fix the emission of open-existential-metatype l-values #9852
Conversation
|
@swift-ci Please smoke test and merge. |
rdar://32288618
rjmccall
force-pushed
the
open-existential-metatype-lvalue
branch
from
6e24ca7 to
bb5fe80
Compare
|
@swift-ci Please smoke test and merge. |
|
Thanks for fixing this! There might be a simpler solution though, to have Sema coerce the base to an rvalue if its a metatype. You cannot have a mutating static method after all. |
|
That's true, but the fix seems reasonable, and I'm reluctant to add any extra requirements to CSApply's treatments of member expressions. :) |
|
@swift-ci Please smoke test. |
|
Honestly, in general I feel a lot more comfortable with doing heroics in SILGen, where we implement most of Swift's evaluation semantics anyway, than in Sema, where we often have to come up with super-awkward abstractions just to make things fit the conceit of a syntax tree. It would be much simpler if CSApply just assigned each source expression a type and then, maybe, recorded how to do a particular implicit conversion in some way that didn't interfere so much with the expression tree. |
|
Hmm. The downside is that I can't really pull this change into swift-4.0-branch without pulling at least #9637. |
rdar://32288618