[Sema][CSApply] Be more conservative while emitting always true conversion diagnostic

[LucianoPAlmeida]

Be more conservative on always true conversion diagnostics when functions are involved.

Resolves SR-13789.

[tbkka]

@hborla -- Do you think this is the right approach? The question here is how to best deal with the fact that function casts behave differently in the runtime and compiler (the compiler can perform argument/return type conversions that the runtime cannot perform). The previous code emitted some inappropriate diagnostics because of this distinction.

[LucianoPAlmeida]

@hborla -- Do you think this is the right approach? The question here is how to best deal with the fact that function casts behave differently in the runtime and compiler (the compiler can perform argument/return type conversions that the runtime cannot perform). The previous code emitted some inappropriate diagnostics because of this distinction.

@tbkka I don't have a deep understanding of the runtime, but from the tests I ran, the runtime couldn't perform any cast other than when the function types were equal, is that really the case? Or there are other situations when the runtime cast is possible?
Because if this always fails, I thought that maybe we can emit a tailored "runtime cannot perform, cast always fails" warning, but not sure, I end up being more conservative to avoid false positives.

[tbkka]

The runtime can only cast functions when the arguments and return type are exactly identical. The only permitted difference is in throwing. You can read the details here:
https://github.com/apple/swift/blob/d5d9e48d37654054cd6fdcc834fd98c54e64a051/stdlib/public/runtime/DynamicCast.cpp#L1167

[LucianoPAlmeida]

Awesome! Thanks for the reference :)
We can definitely detect those situations and otherwise emit a warning that the runtime cannot convert the types. WDYT?

[xedin]

I think makes sense to align solver with runtime but instead of CSApply do more checking simplifyCheckedCastConstraint instead and record an appropriate warning there.

[LucianoPAlmeida]

This is not ready yet, as it still has a lot of things to fix as it requires some changes involving TypeChecker::typeCheckCheckedCast, I'll ping you as soon as I can make this work :)

[LucianoPAlmeida]

@swift-ci Please smoke test

[xedin]

I think we are almost there, just a couple more minor comments.

[LucianoPAlmeida]

@xedin Thanks for the review! I think all points were addressed :)

[LucianoPAlmeida]

@swift-ci Please smoke test

[LucianoPAlmeida]

@swift-ci Please smoke test

[LucianoPAlmeida]

There is a lot of commits, I think will be a good idea just to squash this one. WDYT?

[xedin]

Sounds good to me!

[LucianoPAlmeida]

Squashed! Thank you @xedin :)