Account for the generic zero-sized payload enum cases.

[hjyamauchi]

This fixes an assert failure:

// At least one payload is non-empty (otherwise this
// would get laid out as a non-payload enum)
assert(getNumNonEmptyPayloadCases() > 0);

in TaggedMultiPayloadEnumTypeInfo when you have a generic but zero-sized payload enum cases.

[compnerd]

@swift-ci please test

[tbkka]

I'm not sure about this. My understanding is that the compiler treats all generic payloads as if they were non-zero-sized for the purposes of determining a layout. Is this right, @slavapestov ?

If my understanding is correct, then it's probably the assert here that's actually wrong, not the layout determination logic. I'll take a look.

[tbkka]

I did a little bit of poking around, and I'm pretty sure that:

• This PR is actually wrong
• (The test case is valuable; we should keep that)
• The assert is wrong and needs to be removed (or even better, commented out with an explanation of why it's wrong)

[tbkka]

I believe the changes to TypeLowering.cpp here are all wrong.

However, the compiler support that this is trying to match is complex; it is quite likely that I don't entirely understand it either. I'd be happiest if you could come up with a test case that actually shows an incorrect enum reflection for some version of this (not just the failing assertion). See the tests in validation-tests/Reflection for examples of how to verify enum reflection.

[tbkka]

With your changes, validation-tests/Reflection/reflect_Enum_MultiPayload_generic7.swift and .../reflect_Enum_MultiPayload_generic8.swift both fail on my local system. Do you also see this?

[tbkka]

These cases also failed in CI. Only for macOS ... I wonder if these tests even run on Linux/Windows?

Failed Tests (3):
  Swift-validation(macosx-x86_64) :: Reflection/reflect_Enum_MultiPayload_generic2.swift
  Swift-validation(macosx-x86_64) :: Reflection/reflect_Enum_MultiPayload_generic7.swift
  Swift-validation(macosx-x86_64) :: Reflection/reflect_Enum_MultiPayload_generic8.swift

[compnerd]

https://github.com/apple/swift/blob/main/validation-test/Reflection/reflect_Enum_MultiPayload_generic2.swift#L8
https://github.com/apple/swift/blob/main/validation-test/Reflection/reflect_Enum_MultiPayload_generic7.swift#L8

So, generic2 and generic7 do not run on Windows, generic8 does though.

[tbkka]

So, generic2 and generic7 do not run on Windows, generic8 does though.

That looks like a lazy copy/paste. I don't see any reason those two tests would actually require ObjC interop. Can someone try removing those lines and see if they work today?

[hjyamauchi]

Update this PR to comment out the assert with a comment and to move the test to validation_test. Also removed the objc_interop requirements from several enum reflection tests.

Yes I was able to see the failures with the original PR.
Yes the enum reflection tests didn't need objc_interop.

PTAL.

[tbkka]

Nice! Thank you so much for sorting through all those tests and getting them to run on Windows and Linux!

[tbkka]

@swift-ci Please test

[hjyamauchi]

I was able to repro those failures on Linux. Updated the PR to disable them. PTAL. Can you rerun the CI?

[tbkka]

@swift-ci Please test

[hjyamauchi]

The windows test timed out. Can anyone rerun it?

[hjyamauchi]

@tbkka Would you rerun the windows CI? This is blocking a debugger scenario on windows that we'd like to fix soon.

[tbkka]

@swift-ci Please test Windows

[hjyamauchi]

@tbkka Thanks! Do you mind merging this?

[tbkka]

Done! Thanks for following through on this!

[hjyamauchi]

Thank you so much!