[android][test] Disable new SwiftToCxxToSwift Interop test

[finagolfin]

@drodriguez, this should get the Android CI green again, we've already disabled several Interop tests that show this mbstate_t error.

[drodriguez]

@swift-ci please test and merge

[finagolfin]

Looks like Mac CI broke in the morning, unrelated to this pull, CI will need to be run again.

[egorzhdan]

@swift-ci please smoke test macOS

[finagolfin]

Looks like the Mac CI was broken again yesterday evening, will need to be run again.

[hyp]

What's the mbstate_t error out of curiosity?

[hyp]

@swift-ci please smoke test macOS

[finagolfin]

@hyp, this one, several of the C++ Interop tests are currently disabled for Android with that same error:

/home/ubuntu/android-ndk-r19c/toolchains/llvm/prebuilt/linux-x86_64/sysroot/usr/include/uchar.h:66:57: error: missing '#include <bits/mbstate_t.h>'; 'mbstate_t' must be declared before it is used
size_t c16rtomb(char* _Nullable __buf, char16_t __ch16, mbstate_t* _Nullable __ps) __INTRODUCED_IN(21);
                                                        ^
/home/ubuntu/android-ndk-r19c/toolchains/llvm/prebuilt/linux-x86_64/sysroot/usr/include/bits/mbstate_t.h:47:3: note: declaration here is not visible
} mbstate_t;
  ^

[hyp]

I see, that's just bad SDK then :D

[finagolfin]

Yeah, there's some issue with how the Bionic libc module map we use interacts with the Android libc++ modulemap and Swift now, which can be seen in #60272 also, that I'll need to look into at some point. Strangely, I didn't find any way to get clang to list what symbols are actually included in different modules, other than dumping the pcm files manually.

@hyp, is there any better way that you use for such module debugging? I tried -module-file-info but that doesn't list all the symbols either.

[hyp]

Not particularly. You would just need to figure out what module the mbstate_t.h header is in. It sounds like it might be just an issue in uchar.h - it should either forward declare mbstate_t or include that header, so it sounds like an SDK bug, not module map bug.

[finagolfin]

No, it can't be an SDK issue, this isn't some fly-by-night SDK here: we're talking about the most widely used libc in the world in Bionic. I just looked and uchar.h in glibc does something similar, including a bits/types/mbstate_t.h, which is similar to what Bionic does too. Yet, it works for glibc and not for Bionic when building here with Swift.

Obviously there are other differences between the two libcs too, for example, Swift provides its own libstdc++ module map that's used on linux, while the Android CI uses the libc++ module map that the Android NDK provides.

@hyp, if you have any input on how I might track this down, I'd let to get this fixed before the 5.8 branch, particularly so that I can start including the Swift portions of the Swift compiler itself again, as I had to disable them with #60272.

[finagolfin]

Whoo, Android CI is green again, for the first time in months! I don't expect it to stay this way for long, with the constant C++ Interop test regressions- I see 14 that are now disabled, including this one, and another 10 or so that fail natively- and diminished hardware resources on the CI, but at least we got it passing again.