[clang] [clang][python][test] Check if libclang.so is loadable (PR #142353)

[Rainer Orth via cfe-commits]

rorth wrote:

> > trying to load a 32-bit libclang.so into a 64-bit python is always an error, testsuite or no.
> 
> That's why you shouldn't return an exit code of 0 when this error occurs. Moreover, this seems like a workaround for that one specific issue you encountered. To connect this with:
> 
> > Even if the error message were emitted, this doesn't help because ninja check-all is still broken, there being no way to disable check-clang-python even manually.
> 
> Wouldn't it be a better solution to wrap the `check-clang-python` target somehow as to change the exit code to 0? E.g. just pipe to `true` or `echo` or something like that. (is there an issue with that approach? I'm not that familiar with cmake)

That might work, though I'm not sure if this also works on Windows or non-Unix systems in general.  Alternatively, as I suggested, one could wrap the actual `python -m unittest discover` invocation with a check if `libclang.so` is loadable at all, only then running the actual test.  I had something like
```
import os
from clang.cindex import Config
conf = Config()
Config.set_library_path(os.environ["CLANG_LIBRARY_PATH"])
conf.lib
```
initially until I realized that doing this at `cmake` time would disable the test everywhere (no `libclang.so` exists at that point).

One problem with all these approaches, as I'd already mentioned, is that the test outcome isn't properly reported, nor can one use the usual machinery to `XFAIL` the test or declare it `UNSUPPORTED`.  That's all hardcoded inside `test/CMakeLists.txt`.

> Probably not relevant then, but is there a reason you called `os._exit` instead of `sys.exit`? `os._exit` skips exit handlers etc. so `sys.exit` is usually preferred (though I doubt it makes a difference here)

I tried that at first, but got the `warning: ` many times for every single subtest, which seems excessively verbose to me ;-)


https://github.com/llvm/llvm-project/pull/142353