distutils.ccompiler: Make has_function work with more C99 compilers

[fweimer-rh]

C99 removed support for implicit function declarations. This means that just calling a function, without declaring the function first, can result in a compilation error. Today, has_function works with most compilers because they issue just a warning, create an object file, and attempt a link, which then detects available of the symbol at link time, as intended. With future compilers, compilation will already fail, and no link test is performed.

The has_function interface provides the caller with a way to supply a list of header files to include. However, even with today's compilers, this only works if the function does not expect any parameters. Otherwise, the function call in the C fragment created by has_function will not supply the correct argument list and fail compilation. Therefore, this change supplies and incorrect prototype without arguments. This is what autoconf does today in a very similar situation, so it is quite likely that compilers will support this construct in this context in the future.

The includes and include_dirs arguments are deprecated because of the parameter list mismatch issue.

Fixes pypa/setuptools#3648.

[fweimer-rh]

Ping? This issue breaks setting of LIB_RT_AVAILABLE with C99 in yappi:

    if compiler.has_function('timer_create', libraries=('rt', )):
        user_macros.append(('LIB_RT_AVAILABLE', '1'))
        user_libraries.append('rt')

https://github.com/sumerc/yappi/blob/38afdacf526410f970afc58e147c7377c6c7112c/setup.py#L26

[fweimer-rh]

Here is another broken package:

ceph_version_map = collections.OrderedDict(sorted({
    "hammer": "rados_pool_get_base_tier",
    "jewel": "rados_inconsistent_pg_list",
    "kraken": "rados_aio_exec",
    "luminous": "rados_read_op_omap_get_keys2",
}.items(), key=lambda t: t[0]))


def log(msg):
    print(msg, file=sys.stderr)


def setup_hook(cmd_obj, version=None):
    if version == "latest":
        version = sorted(ceph_version_map.keys())[-1]
    elif version is None:
        comp = ccompiler.new_compiler(force=True, verbose=True)
        for potential_version, method in ceph_version_map.items():
            msg = "* checking for librados >= %s (with function %s)" % (
                potential_version, method)
            log(msg)
            found = comp.has_function(method, libraries=['rados'])
…

https://github.com/sileht/pycradox/blob/master/setup.py#L16

[fweimer-rh]

I've asked for help here:

• Need help with resolving has_function issue in distutils/setuptools

[jaraco]

I missed that this PR hadn't run CI tests successfully. The tests are failing on Windows. In 2f16327, I've marked the test as failing on Windows. Any chance you can adopt the test to also support Windows?

[fweimer-rh]

I missed that this PR hadn't run CI tests successfully. The tests are failing on Windows. In 2f16327, I've marked the test as failing on Windows. Any chance you can adopt the test to also support Windows?

Huh. Do you have a link to the build and tests logs?

[jaraco]

I was basing my assessment on this run.

[fweimer-rh]

Looks like the test is producing a.out.exe:

INFO     root:spawn.py:38 "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Tools\MSVC\14.34.31933\bin\HostX86\x64\link.exe" /nologo /INCREMENTAL:NO /LTCG /MANIFEST:EMBED,ID=1 "/LIBPATH:C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Tools\MSVC\14.34.31933\ATLMFC\lib\x64" "/LIBPATH:C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Tools\MSVC\14.34.31933\lib\x64" "/LIBPATH:C:\Program Files (x86)\Windows Kits\NETFXSDK\4.8\lib\um\x64" "/LIBPATH:C:\Program Files (x86)\Windows Kits\10\lib\10.0.22621.0\ucrt\x64" "/LIBPATH:C:\Program Files (x86)\Windows Kits\10\\lib\10.0.22621.0\\um\x64" C:\Users\RUNNER~1\AppData\Local\Temp\abortwkowtay8.obj /OUT:a.out.exe

But is trying to delete a.out:

            else:
>               os.remove(os.path.join(self.output_dir or '', "a.out"))
E               FileNotFoundError: [WinError 2] The system cannot find the file specified: 'a.out'

This looks like a pre-existing bug? This must be the first test in the test suite that expects a successful has_function result on Windows. This may be the reason why the previous abort test was in distutils/tests/test_unixccompiler.py.

It probably needs to use CCompiler.executable_filename to construct the file name to be deleted.

[fweimer-rh]

Let's see how #203 fares (I didn't even test it locally, sorry).

[mgorny]

@fweimer-rh, do you happen to have a version of this patch that applies to CPython's version of distutils? If not, don't worry. I was mostly wondering if RedHat's planning to patch the distutils from stdlib as well.

[fweimer-rh]

@mgorny I don't have such a patch yet. I'm still slightly confused whether we need it, either for our distribution package builds or pip usage.

[hroncok]

We are considering to do that. Do I assume correctly you do as well? Maybe we can convince upstream to take the change into Pytho. 3.11.3?

[mgorny]

@mgorny I don't have such a patch yet. I'm still slightly confused whether we need it, either for our distribution package builds or pip usage.

I'm not sure either. FWICS there are still a few packages that require forcing "stdlib" distutils but from a quick grep they don't seem to be using has_function().

We are considering to do that. Do I assume correctly you do as well? Maybe we can convince upstream to take the change into Pytho. 3.11.3?

Right now we're at "if it's easy, no harm in backporting it but we're not going to go out of our way to do that" ;-). Convincing upstream would be nice but I wouldn't hold my hopes high.