Repository navigation
BUG: Use of #undef complex in npy_common.h breaks extension modules that use #include <complex.h> #24344
Description
Describe the issue:
The line #undef complex in the file npy_common.h that was added in #24085 breaks the compilation of any extension module that uses both npy_common.h and the standard header <complex.h>. (See, for example, the extension module cross_gufunc.c.src in ufunclab.)
A fix would be to remove that line from npy_common.h, but that will break any code that the includes npy_common.h that also uses the name complex to define a variable or struct. This is the case for several files in NumPy, including the f2c-related files in the lapack_lite code (see my naive attempt in #24343).
Transitive includes and the global namespace in C... such fun.
Runtime information:
In [1]: import sys, numpy; print(numpy.__version__); print(sys.version)
2.0.0.dev0+766.g8627dc962
3.11.1 (v3.11.1:a7a450f84a, Dec 6 2022, 15:24:06) [Clang 13.0.0 (clang-1300.0.29.30)]
In [2]: print(numpy.show_runtime())
[{'numpy_version': '2.0.0.dev0+766.g8627dc962',
'python': '3.11.1 (v3.11.1:a7a450f84a, Dec 6 2022, 15:24:06) [Clang 13.0.0 '
'(clang-1300.0.29.30)]',
'uname': uname_result(system='Darwin', node='Warrens-MacBook-Air.local', release='22.5.0', version='Darwin Kernel Version 22.5.0: Thu Jun 8 22:21:34 PDT 2023; root:xnu-8796.121.3~7/RELEASE_ARM64_T8112', machine='arm64')},
{'simd_extensions': {'baseline': [], 'found': [], 'not_found': []}}]
None
Context for the issue:
No response
Activity
WarrenWeckesser commentedon Aug 5, 2023
In theory, the line
#undef complexcould be removed fromnpy_common.h, and all existing uses of the namecomplexin NumPy could be fixed (e.g. just change the names to, say,complex_). That would fix the NumPy build, but the change tonpy_common.hwill also break third party extension modules that happen to use the namecomplexin their code. On the other hand, we are currently breaking third party extension modules that use#include <complex.h>.rgommers commentedon Aug 6, 2023
That sounds like the right fix at first sight.
We're requiring C99 and
complexis a keyword there, so usingcomplexas a variable name seems bad practice to me. I think it's fine to require those users to fix up their code.That seems like something we should not do indeed, arguably that's a bug in NumPy.
WarrenWeckesser commentedon Aug 7, 2023
Now I see that @mattip also pointed out this problem in #24198.
seberg commentedon Aug 7, 2023
Ping @lysnikolaou for awareness.
lysnikolaou commentedon Aug 7, 2023
Agreed that we should leave the
#undefout and fix the rest of the issues. The problem with fixing the ones in #24198 is that it's in code we vendor in fromf2c, so it's unclear to me how we would go about it. I can try to have a look either way.rgommers commentedon Aug 7, 2023
I think post-processing the
f2c-generated code inside make_lite.py probably?lysnikolaou commentedon Aug 7, 2023
The problem with that is that it's not only in
f2c-generated files, but in the version off2citself that we vendor as well (in the filesf2c.handf2c.c).mattip commentedon Aug 7, 2023
There is an issue about f2c maintenance on a (the?) github mirror, where it seems the way to contact the netlib maintainer is via email (see the bottom of the README).
mattip commentedon Aug 7, 2023
with that, I don't think we have to solve all of the problems. We can do what we need to do, and point users to the potential pitfalls of using
f2c.rgommers commentedon Aug 7, 2023
We're not exposing
f2cfor anyone else to use, so I'm not sure that's all that relevant. We should be able to add anf2c.c.patchlike for some of the other.patchfiles underlinalg/lapack_lite/I'd think? Either that or post-processing via Python code, a sed script, or whatever works.lysnikolaou commentedon Aug 7, 2023
Not sure whether that's necessary. As far as I understand, the
f2c.[ch]files are just copy-pasted from the libf2c source, so that they can be used in casef2cis not installed. Just adding/modifying the patches forf2c-generated files and changingf2c.candf2c.hdirectly (maybe together with a note in the README in that directory) seems good enough for me.BUG: Remove #undef complex from npy_common.h
BUG: Remove #undef complex from npy_common.h