getting OpenMP flags from Meson(-python) after OpenMP detection

[dimpase]

While the standard Mesonic way to configure OpenMP is by putting it in dependencies:, not by setting up flags,
can we still get them extracted in some way?

While Sage is built with Meson, we do support %%cython magic at the (i)Python prompt (a clone of https://ipython.org/ipython-doc/2/config/extensions/cythonmagic.html), which has to be configured with flags. Surely we can do our own platform detection,
but would rather use what Meson figures out via dependency('openmp') call.
Cf. sagemath/sage#40898

[dnicolodi]

Currently there is no way to serialize the properties of a Meson dependency object in a form that can be passed to something else. However, even if this would be possible, what you would get are the dependency properties (compilation flags and whatnot) valid for the system and for the compiler used to compile the project that is being configured. These most likely are different from the ones valid for the system and for the compiler used to execute the %%cython magic. Therefore, capturing the flags as used to compile the Sage components and using them to compile user code on the user system is most likely not going to work (unless Sage is distributed only in source form and all users are expected to recompile it, or Sage is supposed to have quite strict control over the environment in which it is executed).

AFAIK, the %%cython magic is based on the cythonize command line script, which uses setuptools and distutils for the actual compilation. If you want to use Meson dependency discovery capabilities, the best thing is to replace that pile of hacks with something based on meson. Most likely you would not even need meson-python for compiling and importing a single Python extension module: you can just synthesize a simple meson.build and pick the compiled extension module out of the Meson build directory.

[dnicolodi]

IIRC there have been already talks of replacing the setuptools machinery behind cythonize with something else.

[dimpase]

Thanks, it's a good point. On the other hand, one can run Meson merely to setup the data on the box the package is installed, and refrain from doing an actual build.
(there is nothing for meson compile, one only needs meson setup)

This way Meson ends up in runtime dependencies, but this doesn't look like a problem.

[dnicolodi]

On the other hand, one can run Meson merely to setup the data on the box the package is installed, and refrain from doing an actual build.

You still need to make sure that the compiler and the compilation environment used when meson is invoked are the same as when cythonize is invoked, which does not seem a trivial issue to solve: AFAIK setuptools, on which cythonize is based, does not have any public interface for passing compiler selection, compilation flags, library search paths, etc...

It seems much easier to use meson all the way than use meson to do the build setup and then find a way to transplant the information that mesom collected into the setuptools build.

In any case, I don't see how meson-python can help.

[dnicolodi]

I don't think the use case described here is valid. Closing.