Win meson clang

[xhochy]

Checklist

• Used a personal fork of the feedstock to propose changes
• Bumped the build number (if the version is unchanged)
• Reset the build number to 0 (if the version changed)
• Re-rendered with the latest conda-smithy (Use the phrase @conda-forge-admin, please rerender in a comment in this PR for automated rerendering)
• Ensured the license file is being packaged.

[conda-forge-webservices]

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

[h-vetinari]

Great job in getting to the fortran compilation! Now errors with

[133/1628] Compiling Fortran object scipy/special/libcdflib.a.p/cdflib_algdiv.f.obj
FAILED: scipy/special/libcdflib.a.p/cdflib_algdiv.f.obj 
"flang-new" "-Iscipy\special\libcdflib.a.p" "-Iscipy\special" "-I..\scipy\special" "-fcolor-diagnostics" "-DNDEBUG" "-D_FILE_OFFSET_BITS=64" "-Minform=inform" "-O3" "-module" "scipy\special\libcdflib.a.p" -MD -MQ scipy/special/libcdflib.a.p/cdflib_algdiv.f.obj -MF "scipy\special\libcdflib.a.p\cdflib_algdiv.f.obj.d" -o scipy/special/libcdflib.a.p/cdflib_algdiv.f.obj "-c" ../scipy/special/cdflib/algdiv.f
flang-new: error: unknown argument: '-Minform=inform'

flang-new: error: unknown argument: '-module'

flang-new: warning: scipy\special\libcdflib.a.p: 'linker' input unused

From a quick google search, -Minform=inform seems to be a PGI-ism (not even gfortran), and I cannot find where it would be added in scipy. Might be something meson is adding... CC @eli-schwartz

[rgommers]

From a quick google search, -Minform=inform seems to be a PGI-ism (not even gfortran), and I cannot find where it would be added in scipy. Might be something meson is adding... CC @eli-schwartz

It's coming from the FlangFortranCompiler class in Meson: https://github.com/mesonbuild/meson/blob/74407670e1a4fb12605e17806b53c42b0559b343/mesonbuild/compilers/fortran.py#L462. I'm guessing that it may have worked in old-style Flang, otherwise that compiler support would be completely broken. However, I can't find it in the docs. @h-vetinari if you have an old-style Flang at hand, can you check?

[rgommers]

Oh, old-style Flang was the PGI compiler pretty much, right? The returned link flags are ['-lflang', '-lpgmath'].

The -module error is similar, it's the default return value for Fortran compilers of get_module_outdir_args (same file I linked above). So I guess this is running into the limitations of pretending new-style and old-style Flang are the same.

[eli-schwartz]

I guess this is a compelling sign that supporting new flang for SciPy is going to need a pull request to meson to make it work.

If it was just the first error (-Minform) I'd suggest building with warning_level=0 to suppress meson adding any warning flags of its own (including that one).

[xhochy]

Given the problematic invocation below, what would be the correct one with new-style clang? I would then try to write a minimal translator so that we could move forward in the build at least by a bit.

"flang-new" "-Iscipy\special\libamos.a.p" "-Iscipy\special" "-I..\scipy\special" "-fcolor-diagnostics" "-DNDEBUG" "-D_FILE_OFFSET_BITS=64" "-Minform=inform" "-O3" "-module" "scipy\special\libamos.a.p" -MD -MQ scipy/special/libamos.a.p/amos_dgamln.f.obj -MF "scipy\special\libamos.a.p\amos_dgamln.f.obj.d" -o scipy/special/libamos.a.p/amos_dgamln.f.obj "-c" ../scipy/special/amos/dgamln.f

[xhochy]

Any ideas why meson / flang-new would behave differently when I add the wrapper script? It should be a pass-through.

[h-vetinari]

So I did a bit of research, and it seems flang uses -module-dir instead of -module. It's not clear to me why the latter is not supported anymore, as that's apparently what "classic" flang was using.

@eli-schwartz, do you think it's worthwhile introducing a new LLVMFlang (or whatever) flavour in meson already, where we override get_module_outdir_args appropriately? I'm almost tempted to just patch meson-in-conda-forge for now (we can even push it to a dev-label that doesn't get installed for users by default), just so we can move a bit more quickly here, as Python 3.12 is coming up like a freight train...

[eli-schwartz]

I'm fine with either solution. In the event of a conda side patch I expect it would be used to test things for later merging into meson, of course. :)