Re: CVS commit: pkgsrc/math/py-numpy

[Adam Ciarciński <adam%NetBSD.org@localhost>]

>>> [...]
>>> Log Message:
>>> py-numpy: updated to 1.18.4
>>> [...]
>>> @@ -24,9 +24,7 @@ PYTHON_VERSIONS_INCOMPATIBLE=       27 # since
>>> # XXX Avoid picking up other compilers when installed
>>> .include "../../mk/compiler.mk"
>>> 
>>> -.if defined(PKGSRC_FORTRAN) && ${PKGSRC_FORTRAN} == "g95"
>>> -PYSETUPBUILDARGS+=   --fcompiler=g95
>>> -.elif defined(PKGSRC_FORTRAN) && ${PKGSRC_FORTRAN} == "gfortran"
>>> +.if defined(PKGSRC_FORTRAN) && ${PKGSRC_FORTRAN} == "gfortran"
>>> PYSETUPBUILDARGS+=   --fcompiler=gnu95
>>> .elif !empty(PKGSRC_COMPILER:Mgcc)
>>> PYSETUPBUILDARGS+=   --fcompiler=gnu
>>> [...]
>> 
>> Why the logic for `g95' was removed?
>> (please also document such rationale in commit messages)
> 
> This also applies to py-scipy (g95 was removed there at the same time),
> so we'd want parity. (Or, I'd like parity, anyway. I don't know the
> rationale for the removal, either.)
> 
> Thanks,
> 
> Dave

g95 generates incorrect symbols, so these packages build but are unusable.
Try building scipy and running the tests.
As far as I remember, we agreed to remove g95 support for numpy and scipy some time ago.

Kind regards,
Adam