allow specifying non-default Fortran compiler with FC env var #13462

Activity

1. rgommers

   added

   01 - Enhancement

   component: numpy.distutils

   on May 3, 2019
2. isuruf commented on May 3, 2019

   

   isuruf

   on May 3, 2019

   Contributor

   More actions

   flang is not added to default_compiler

   numpy/numpy/distutils/fcompiler/__init__.py

   Line 747 in 7dbb325

   _default_compilers = (

   
   and is only defined for python 3 for some weird reason,

   numpy/numpy/distutils/fcompiler/pg.py

   Line 67 in b1d7001

   if sys.version_info >= (3, 5):
3. dimpase commented on May 12, 2020

   

   dimpase

   on May 12, 2020

   Contributor

   More actions

   Another error I am getting with flang in OpenBSD 6.6 (based on clang 8.0.1) is that it attempts to use it with -g flag, which results in inability to compile anything with a named COMMON block (a known flang bug), e.g. an attempt to build scipy ends with

       compiling Fortran sources
       Fortran f77 compiler: flang -Wall -g -ffixed-form -fno-second-underscore -I/usr/local/include -fPIC
       Fortran f90 compiler: flang -Wall -g -fno-second-underscore -I/usr/local/include -fPIC
       Fortran fix compiler: flang -Wall -g -ffixed-form -fno-second-underscore -Wall -g -fno-second-underscore -I/usr/local/include -fPIC
       creating build/temp.openbsd-6.6-amd64-3.7/scipy/integrate
       creating build/temp.openbsd-6.6-amd64-3.7/scipy/integrate/mach
       compile options: '-I/home/dima/sagetrac-mirror/local/lib/python3.7/site-packages/numpy/core/include -c'
       flang:f77: scipy/integrate/mach/d1mach.f
       flang: warning: argument unused during compilation: '-fno-second-underscore' [-Wunused-command-line-argument]
       /tmp/d1mach-f87f56.ll:894:17: error: expected metadata type
       !121 = distinct !DICommonBlock(scope: !17, declaration: !126, name: "d9mach")
                       ^
   

   Given that the only other Fortran provided on OpenBSD is g95 (very old), it's a show-stopper.

   Is there a graceful way to remove -g? If not, what should I patch?
4. dimpase commented on May 12, 2020

   

   dimpase

   on May 12, 2020

   Contributor

   More actions

   A related question - what is the need for that COMMON block? It does not seem to be used anywhere in scipy. Can it just be removed?
5. rgommers commented on Nov 21, 2022

   

   rgommers

   on Nov 21, 2022

   MemberAuthor

   More actions

   This is another feature request for numpy.distutils. Bug reports about numpy.distutils are still needed until it is completely removed from numpy, however feature requests we can close now.
6. rgommers

   closed this as completedon Nov 21, 2022
7. rgommers commented on Nov 21, 2022

   

   rgommers

   on Nov 21, 2022

   MemberAuthor

   More actions

   I'll also note that Meson already does the right thing here.