On Thu, Feb 03, 2022 at 07:36:21PM +0000, jmeis...@posteo.ch wrote:
> Hi,
> 
> I'd like to propose an update to current math/hdf5 from 1.12.0 to
> 1.12.1. This removes the need of all patches related to non-existing
> $(DESTDIR) at the time of 1.12.0 version.
> All tests PASSED (make test + gmake check-install in ports obj dir).
> The only new patch added is a missing CLOSE statement
> (hdf5-current has corrected this problem so it won't stay for next
> version bump).
> 
> Tested on -current, amd64 only.
> 
> Also, with the port version of hdf5 (1.12.0), fortran
> binary bin/h5fc is linked to "gfortran" compiler name instead of
> egfortran (which gets built/installed from source with
> modules=fortran, at least on amd64 for now until flang comes,
> with the name "egfortran").
> 
> $ h5fc -show
>  gfortran -I/usr/local/include ..........
> 
> which results in a "missing gfortran" error. (gmake check-install is
> failing)
> 
> In the diff below, I specified the env variable FC=egfortran to get:
> 
> $ h5fc -show
> egfortran -I/usr/local/include ..............
> 
> (gmake check-install is now PASSED)
> 
>  but I'm wondering if that's the right way of dealing with this issue.
> As MODFORTRAN_COMPILER is reporting 'gfortran' by default,
> we cannot use that to get the correct name for FC env variable.
> So, do we have to hardcode it this way or let the user deal with
> some symbolic links from egfortran -> gfortran (and add a note
> for that in the readme?)?
> 
> Comments/advises/corrections are welcomed/needed as I'm new
> in the writing/updating ports world.
> 
> Thank you!

Good diff. Thank you. HDF5 updates need thorough checking. Here is my
remaining TODO list for a similiar diff:

- at least compile check on more GCC49_ARCHS (sparc64, arm64 ..)
  geo/gdal
  geo/gdal,-main
  geo/gdal,python3
  geo/gdal,python3,-main
  math/armadillo
  math/matio
  math/netcdf
  math/octave
  math/py-h5py,python
- bin/mirror_server* in PLIST?
- Fortran: maybe it is time to set MODFORTRAN_COMPILER to egfortran? Three ports
  in-tree, three differente variants: symbolic link (devel/meson), FC (openmpi),
  COMPILER_LINKS (gcc8+11)

Reply via email to