Incorrect "has_fortran" and "has-c++4" reporting

[gsjaardema]

If I run nc-config --all, I get the following (editted) output:

s1041980:build()> ../bin/nc-config --all

This netCDF 4.8.2-development has been built with the following features:

  --cc            -> /opt/local/bin/gcc
  --cflags        -> -I/Users/gdsjaar/src/seacas/include
  --libs          -> -L/Users/gdsjaar/src/seacas/lib -lnetcdf
  --static        -> -lhdf5_hl-shared -lhdf5-shared -lm -lz -lzstd -lbz2 -lsz -lcurl -lxml2

  --has-c++       -> no
  --cxx           ->

  --has-c++4      -> yes
  --cxx4          -> clang++
  --cxx4flags     -> -I/usr/local/Cellar/netcdf/4.8.1_2/include
  --cxx4libs      -> -L/usr/local/Cellar/netcdf/4.8.1_2/lib -lnetcdf-cxx4 -lnetcdf

  --has-fortran   -> yes
  --fc            -> /usr/local/bin/gfortran
  --fflags        -> -I/usr/local/Cellar/netcdf/4.8.1_2/include -I/usr/local/Cellar/netcdf/4.8.1_2/include
  --flibs         -> -L/usr/local/Cellar/netcdf/4.8.1_2/lib -lnetcdff -lnetcdf
  --has-f90       -> TRUE
  --has-f03       -> yes

Note that this indicates that has-fortran is true, but points to a fortran netcdff that is in a different location than my local install. If I look at how nc-config is determining this, it simply searches to see if there is a nf-config in my path somewhere.

I don't think that this is correct behavior since the versions of the libraries don't match and the -flibs setting would result in the application linking with a different libnetcdf (C API) then the local one that this nc-config is giving details for.

Maybe the correct behavior is to search for a nf-config that is in the same path as the nc-config? I'm not sure of the correct behavior, but this output was confusing to me since I intentionally disable fortran when I am building netcdf-c.

The C++ / C++4 reporting is also confusing since they are also picking up a ncxx-config and ncxx4-config anywhere in the path and the resulting library they find may be different versions than what I am building locally.

[edwardhartnett]

Since netcdf-fortran is installed separately, I don't know how nc-config can be expected to know what settings to use for Fortran.

Similarly, I don't see how nc-config can know anything about the netcdf-cxx library or where it might be installed.

Should these values even be in nc-config?

[gsjaardema]

Agree. I was surprised to see them in the output.

[edwardhartnett]

Is is possible that these are historical artifacts from the time when netCDF C and Fortran libraries were distributed as one package? (But even if so, netcdf-cxx was never part of it.)

Anyway, I believe these fields should be removed from nf-config. @WardF what do you think?

[DennisHeimbigner]

Historical note. The current c++ implementation was not the first. There was an older version.
So perhaps this comes from that version support.

[edwardhartnett]

If we all agree they should be removed, I will submit a PR. Just waiting to see if @WardF wants to weigh in...

[edwardhartnett]

OK, I see what is going on here. In nc-config.in we have:

if [ -f "$nfconf" ]; then
  has_fortran="yes"
  fc=`nf-config --fc`
  fflags=`nf-config --fflags`
  flibs=`nf-config --flibs`
  has_f90=`nf-config --has-f90`
  has_f03=`nf-config --has-f03`
fi

So if you have nf-config in your path, you will get the correct answers.

I think this is fine the way that it is. @gsjaardema I suggest you close this issue.

[gsjaardema]

I disagree that you will get the "correct" answers. Please read the description in the original report of how I think you get "incorrect" results.

[edwardhartnett]

OK, fair point. Perhaps removing them would be the best approach, and let the user find the nf-config that he or she wants to use...

[WardF]

I've been thinking about this, and it occurs to me that removing it is probably the most straightforward way to go.