Packaging NetCDF with fpm

[pgierz]

Hi,

I really like the idea of fpm, and it seems to bring Fortran much more into the modern world.

I mostly work with scientific numerical software, and a common data format there is NetCDF; so I am trying to get a fpm-friendly version of NetCDF prepared. Basically it is just writing the fpm.toml file. I've been reading through the documentation but I might need some guidance.

After a fork of the official NetCDF repo, I tried my hand at the very basics:

name = "NetCDF"
version = "4.5.4"
license = "GPL v3"                       # Probably wrong
author = "Unidata"
maintainer = "paul.gierz@awi.de"         # Only for the fpm version
copyright = "Copyright 2022, Paul Gierz" # Only for the fpm version
[build]
auto-executables = true
auto-tests = true
auto-examples = true
[install]
library = true
[library]
source-dir = "fortran"

Unfortunately, fpm build dies right away:

 Warning: Module netcdf in ./fortran/netcdf.f90 is a duplicate
<ERROR>*build_model*:Error: One or more duplicate module names found.
STOP 1

Probably I am just writing down the fpm.toml incorrectly, any help would be hugely appreciated!

All work so far can be found here: https://github.com/pgierz/netcdf-fortran

Best,
Paul

[zoziha]

Maybe you can use the nc4fortran project, you can install the netcdf link library first, and then specify the FPM_FFLAGS environment variable to specify the link library and header file location to use NetCDF in fpm.

[pgierz]

I would like to, but the project I am working on still uses vintage Fortran, and changing the NetCDF library is not an option.

[LKedward]

If netcdf and netcdf-fortran are installed to the local system, then this fpm package should provide interfaces to that: https://github.com/LKedward/netcdf-interfaces

[vmagnin]

I had the same problem with my library.
fpm is building all the Fortran files, and it can not manage two files with two modules with the same name.
In your case the files netcdf.f90 and netcdf4.f90 define the same module netcdf.

[pgierz]

So, is this a bug in fpm? Should this instead be an issue I stead of just a Q&A?

[vmagnin]

It's not a bug but a functionality! fpm is looking for all the Fortran files in the source directory and compiles all the modules.
You may need to restructure the repository as it seems that two versions are stored in the same directory: NetCDF 3 and NetCDF 4.
In the CMakeLists.txt I have found:

IF (${USE_NETCDF4})
  SET(netcdff_SOURCES ${netcdff_SOURCES}
    netcdf4.f90
    )
  # SET(netcdff_DEPENDENCIES ${netcdff_DEPENDENCIES}
  #   netcdf4_func.f90 netcdf4_externals.f90 netcdf4_visibility.f90
  #   netcdf4.f90 netcdf4_eightbyte.f90
  #   netcdf4_variables.f90 netcdf4_file.f90
  #   )
ELSE ()
  SET(netcdff_SOURCES ${netcdff_SOURCES}
    netcdf.f90 netcdf3_file.f90
    )
ENDIF(${USE_NETCDF4})

[vmagnin]

The two versions should be stored in two different directories or git branches. In my opinion that last solution is probably better for fpm. But I think @LKedward, @awvwgk and other main fpm developers will tell you more as they know fpm better than me.

[pgierz]

Two different git branches will probably not fly with the folks who actually run NetCDF, so my PR to make their library fpm friendly will almost certainly be rejected. Two different directories might be possible.

To mimic the CMake way of doing things by selectively including or excluding some sources is probably not possible in the fpm world? And probably, by design?

[LKedward]

To mimic the CMake way of doing things by selectively including or excluding some sources is probably not possible in the fpm world? And probably, by design?

It's not by design necessarily. Something similar will hopefully be possible in the future with #609