Installation with Python 3 and Fortran 90

[mwoussen]

The file pyratp.pyd is not working on my configuration and doesn't seem to be built while installing the package with scons.
My processor is a Intel i7-4800MQ 64 bits, with Windows 10. I installed a conda environment with Python 3.7 and the following packages :

• openalea.plantgl
• openalea.sconsx
• sphinx -> it seems like one of the packages installed with Sphinx allows a correct interaction between plantgl and sconsx, otherwise, plantgl is broken.
• openalea.deploy
• numpy, pandas, scipy, matplotlib : last version

Under this conda env, I did this modifications to adapt the code in Python 3 :

• print " -> print(
• <> -> !=
• setup.py l.12 : metadata.iteritems() -> metadata.iter()

The command python setup.py install correctly installs the pakage but without rebuilding the pyratp.pyd file, only copying it.

The existing pyratp.pyd seems to be compiled for Python 2, so I tried to build it with f2py by typing the commands described in src/f90/SConscript :

1. f2py -m pyratp --overwrite-signature -h [source_64bits]
2. f2py -c --build-dir . --compiler=mingw32 --fcompiler=gnu95 [sources_64bits]

Where [sources_64bits] are the files written in src/f90/SConscript l.18-32 . The compile command (f2py -c) ends up with an error :

pyratpmodule.c: 3335: undefined reference to '__intrinsic_setjmpex'
collect2.exe: error: ld returned 1 exit status

I also tried using a conda environnement with Python 2, but openalea.sconsx is no longer available for this version of Python which makes the python setup.py install command obsolete. Then, when I recreate the pyratp.pyf file, it calls unknown modules in the environnement so f2py -c can't build.

[mwoussen]

I managed to build the pyratp.pyd file, the procedure was :

1. f2py -m pyratp --overwrite-signature -h [source_64bits]
2. f2py -c --build-dir . --compiler=mingw32 --fcompiler=gnu95 [sources_64bits]
3. the process ends on the ld error as before
4. change #include <setjmp.h> to #include <setjmpex.h> in the src.win-amd64-3.7/pyratpmodule.c file
5. rerun f2py -c --build-dir . --compiler=mingw32 --fcompiler=gnu95 [sources_64bits], now it's building the file pyratp.src.win-amd64-3.7.pyd
6. run the last command of f2py -c ... , which is gfortran -Wall -g -shared .\Release\src.win-amd64-3.7\pyratpmodule.o .\Release\src.win-amd64-3.7\src.win-amd64-3.7\fortranobject.o .\Release\mod_[sources_64bits].o .\Release\src.win-amd64-3.7\pyratp-f2pywrappers.o -L"C:\Program Files\mingw-w64\x86_64-8.1.0-posix-seh-rt_v6-rev0\mingw64\lib\gcc\x86_64-w64-mingw32\8.1.0" -LC:\Users\mwoussen\Anaconda3\envs\ratp\libs -LC:\Users\mwoussen\Anaconda3\envs\ratp\PCBuild\amd64 -lpython37 -lgfortran -o .\pyratp.pyd, where ratp is the name of my conda env
7. Then, I copy pyratp.pyd among the Python files

[pradal]

Great!
Thanks again for the tip.
We will look at building a conda package from this work.

[pradal]

Use: https://numpy.org/devdocs/f2py/usage.html#automatic-extension-module-generation

[pradal]

In the branch meson, we will test the automatic build of PyRATP using meson with @thomasarsouze

• Use meson.build only to run f2py and build an extension module
• use automatic build in meson to run f2py before building Fortran
• use meson-python to run everything in a project.toml

[pradal]

@pellem Here is the next challenge.
You can create a build.sh to run f2py in conda using the code in the Makefile

[pradal]

@pellem

1. Read and understand meson.build

mamba install meson compilers
meson setup builddir
meson compile -C builddir
cd builddir

ipython 
import pyratp

Read https://numpy.org/doc/stable/f2py/buildtools/meson.html