pip or conda version - Githubissues

madebr / pyOpt

Fork of http://www.pyopt.org/ (python3 compatible)

Other

84 stars 35 forks source link

pip or conda version #13

Open jonas-eschle opened 3 years ago

jonas-eschle commented 3 years ago

Hi all, are there any plans for a release on pip or conda-forge?

UPDATE: keeping track of what we deactivated

having problems with gcc10 as it won't compile all fortran code: https://github.com/mayou36/pyOpt/runs/2281419631?check_suite_focus=true#step:4:5272
NSAG2 in unittests: while it compiles, it can't be imported

madebr commented 3 years ago

Not currently, but it would be nice to have a github workflow to create a new pypy release when creating a git tag.

jonas-eschle commented 3 years ago

Okay, I think the library has a lot of potential. Since it is your fork, what are your plans and willingness to maybe change and improve things? I am glad to contribute something, but it requires of course also readiness to accept PRs etc.

madebr commented 3 years ago

I'm willing to accept anything as long as it's within scope. Having a rudimentary CI should prove that the code is working.

I think the

Use https://github.com/actions/setup-python to build PyOpt for all relevant python versions
On Ubuntu, use gfortran from the package manager
On Windows, use fortran from msys2 (https://github.com/msys2/setup-msys2) Windows is a stretch goal, imho.
Some care should be taken to ensure PyOpt runs without any fortran runtime libraries.
Use https://github.com/pypa/gh-action-pypi-publish to publish the package.

I think this will make it possible to easily test + publish PyOpt.

jonas-eschle commented 3 years ago

I agree to the CI and github actions, I would however opt for the conda compilers for the testing as this are available everywhere and may allows for a deployment with conda-forge at some point

Some care should be taken to ensure PyOpt runs without any fortran runtime libraries.

Why this point? Like not to force the usage of fortran?

madebr commented 3 years ago

I agree to the CI and github actions, I would however opt for the conda compilers for the testing as this are available everywhere and may allows for a deployment with conda-forge at some point

I'm fine with whatever a pr author comes up with. As long as it's robust and works with minimum differences across platforms.

Some care should be taken to ensure PyOpt runs without any fortran runtime libraries.

Why this point? Like not to force the usage of fortran?

I meant to allow usage of pyopt without having to install a distribution specific fortran library. But I don't think a fortran standard library exists, so this point can be ignored.

I think I was confused with the library component of f2c