New conda-forge recipe for climlab

[brian-rose]

This is a recipe for the climlab packge, which is a climate modeling toolkit written in Python with some underlying numerics in Fortran.

climlab builds against numpy using numpy.distutils to compile Fortran extension modules.

I have tested this recipe on OSX and linux and posted the resulting binaries to https://anaconda.org/brian-rose/climlab.

I would love to get climlab integrated into conda-forge going forward for easier deployment.

Thanks.

[conda-forge-linter]

Hi! This is the friendly automated conda-forge-linting service.

I wanted to let you know that I linted all conda-recipes in your PR (recipes/climlab) and found some lint.

Here's what I've got...

For recipes/climlab:

• Selectors are suggested to take a <two spaces>#<one space>[<expression>] form.

[conda-forge-linter]

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipes/climlab) and found it was in an excellent condition.

[ocefpaf]

Add another skip statement above like: # [win or not py2k] instead of pinning the python version here.

[ocefpaf]

gcc brings libgfortran so there is no need to specify it here. (It is a run time dependency only.)

[brian-rose]

Thanks @ocefpaf for these suggestions!

The build is still failing on circleci and I'm not sure why.

[ocefpaf]

Usually Linux (CircleCI) is easier than OS X 😄

Do you know how to run the docker image locally? (http://conda-forge.github.io/docs/testing.html#run-docker-tests-locally-for-feedstock) That can help you debug it.

If Linux is not mandatory for you now just add another skip condition so we can merge this and you can work on fixing it in the feedstock.

[brian-rose]

The linux build is failing with linker errors:

/staged-recipes/build_artefacts/climlab_1491269046894/_b_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_plac/bin/../lib/gcc/x86_64-unknown-linux-gnu/4.8.5/crt1.o: In function `_start':
(.text+0x20): undefined reference to `main'
collect2: error: ld returned 1 exit status

I am struggling with this. Anyone have suggestions? I encounter the same type of error when running conda-build with this recipe on my local linux system.

On the other hand, the package will build without complaint by manually running python setup.py install using the default ifort compiler on my linux system.

[ocefpaf]

On the other hand, the package will build without complaint by manually running python setup.py install using the default ifort compiler on my linux system.

I don't have time to look at it right now but you probably need to either adapt the Fortran code or some compiler flag that will make it work with gfortran on Linux.

For now I recommend you to skip the build on Linux and open an issue in the feedstock.

[brian-rose]

Thanks @ocefpaf. I am skipping linux build for now.

[ocefpaf]

Thanks @ocefpaf. I am skipping linux build for now.

When you create the issue in the feedstock ping me and I will take a look. Maybe it is something simple that we are missing now.

[brian-rose]

Build is now working on OSX and linux!

I mimicked the scipy recipe as closely as possible. The trick for building on linux seems to be the inclusion of yum_requirements.txt

[brian-rose]

Anything else needed before this cam be merged?

[brian-rose]

I should mention that building climlab on Windows is definitely on my wish-list (this is educational software designed for wide deployment on students' laptops).

But from what I can gather, building Fortran extensions for Windows is not yet possible on conda-forge, and probably won't be until there's a working conda-forge recipe for numpy on Windows, discussed here: conda-forge/numpy-feedstock#46

[brian-rose]

ping @ocefpaf