Add a fortran compiler as a dependency

[kmod]

The build script ends up pip installing a number of packages
such as scipy. Typically these packages will be installed via
pre-compiled wheels, but if a wheel is not available they
will be compiled from source. This process is opaque to conda
so it doesn't help with installing the dependencies, so we have
to list them explicitly. In this case, building scipy requires
a fortran compiler.

This is particularly important for building for Pyston because
there are currently no pre-compiled wheels available in pypi.

[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 (recipe) and found it was in an excellent condition.

I do have some suggestions for making it better though...

For recipe:

• Old-style Python selectors (py27, py34, py35, py36) are deprecated. Instead, consider using the int py. For example: # [py>=36]. See lines [19]

[jakirkham]

@conda-forge-admin, please re-render

(needed for the compiler addition)

[isuruf]

The build script ends up pip installing a number of packages
such as scipy. Typically these packages will be installed via
pre-compiled wheels, but if a wheel is not available they
will be compiled from source. This process is opaque to conda
so it doesn't help with installing the dependencies, so we have
to list them explicitly. In this case, building scipy requires
a fortran compiler.

This is particularly important for building for Pyston because
there are currently no pre-compiled wheels available in pypi.

This makes no sense to me. scipy is pre-compiled as a conda
package and there's no need to include a Fortran dependency.

[leofang]

This is bizarre. Looks like @kmod was right that a ton of wheels were installed from PyPI:
https://dev.azure.com/conda-forge/feedstock-builds/_build/results?buildId=449808&view=logs&j=d0d954b5-f111-5dc4-4d76-03b6c9d0cf7e&t=841356e0-85bb-57d8-dbbc-852e683d1642&l=230
I think scikit-build may be incorrectly configured? As @isuruf mentioned this should not happen. All of the deps pulled by scikit-build are already on conda-forge.

[leofang]

maybe the correct fix is to add scikit-build to the host section...

[benfred]

Yeah - I also see a ton of wheels being pulled from pypi and that shouldn't be required. Like @leofang pointed out, I might have configured this wrong.

I'm wondering if I need to update this line here:

implicit-feedstock/recipe/meta.yaml

Line 42 in 92182a0

script: python -m pip install . --no-deps -vv

This is using pip to build, and I'm wondering if I should update to not use pip here and maybe change to something like

script: python setup.py install

and then make sure that the build dependencies like scikit-build/cmake etc are all listed in the conda host requirements.

Though, I'm really not a conda expert and would appreciate suggestions from everyone here

[jakirkham]

Ben, if you have some time next week, maybe we can take a look at this together 🙂

[isuruf]

@conda-forge-admin, rerender

[benfred]

Ben, if you have some time next week, maybe we can take a look at this together 🙂

That would be great! I also have some questions about the CUDA builds that maybe you could point me in the right direction on -

[isuruf]

This is probably conda/conda-build#3993