Travis build failing - libgfortran related?

[sonyahanson]

Seems like the travis build is currently failing, potentially because of this line in our conda recipe:
- libgfortran ==1.0.0 [linux]

This is the commit that changed this: 8e6f35e

Seems like we are no longer pinning the pymc version. Do we have to keep this libgfortran version pinned for some other reason?

[bas-rustenburg]

The commit you point to pins pymc, and gfortran.

This one unpins only pymc. Has it been broken for that long,, or did something else change too?

7adbe6e#diff-23fca1736cbf667a8319e570051bbe20

[MehtapIsik]

This problem looks similar to assaytools installation problems I have been having, because of pymc dependency.
The combination of versions that worked for me in the end is:

pymc                      2.3.6 
libgfortran               3.0.0  
gcc                       4.8.5 
libgcc                    5.2.0           

[sonyahanson]

I wonder if @jchodera noticed if this was happening when he cut a release in June: #111

[bas-rustenburg]

It looks like that PR might have been merged before the travis tests had a chance to run, which would lead to the travis logs you see there.

[jchodera]

Looks like this is because libgfortran 1.0 has been removed from standard channels. Now that we're requiring conda-forge, we need to:

• Add conda-forge to the channels in .travis.yaml according to these instructions:

conda config --add channels omnia --add channels conda-forge

This should cause pymc to be picked up from conda-forge

• Remove the libgfortran explicit dependency in devtools/conda-recipe/meta.yaml, since I don't think this should be explicitly specified anymore; we should rely on the conda-forge pymc to include its own dependencies.