choderalab/assaytools

Travis build failing - libgfortran related?

Closed this issue · 5 comments

sonyahanson commented

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 commented

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 commented

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 commented

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

bas-rustenburg commented

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 commented

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

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.