The Bayeslite build process is based on Travis and Conda. The (main) problem was that the Numpy dependency could not be resolved -- it didn't resolve it's own dependency on Cython. By adding the channel that is recommended on Numpy's Anaconda page used to install Numpy instead of relying to different Conda default channel, we can fix this.
Why should we do this?
... because we want to maintain our software?
How do I test this?
Push something to a branch which is branched off this branch and checkout what happens on Travis.
I finally found a fix that is not a hack.
What does this do?
The Bayeslite build process is based on Travis and Conda. The (main) problem was that the Numpy dependency could not be resolved -- it didn't resolve it's own dependency on Cython. By adding the channel that is recommended on Numpy's Anaconda page used to install Numpy instead of relying to different Conda default channel, we can fix this.
Why should we do this?
... because we want to maintain our software?
How do I test this?
Push something to a branch which is branched off this branch and checkout what happens on Travis.