TST: Travis 3.6 Job Failing Across Multiple PR's

[gfyoung]

https://travis-ci.org/pandas-dev/pandas/jobs/278263195
https://travis-ci.org/pandas-dev/pandas/jobs/278232682

These are a couple of examples. Seems like numpy is not compatible for some strange reason.

[jreback]

conda forge appears to have updated the numpy build ; i think other deps need updating (iow anything that depends on numpy)

you can check in there recently merged PR list to see

[gfyoung]

Hmm...numpy hasn't made a new release since July...that's pretty late to update on conda.

[jorisvandenbossche]

Isn't this is something we rather should report to conda-forge instead of that 'fix' ? (which can certainly be done temporarily)
As conda should not just downgrade numpy?

[jreback]

no its not really conda-forge's issue. its the downstream packages that should update. and we use quite a lot of these (e.g. numexpr, pytables, pyarrow, netcdf4, matplotlib, statsmodels, bottleneck). IOW anything that specifically targets the ABI (like pandas does).

Whenever a new ABI version (or version), then the downstreams need to pin. Sure you can 'report' it if you want. I am going to temporarily pin to make our builds work (and will revert at some later point).