Travis 27 SciPy Failures

[WillAyd]

As noted by @gfyoung in #22515 (comment) something appears to be awry with the SciPy fixtures and the Travis 2.7 env

[TomAugspurger]

Two things compared to https://travis-ci.org/pandas-dev/pandas/jobs/421543071

1. mkl and nomkl are installed are installed, which seems strange. https://travis-ci.org/pandas-dev/pandas/jobs/421777999#L1587. The numpy seems to have mkl, while the scipy one doesn't
2. libopenblas went from 0.2.20 to 0.3.2

[WillAyd]

I noticed the build for scipy is also different when compared back to the working log you linked to. Do you know of a way to install the exact build from conda locally to see if that's it? I didn't see it in a conda search scipy

[TomAugspurger]

Running `conda env create --file = ci/travis-27.yaml` should get you most of the way there, but there might still be some platform differences.

…

On Wed, Aug 29, 2018 at 9:42 PM William Ayd ***@***.***> wrote: I noticed the build for scipy is also different when compared back to the working log you linked to. Do you know of a way to install the exact build from conda locally to see if that's it? I didn't see it in a conda search scipy — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#22538 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABQHInbj_0FN8v78PLu4ZcgRp50b2gcEks5uV1D5gaJpZM4WSm4c> .

[TomAugspurger]

I'm taking a look at this now.

[TomAugspurger]

Seems like this package needs a blas metapackage constraint. The metadata is being fixed on anaconda's end.

[TomAugspurger]

This has been fixed. I restarted the build for https://travis-ci.org/pandas-dev/pandas/builds/422535976 and it's passing now.