seemingly random mac scipy failures

[gidden]

E.g., one run on a PR produced: http://submit-1.batlab.org/nmi/results/details?runID=235435

while another produced: http://submit-1.batlab.org/nmi/results/details?runID=235585

[gidden]

note this is why pyne/#372 is failing, @pshriwise and @elliottbiondo

[crbates]

@gidden I think it's correlated with which machine the full build used to do the fast build was done on. I've looked at it and the full build passed last night on exec-19 but then fast builds running on exec-19 have failed but the fast builds from yesterday on exec-18 all succeeded.

[scopatz]

Oh... that would be awful.

[scopatz]

Not surprising - but awful.

[gidden]

Sigh... terrible. There are only two mac boxes on batlab at the moment...
we should be able to do the naive solution and have fast builds for both
boxes. Any idea what the sophisticated solution would take, @crbates? I
assume some (possibly overly complicated) sedding?

On Sat, Mar 15, 2014 at 12:48 PM, Anthony Scopatz
[email protected]:

Not surprising - but awful.

Reply to this email directly or view it on GitHubhttps://github.com//issues/7#issuecomment-37732643
.

Matthew Gidden
Ph.D. Candidate, Nuclear Engineering
The University of Wisconsin -- Madison
Ph. 225.892.3192

[scopatz]

We should confirm that there are only two mac boxes before going further.

[gidden]

e.g., http://submit-1.batlab.org/quality-control/2014-03-15.html (exec-18
and exec-19).

On Sat, Mar 15, 2014 at 1:49 PM, Anthony Scopatz
[email protected]:

We should confirm that there are only two mac boxes before going further.

Reply to this email directly or view it on GitHubhttps://github.com//issues/7#issuecomment-37734402
.

Matthew Gidden
Ph.D. Candidate, Nuclear Engineering
The University of Wisconsin -- Madison
Ph. 225.892.3192

[scopatz]

Nice find. Maybe we should switch to MacOSX7? Those are green.

[crbates]

Well one option would be to use conda to install everything except for pyne, moab, and pytaps. If scipy is really the only troublemaker this could solve the problem. It would also be way faster for the full build.

[scopatz]

I am +1 on this suggestion. We'll want to do that anyway eventually. No reason to not do it in stages with the stable conda packages.