1.0.x broken on macOS

[rgommers]

See https://travis-ci.org/MacPython/scipy-wheels/builds/357678450, ends repeatably with sudo: pip: command not found. Looks like multibuild was updated for various macOS issues, so looks like the right fix is to update to latest multibuild also on the 1.0.x branch. Opening this issue to ask if anyone sees an issue with that.

[rgommers]

Okay, gave that a try in 170955c

[rgommers]

Great, those builds errored after 5 minutes and TravisCI is not showing the full build log even with "raw log". That needs reverting (tomorrow), guess we'll try and cherry-pick the relevant macOS fixes. Not sure how well that works with a git submodule - is it a good idea to just add some commits directly to a submodule??

[pv]

I think it's not possible to add commits to a submodule, it always has to refer to a commit available on upstream.

[matthew-brett]

I will try and fix this up later today.

…

On Sat, 24 Mar 2018 12:14 PM Pauli Virtanen, ***@***.***> wrote: I think it's not possible to add commits to a submodule, it always has to refer to a commit available on upstream. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#27 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAEIHN0y_cn9wFsG3EiOtw-qx4D90R9Tks5thjitgaJpZM4S5mia> .

[pv]

Note that the new travis builds were terminated with an exclamation mark, not x --- I restarted the build in case this is just intermittent failure.

[pv]

Nope, still fails. Pushed a commit into v1.0.x that makes it use the same multibuild version as the weekly master branch builds, which seemed to be in working order.

[pv]

Travis build seems to pass OK with that version of multibuild.

[rgommers]

Ah I updated to multibuild/master while multibuild/devel is the one that worked. Thanks Pauli.

@matthew-brett would be useful to fast-forward master to devel perhaps?

[matthew-brett]

Multibuild master is nearly the same as multibuild devel though - just a few tiny differences. The version of multibuild in the scipy-wheels / master branch is a fair bit behind. I'm just trying to work out where the damage comes from :

• https://github.com/matthew-brett/scipy-wheels/tree/mb-1.0.x-testing

[matthew-brett]

Now building correctly on multibuild master - so not sure what the problem was: https://travis-ci.org/matthew-brett/scipy-wheels/builds/357931159 - maybe https://www.traviscistatus.com/incidents/n0f1kmm5k4x5

[matthew-brett]

Aha - sorry - there is a bug in the suppress macro - but the libwebp and szip builds are failing - search the log for tar: Unrecognized archive format