LGTM, feel free to commit if the tests pass. A backport to 1.14.x would be helpful.

Does look OK, but the tests fail...

Failed with exit code 2, misuse of shell builtins or some such, not sure why. Looked like the failure was in the bootstrap.

I tried rerunning the travis test run, still fails. I tried running the command sudo debootstrap --variant=buildd --include=fakeroot,build-essential --arch=i386 --foreign bionic /chroot on my local machine, it succeeds. Not sure what to do, maybe some disk space issue? @juliantaylor - ping?

I note that the previous successful runs continued

I: Chosen extractor for .deb packages: dpkg-deb
I: Extracting base-files...
I: Extracting base-passwd...

So the failure is immediate on attempting extraction. Maybe we should pip install pytest?

Yeah, when @mattip ran it locally then that's what it did too (print "Extracting ..
" a bunch of times). Hence the guess that it could maybe possibly be a disk space issue, since it's randomly crashing in the middle of writing out files?

Could try removing build-essential, documentation says it is only needed for building debian packages.

I reached out to travis-ci, their response is

Do you know about our Debug feature (https://docs.travis-ci.com/user/running-build-in-debug-mode/)? This will allow you to restart one of the affected jobs in debug mode, and then access to the VM running your build via SSH. In this case, I believe that it could be an interesting experiment since it would allow you to interact further with the system and make further checks on the processes that are running.

This feature is deactivated by default for open source projects, but please let me know if you'd like to have it activated for numpy/numpy and I'll be glad to do so!

Who is our "official" travis-ci POC?

odd, we should be able to replace this multiarch package installations in a regular non sudo container by now. This would simplify the setup of the test by quite a bit.
This assumes travis allows installing foreign architecture packages in their non root instances, if not a sudo vm would work too.

of course doing that would set the test back to whatever travis is actually runing which I think is ubuntu trusty. So we would lose testing newer compilers and libraries.

If we want to keep going with debootstrap (as opposed to switching to multiarch packages, or I don't know, using docker or something), then the only thing I can think of to try next is strace -f debootstrap ...

(Possibly with some hacks to avoid printing the whole log, in case that triggers Travis's limit on log output.)

might have to do with dpkg 1.17 (the one in Ubuntu 14.04) not being able to deal with packages for bionic, so I enabled apt update in that specific build

What is the procedure to backport a PR ?

Depends. In this case I would rebase the top commit onto maintenance/1.14.x and push another PR.

$ git rebase HEAD^  --onto maintenance/1.14.x

Or you could just cherry-pick the commit to a branch based on maintenance/1.14.x.

Actually, we can probably ignore this for 1.14.4, it was only needed for pytest, correct?

Although it cannot hurt, but I don't think we will do anymore 1.14.x releases.

This PR is needed only if PR #10142 is merged to update pytest to include warning filters so I hink we can ignore it for 1.14.x, although I am intrigued how to do backports

There is some discussion of backports in doc/source/dev/gitwash/development_workflow.rst, which is probably a good starting point. It may also be somewhat outdated. For other peoples commits, I first make sure I have the merge in my master branch, then, say

charris@fc [numpy.git (maintenance/1.14.x)]$ git co -b backport-11043 maintenance/1.14.x
Switched to a new branch 'backport-11043'
charris@fc [numpy.git (backport-11043)]$ git cherry-pick 2f4e5fdb3b1d4ecff2af2baae191f963a3df0470
[backport-11043 32610b9d0e] BUG: reduce using SSE only warns if inside SSE loop
 Author: mattip <matti.picus@gmail.com>
 Date: Sun May 6 00:04:16 2018 +0300
 3 files changed, 10 insertions(+), 3 deletions(-)
charris@fc [numpy.git (backport-11043)]$ python runtests.py
<snip>
charris@fc [numpy.git (backport-11043)]$ git push origin HEAD

You can get the needed hash from the PR by hitting the commits button, then the copy hash button. Backports get more complicated when there are more than one commit, but if you read up on cherry-pick you will see how to do a range of commits.