[SciPy-Dev] moved 0.19.1 tag

[Matthew Brett]

On Thu, Jun 22, 2017 at 12:39 PM, Ralf Gommers <ralf.gommers at gmail.com> wrote:
>
>
> On Thu, Jun 22, 2017 at 11:36 PM, Matthew Brett <matthew.brett at gmail.com>
> wrote:
>>
>> Hi,
>>
>> On Thu, Jun 22, 2017 at 12:32 PM, Ralf Gommers <ralf.gommers at gmail.com>
>> wrote:
>> >
>> >
>> > On Thu, Jun 22, 2017 at 11:26 PM, Jordi Montes <jomsdev at gmail.com>
>> > wrote:
>> >>
>> >> Hi,
>> >>
>> >> What do you mean when you say that it is not tested in your regular
>> >> TravisCI? Maybe I could do a pull request fixing it and to skip this
>> >> kind of
>> >> problems the next time.
>> >
>> >
>> > It was a 32-bit platform specific issue that turned up at the last
>> > moment,
>> > and we don't have 32-bit builds on TravisCI. We build the wheels for
>> > release
>> > from a separate repo (https://github.com/MacPython/scipy-wheels), which
>> > does
>> > do 32-bit. And those builds were failing after I tagged the release.
>> >
>> > Copying the TravisCI config (PLAT=i686) from the scipy-wheels repo and
>> > updating one of the existing entries in the build matrix of the scipy
>> > repo
>> > to use that would be a helpful contribution.
>>
>> We do test master every day with cron jobs on the scipy-wheels repo.
>> Did those tests miss this problem?
>
>
> That doesn't test 0.19.x (and I'm not suggesting it should); I backported a
> PR from master with the issue to 0.19.x. On master it was fixed later, after
> the daily builds found the problem. Didn't see that second PR until it was
> too late.

Maybe worth adding instructions to trigger a build on the scipy-wheels
repo with the relevant commit before tagging?

Cheers,

Matthew