Re: [Numpy-discussion] Daily numpy wheel builds - prefer to per-commit builds?

2017-02-28 Thread Olivier Grisel 
+1 as well.

-- 
Olivier
___
NumPy-Discussion mailing list
NumPy-Discussion@scipy.org
https://mail.scipy.org/mailman/listinfo/numpy-discussion

Re: [Numpy-discussion] Daily numpy wheel builds - prefer to per-commit builds?

2017-02-28 Thread Ralf Gommers 
On Tue, Feb 28, 2017 at 9:43 PM, Matthew Brett 
wrote:

> Hi,
>
> I've been working to get daily travis-ci cron-job manylinux builds
> working for numpy and scipy wheels.   They are now working OK:
>
> https://travis-ci.org/MacPython/numpy-wheels
> https://travis-ci.org/MacPython/scipy-wheels
> https://7933911d6844c6c53a7d-47bd50c35cd79bd838daf386af554a
> 83.ssl.cf2.rackcdn.com
>
> These are daily builds of the respective numpy and scipy master branches.
>
> Numpy already has a system, kindly worked up by Olivier Grisel, which
> uploads wheel builds from each travis-ci run. travis-ci uploads these
> wheels for every commit, at the
> "travis-dev-wheels" container on Rackspace, visible at
> https://f66d8a5767b134cb96d3-4ffdece11fd3f72855e4665bc61c74
> 45.ssl.cf2.rackcdn.com
> . These builds are specific to the Linux they were built on - in this
> case Ubuntu 12.04.
>
> Some projects use these per-commit builds for testing compatibility
> with development Numpy code.
>
> Now I'm wondering what the relationship should be between the current
> every-commit builds and the new wheel builds.
>
> I think that we should prefer the new wheel builds and deprecate the
> previous per-commit builds, because:
>
> * the new manylinux wheels are self-contained, and so can be installed
> with pip without extra lines of `apt` installs;
> * the manylinux builds work on any travis container, not just the current
> 12.04 container;
> * manylinux builds should be faster, as they are linked against OpenBLAS;
> * manylinux wheels are closer to the wheels we distribute for
> releases, and therefore more useful for testing against.
>
> What do y'all think?
>

Uploading your daily Linux and OS X builds and just turning off uploads of
the per-commit builds sounds like an improvement to me.

Ralf
___
NumPy-Discussion mailing list
NumPy-Discussion@scipy.org
https://mail.scipy.org/mailman/listinfo/numpy-discussion

[Numpy-discussion] Daily numpy wheel builds - prefer to per-commit builds?

2017-02-28 Thread Matthew Brett 
Hi,

I've been working to get daily travis-ci cron-job manylinux builds
working for numpy and scipy wheels.   They are now working OK:

https://travis-ci.org/MacPython/numpy-wheels
https://travis-ci.org/MacPython/scipy-wheels
https://7933911d6844c6c53a7d-47bd50c35cd79bd838daf386af554a83.ssl.cf2.rackcdn.com

These are daily builds of the respective numpy and scipy master branches.

Numpy already has a system, kindly worked up by Olivier Grisel, which
uploads wheel builds from each travis-ci run. travis-ci uploads these
wheels for every commit, at the
"travis-dev-wheels" container on Rackspace, visible at
https://f66d8a5767b134cb96d3-4ffdece11fd3f72855e4665bc61c7445.ssl.cf2.rackcdn.com
. These builds are specific to the Linux they were built on - in this
case Ubuntu 12.04.

Some projects use these per-commit builds for testing compatibility
with development Numpy code.

Now I'm wondering what the relationship should be between the current
every-commit builds and the new wheel builds.

I think that we should prefer the new wheel builds and deprecate the
previous per-commit builds, because:

* the new manylinux wheels are self-contained, and so can be installed
with pip without extra lines of `apt` installs;
* the manylinux builds work on any travis container, not just the current
12.04 container;
* manylinux builds should be faster, as they are linked against OpenBLAS;
* manylinux wheels are closer to the wheels we distribute for
releases, and therefore more useful for testing against.

What do y'all think?

Cheers,

Matthew
___
NumPy-Discussion mailing list
NumPy-Discussion@scipy.org
https://mail.scipy.org/mailman/listinfo/numpy-discussion