Any progress on binary installer for OSX?

In article
<CAH6Pt5q=Z_mACAy-DgRoz38JiMo6ucLv+amxkXhvT+qoe+CEHg@...288...>,
Matthew Brett <matthew.brett@...287...>
wrote:

Hi,

> Matthew Terry, as part of his Mac testing project, has done a great deal of
> reconnaissance on this.
>
> https://github.com/matplotlib/mpl_mac_testing
>
> I know he was looking into statically linking some of the C dependencies
> (freetype, libpng etc.) as a way to make the installer more robust to
> different environments.

Thanks - that looks like a useful testing grid.

Are there any near-term plans for something like a .dmg or .mpkg or
.pkg installer?

Building a binary installer with statically linked libraries is not
terribly hard (see
<http://www.astro.washington.edu/users/rowen/BuildingMatplotlibForMac.htm

). There are two problem:

- As of 1.3.0 mpl does not include python-dateutil, pytz or six (for
good reasons) and that makes it harder to make a really usable binary
installer. This interacts with the next problem:
- For unknown reasons running the 1.3.0 installer breaks existing
installations of python-dateutil if those packages were installed using
an older mpl binary installer.

The missing packages can be added to the binary installer after it is
produced by bdist_mpkg by post-processing the mpkg. That would take care
of the second issue for most users (who would use the default
installation and get everything). I have not had time to deal with that.
Thus I never uploaded an official binary installer for 1.3.0 and stopped
providing them. Matthew Terry has taken over that task.

I did put an unofficial binary installer for 1.3.0 here:
<http://www.astro.washington.edu/users/rowen/python/>
just be prepared to reinstall python-dateutil after you use it.

-- Russell

···

On Thu, Oct 3, 2013 at 5:59 AM, Michael Droettboom > <mdroe@...86...> wrote:

Hi,

···

On Thu, Oct 3, 2013 at 1:29 PM, Russell E. Owen <rowen@...2756...> wrote:

In article
<CAH6Pt5q=Z_mACAy-DgRoz38JiMo6ucLv+amxkXhvT+qoe+CEHg@...288...>,
Matthew Brett <matthew.brett@...287...>
wrote:

Hi,

On Thu, Oct 3, 2013 at 5:59 AM, Michael Droettboom >> <mdroe@...86...> wrote:
> Matthew Terry, as part of his Mac testing project, has done a great deal of
> reconnaissance on this.
>
> https://github.com/matplotlib/mpl_mac_testing
>
> I know he was looking into statically linking some of the C dependencies
> (freetype, libpng etc.) as a way to make the installer more robust to
> different environments.

Thanks - that looks like a useful testing grid.

Are there any near-term plans for something like a .dmg or .mpkg or
.pkg installer?

Building a binary installer with statically linked libraries is not
terribly hard (see
<http://www.astro.washington.edu/users/rowen/BuildingMatplotlibForMac.htm
>). There are two problem:
- As of 1.3.0 mpl does not include python-dateutil, pytz or six (for
good reasons) and that makes it harder to make a really usable binary
installer. This interacts with the next problem:
- For unknown reasons running the 1.3.0 installer breaks existing
installations of python-dateutil if those packages were installed using
an older mpl binary installer.

The missing packages can be added to the binary installer after it is
produced by bdist_mpkg by post-processing the mpkg. That would take care
of the second issue for most users (who would use the default
installation and get everything). I have not had time to deal with that.
Thus I never uploaded an official binary installer for 1.3.0 and stopped
providing them. Matthew Terry has taken over that task.

Aha - yes - postprocessing the mpkg would be pretty easy.

So - I guess I should just build the installer myself and post it for
testing? Is that the best way forward?

Cheers,

Matthew

Hi,

Hi,

In article
<CAH6Pt5q=Z_mACAy-DgRoz38JiMo6ucLv+amxkXhvT+qoe+CEHg@...288...>,
Matthew Brett <matthew.brett@...287...>
wrote:

Hi,

> Matthew Terry, as part of his Mac testing project, has done a great deal of
> reconnaissance on this.
>
> https://github.com/matplotlib/mpl_mac_testing
>
> I know he was looking into statically linking some of the C dependencies
> (freetype, libpng etc.) as a way to make the installer more robust to
> different environments.

Thanks - that looks like a useful testing grid.

Are there any near-term plans for something like a .dmg or .mpkg or
.pkg installer?

Building a binary installer with statically linked libraries is not
terribly hard (see
<http://www.astro.washington.edu/users/rowen/BuildingMatplotlibForMac.htm
>). There are two problem:
- As of 1.3.0 mpl does not include python-dateutil, pytz or six (for
good reasons) and that makes it harder to make a really usable binary
installer. This interacts with the next problem:
- For unknown reasons running the 1.3.0 installer breaks existing
installations of python-dateutil if those packages were installed using
an older mpl binary installer.

The missing packages can be added to the binary installer after it is
produced by bdist_mpkg by post-processing the mpkg. That would take care
of the second issue for most users (who would use the default
installation and get everything). I have not had time to deal with that.
Thus I never uploaded an official binary installer for 1.3.0 and stopped
providing them. Matthew Terry has taken over that task.

Aha - yes - postprocessing the mpkg would be pretty easy.

So - I guess I should just build the installer myself and post it for
testing? Is that the best way forward?

OK - after a lot of blood, sweat and tears:

http://nipy.bic.berkeley.edu/practical_neuroimaging/matplotlib-1.3.1-py2.7-macosx10.6.mpkg.zip

- a standalone binary installer for matplotlib 1.3.1, including:

tornado
pyparsing
python-dateutil
pytz
six

Please do test.

It imports on my machines (10.6, 10.7 * 2, 10.8), I am just running the tests.

I'm building from a waf build that should be replicable:

https://github.com/matthew-brett/mpl-osx-binaries

On a 10.6 and a 10.8 machine I get a couple of test errors, log attached:

ERROR: matplotlib.tests.test_backend_pgf.test_pathclip
ERROR: matplotlib.tests.test_backend_pgf.test_mixedmode

One (clean) 10.7 passes, another 10.7 machine gives the same errors as
above plus 2 ghostscript errors.

Cheers,

Matthew

···

On Thu, Oct 3, 2013 at 1:33 PM, Matthew Brett <matthew.brett@...287...> wrote:

On Thu, Oct 3, 2013 at 1:29 PM, Russell E. Owen <rowen@...2756...> wrote:

On Thu, Oct 3, 2013 at 5:59 AM, Michael Droettboom >>> <mdroe@...86...> wrote:

Hi,

mpl.log.bz2 (3.15 KB)

···

On Fri, Oct 11, 2013 at 9:29 PM, Matthew Brett <matthew.brett@...287...> wrote:

Hi,

On Thu, Oct 3, 2013 at 1:33 PM, Matthew Brett <matthew.brett@...287...> wrote:

Hi,

On Thu, Oct 3, 2013 at 1:29 PM, Russell E. Owen <rowen@...2756...> wrote:

In article
<CAH6Pt5q=Z_mACAy-DgRoz38JiMo6ucLv+amxkXhvT+qoe+CEHg@...288...>,
Matthew Brett <matthew.brett@...287...>
wrote:

Hi,

On Thu, Oct 3, 2013 at 5:59 AM, Michael Droettboom >>>> <mdroe@...86...> wrote:
> Matthew Terry, as part of his Mac testing project, has done a great deal of
> reconnaissance on this.
>
> https://github.com/matplotlib/mpl_mac_testing
>
> I know he was looking into statically linking some of the C dependencies
> (freetype, libpng etc.) as a way to make the installer more robust to
> different environments.

Thanks - that looks like a useful testing grid.

Are there any near-term plans for something like a .dmg or .mpkg or
.pkg installer?

Building a binary installer with statically linked libraries is not
terribly hard (see
<http://www.astro.washington.edu/users/rowen/BuildingMatplotlibForMac.htm
>). There are two problem:
- As of 1.3.0 mpl does not include python-dateutil, pytz or six (for
good reasons) and that makes it harder to make a really usable binary
installer. This interacts with the next problem:
- For unknown reasons running the 1.3.0 installer breaks existing
installations of python-dateutil if those packages were installed using
an older mpl binary installer.

The missing packages can be added to the binary installer after it is
produced by bdist_mpkg by post-processing the mpkg. That would take care
of the second issue for most users (who would use the default
installation and get everything). I have not had time to deal with that.
Thus I never uploaded an official binary installer for 1.3.0 and stopped
providing them. Matthew Terry has taken over that task.

Aha - yes - postprocessing the mpkg would be pretty easy.

So - I guess I should just build the installer myself and post it for
testing? Is that the best way forward?

OK - after a lot of blood, sweat and tears:

http://nipy.bic.berkeley.edu/practical_neuroimaging/matplotlib-1.3.1-py2.7-macosx10.6.mpkg.zip

- a standalone binary installer for matplotlib 1.3.1, including:

tornado
pyparsing
python-dateutil
pytz
six

Please do test.

It imports on my machines (10.6, 10.7 * 2, 10.8), I am just running the tests.

I'm building from a waf build that should be replicable:

https://github.com/matthew-brett/mpl-osx-binaries

On a 10.6 and a 10.8 machine I get a couple of test errors, log attached:

ERROR: matplotlib.tests.test_backend_pgf.test_pathclip
ERROR: matplotlib.tests.test_backend_pgf.test_mixedmode

One (clean) 10.7 passes, another 10.7 machine gives the same errors as
above plus 2 ghostscript errors.

python3.3 installer, same errors as 2.7 for OSX 10.8:

http://nipy.bic.berkeley.edu/practical_neuroimaging/matplotlib-1.3.1-py3.3-macosx10.6.mpkg.zip

I've attached the log this time (for 2.7, it's similar for 3.3)

Best,

Matthew