Update Mac build process. Fixes #751 #1322

dmcdougall · 2012-10-01T23:15:07Z

Previously, make.osx installed necessary dependencies. Recent versions
of OS X already include all those dependencies.

The setupext.py basedir list has been updated to include the default
directory to which macports installs libraries.

Previously, make.osx installed necessary dependencies. Recent versions of OS X already include all those dependencies. The setupext.py basedir list has been updated to include the default directory to which macports installs libraries.

dmcdougall · 2012-10-02T09:39:34Z

Regarding the discussion in #751, I should perhaps add some kind of automation regarding setting the PKG_CONFIG_PATH environment variable.

If the platform has pkg-config then add python's .pc files to the end of the pkg-config search path. If there was a problem setting it or pkg-config doesn't exist then do nothing. It appears that the environment variable change only exists within the build process.

dmcdougall · 2012-10-02T10:59:42Z

Ok, I now have setup.py automatically setting the PKG_CONFIG_PATH variable. This works on my Mac, but it would be really helpful if some of the linux people could try this out and let me know if they run into any problems. Specifically, I don't think that changes to PKG_CONFIG_PATH should persist once the build process has completed. Though a change to PKG_CONFIG_PATH does not persist on my machine, I would like to know if this is not the case on linux machines. To check, do the following:

$ echo $PKG_CONFIG_PATH # look at the output of this
/some/libs:/may/get/listed:/here
$ python setup.py build
...
$ echo $PKG_CONFIG_PATH # should be the same
/some/libs:/may/get/listed:/here

pelson · 2012-10-02T14:49:27Z

setupext.py

+    if pkgconfig_path is None:
+        return
+
+    pkgconfig_path += '/pkgconfig'


os.path.join is the preferred way to do this, but then, I'm not sure that a Windows build can ever get to here...

On Tue, Oct 2, 2012 at 10:49 AM, Phil Elson notifications@github.comwrote:

In setupext.py:

@@ -258,6 +258,20 @@ def get_win32_compiler():
else:
std_libs = ['stdc++', 'm']

+def set_pkgconfig_path():

pkgconfig_path = sysconfig.get_config_var('LIBDIR')

if pkgconfig_path is None:

return

pkgconfig_path += '/pkgconfig'

os.path.join is the preferred way to do this, but then, I'm not sure that
a Windows build can ever get to here...

os.path.join() is also good in case the first string is empty

os.path.join('', 'foobar')
'foobar'

Cheers @pelson and @WeatherGod. I was clearly in noob-mode when I wrote that. Thanks for the feedback. Did you both try out a build with this new change? If so, how did it go?

dmcdougall · 2012-10-03T10:38:16Z

As far as I can tell, this and the related issue (#751) are the only 1.2.x milestones left before the rc3 cut on Monday (according to the mpl calendar). This is a gentle reminder to others that this should at least be tried before the weekend so we can iron out any problems with this approach, should they exist.

I realise we originally discussed this should have been done during rc1 ready for rc2, but it slipped my mind. Apologies.

pelson · 2012-10-03T11:54:21Z

setupext.py

@@ -267,6 +281,10 @@ def has_pkgconfig():
        #print 'environ',  os.environ['PKG_CONFIG_PATH']


We may decide that it will aid debugging if this is uncommented....

pelson · 2012-10-03T11:57:51Z

Looks sensible to me, and works just fine on RHEL6 (every platform except "win32" will be affected by this change).
I would like to get as many +1's from others as possible before this goes in if we want this to be in the next RC (minimum of @efiring @WeatherGod and @mdboom ).

+1

pelson · 2012-10-03T11:59:41Z

There is some associated documentation which we have discussed removing which relates to this. In particular, the https://github.com/dmcdougall/matplotlib/blob/master/README.osx file needs to be updated, removed and/or renamed.

pelson · 2012-10-03T12:10:54Z

There is some thought needed about the whole test directory (i.e. do we still need/want it), in particular regarding this PR is the _buildbot_mac_sage.sh file which uses the make.osx makefile.

FYI from the source root:

$> grep -ir "make.osx" *
MANIFEST.in:include Makefile  make.osx MANIFEST.in MANIFEST
README.osx:The recommended and supported way to build is to use the make.osx file
README.osx:  make -f make.osx PREFIX=/Users/jdhunter/dev PYVERSION=2.6 \
test/_buildbot_mac_sage.sh:make -f make.osx mpl_install

mdboom · 2012-10-03T14:28:05Z

+1.

@pelson: I'm not sure if anyone is using the _buildbot_mac_sage.sh anymore. And with Travis now working (mostly) we have even less incentive to keep it working.

pelson · 2012-10-03T15:17:48Z

I'm not sure if anyone is using the _buildbot_mac_sage.sh anymore.

Any thoughts on the whole test directory? It seems like pretty obsolete stuff (I have never needed to look in there before).

dmcdougall · 2012-10-04T11:51:26Z

@pelson I'll look at the test directory today.

dmcdougall · 2012-10-04T14:51:43Z

Looks like the test stuff is for an old buildbot infrastructure. Some of the scripts required python 2.6. I deleted them because I think it's safe. The tests still, unsurprisingly, passed.

There was a TODO file in test/ and some of the ideas were good, I didn't want to just delete it, so I moved it to source root. We could look into both TODO and TODO_TESTS and figure out what needs to be done. A lot of them have been marked as completed. If so, we could look at removing these in future.

Let me know if I horribly broke something.

dmcdougall · 2012-10-07T10:18:52Z

Just as a reminder, a decision should be made on this today.

mdboom · 2012-10-08T12:58:41Z

As a non-OSX user, I'll stay out of the make.osx question and leave that to others.

As for the buildbot stuff, I think that's fine. We haven't been on buildbot for a while, and Travis seems to be the way forward. We'll always have the buildbot configuration in the repo if we decide to go back to it.

pelson · 2012-10-08T13:12:52Z

+1.

Update Mac build process. Fixes #751

Update Mac build process. Fixes matplotlib#751

98e7a95

Previously, make.osx installed necessary dependencies. Recent versions of OS X already include all those dependencies. The setupext.py basedir list has been updated to include the default directory to which macports installs libraries.

pelson reviewed Oct 2, 2012
View reviewed changes

Make path handling use os.path.join

944a930

pelson reviewed Oct 3, 2012
View reviewed changes

This was referenced Oct 3, 2012

make.osx SDK location needs updating #1041

Closed

make.osx with Xcode 4.3 #825

Closed

dmcdougall added 2 commits October 4, 2012 15:45

Remove unused buildbot files. We use Travis now

98a47ef

Moved test TODO and remove test dir

8662c32

efiring added a commit that referenced this pull request Oct 8, 2012

Merge pull request #1322 from dmcdougall/mac_build

58d8310

Update Mac build process. Fixes #751

efiring merged commit 58d8310 into matplotlib:v1.2.x Oct 8, 2012

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Update Mac build process. Fixes #751 #1322

Update Mac build process. Fixes #751 #1322

dmcdougall commented Oct 1, 2012

dmcdougall commented Oct 2, 2012

dmcdougall commented Oct 2, 2012

pelson Oct 2, 2012

WeatherGod Oct 2, 2012

dmcdougall Oct 2, 2012

dmcdougall commented Oct 3, 2012

pelson Oct 3, 2012

pelson commented Oct 3, 2012

pelson commented Oct 3, 2012

pelson commented Oct 3, 2012

mdboom commented Oct 3, 2012

pelson commented Oct 3, 2012

dmcdougall commented Oct 4, 2012

dmcdougall commented Oct 4, 2012

dmcdougall commented Oct 7, 2012

mdboom commented Oct 8, 2012

pelson commented Oct 8, 2012

		@@ -267,6 +281,10 @@ def has_pkgconfig():
		#print 'environ', os.environ['PKG_CONFIG_PATH']

Update Mac build process. Fixes #751 #1322

Update Mac build process. Fixes #751 #1322

Conversation

dmcdougall commented Oct 1, 2012

dmcdougall commented Oct 2, 2012

dmcdougall commented Oct 2, 2012

pelson Oct 2, 2012

Choose a reason for hiding this comment

WeatherGod Oct 2, 2012

Choose a reason for hiding this comment

dmcdougall Oct 2, 2012

Choose a reason for hiding this comment

dmcdougall commented Oct 3, 2012

pelson Oct 3, 2012

Choose a reason for hiding this comment

pelson commented Oct 3, 2012

pelson commented Oct 3, 2012

pelson commented Oct 3, 2012

mdboom commented Oct 3, 2012

pelson commented Oct 3, 2012

dmcdougall commented Oct 4, 2012

dmcdougall commented Oct 4, 2012

dmcdougall commented Oct 7, 2012

mdboom commented Oct 8, 2012

pelson commented Oct 8, 2012