Allow ffi import when package pip installed

[flacjacket]

Import the built ffi modules from cairocffi._ffi rather that ._ffi, so that if the package has been installed, this will pick up the compiled modules that are installed rather than insisting they be re-built in the source directory. This has the unfortunate side-effect of possibly hiding any problems if the cdefs are changed but the package is not re-built in the source tree.

Also, get the XCB tests running on Travis, though they will be xfailed until Travis bumps to Ubuntu 14.04.

[SimonSapin]

This doesn’t fix the issue described at #60 (comment) . The output is the same. You can try it with pip install tox and then e.g. tox -e py27. See https://github.com/SimonSapin/cairocffi/blob/master/tox.ini

[flacjacket]

Hmm, if I do that then ls .tox/py27/lib/python2.7/site-packages/cairocffi/, I see that _ffi.py and _ffi_pixbuf.py are both there, but they aren't found when tox runs py.test. If I run .tox/py27/bin/python cairocffi/ffi_build.py, re-running the tests, I can get the tests to pass. The problem is tox uses the root cairocffi directory to run the tests, not the installed cairocffi directory. If I apply this:

diff --git a/tox.ini b/tox.ini
index caced90..052754c 100644
--- a/tox.ini
+++ b/tox.ini
@@ -5,7 +5,7 @@ envlist = py26-cairo-1.8.2, py26-cairo-1.10.0, py26, py27, py32, py33, py34, pyp
 deps=
     pytest
     xcffib
-commands=py.test []
+commands=py.test [] {envsitepackagesdir}/cairocffi

 [testenv:py26-cairo-1.8.2]
 ; Building old cairo on an modern system might require setting

to explicitly run from the site-packages installation rather than the source, I get tox to pass.

[SimonSapin]

The problem is tox uses the root cairocffi directory to run the tests, not the installed cairocffi directory.

That was it, good catch. I cherry-picked one of your commits but did the rest different: e374c44

The XCB test have been unreliable for me, though. Sometimes the python process gets killed with SIGABRT.

[flacjacket]

That looks good, thanks for cleaning that up.

Were you getting errors with any version in particular? I was having some problems with 2.7 while I was trying to get tox running, but it may have been how I was getting it to run. I can try to look into it.

If there's nothing else, is there any chance of getting a point release soon? We need to do the same cffi 1.0 changeover in Qtile, and since we use cairocffi for some drawing, we need to depend on a version that also uses out-of-line cffi. Thanks.

[SimonSapin]

No, pretty much all Python versions tested were affected by this SIGABRT… sometimes.

I was kinda lazy to write the changelog, but I just uploaded 0.7 to PyPI without it.

[flacjacket]

Hmm, alright, I can play around with it and come back if I can find
anything. Thanks!

On Fri, Jun 5, 2015 at 11:20 AM, Simon Sapin notifications@github.com
wrote:

No, pretty much all Python versions tested were affected by this SIGABRT…
sometimes.

I was kinda lazy to write the changelog, but I just uploaded 0.7 to PyPI
without it.

—
Reply to this email directly or view it on GitHub
#61 (comment).