test_distutils fails when you're not building in the root of the source
tree.

That is:

mkdir build
cd build
../configure
make
make check

This fails like this:

test test_distutils failed -- Traceback (most recent call last):
  File "/Users/ronald/Projects/python/python-
trunk/Lib/distutils/tests/test_build_ext.py", line 23, in setUp
    shutil.copy(xx_c, self.tmp_dir)
  File "/Users/ronald/Projects/python/python-trunk/Lib/shutil.py", line 
82, in copy
    copyfile(src, dst)
  File "/Users/ronald/Projects/python/python-trunk/Lib/shutil.py", line 
46, in copyfile
    fsrc = open(src, 'rb')
IOError: [Errno 2] No such file or directory: 
'/Users/ronald/Projects/python/python-trunk/build104/Modules/xxmodule.c'

This also happens in py3k.
Attached a patch that fixes this by using the dirname of
sysconfig.project_base if the file is not found, but perhaps a better
solution could be found...

Looks like this has exposed some ugly code. From setup.py:

# Figure out the location of the source code for extension modules
# (This logic is copied in distutils.test.test_sysconfig,
# so building in a separate directory does not break test_distutils.)

I believe the proper fix is to move some of the code from setup.py into
sysconfig so that it does more than just set "python_build" when running
inside a Python build directory. For example, it should also add
<srcdir>/Include to the standard list of includes that distutils uses.

The attached patch fixes test_build_ext.py to use the right source file
(perhaps it should be tested on Windows). The test still fails because
Python.h cannot be found.

I think my previous patch combined with sysconfig_builddir.patch fixes
this issue. Someone will need to test on other platforms. Note that
the messy code in setup.py and in the tests should still be cleaned up.

BTW, distutils is a den of stinking evil. ;-)

One final patch for today (get_python_inc.patch). The patch combines my
previous two patches and also cleans up some ugly code in setup.py and
test_sysconfig.py. The source of the ugliness was that get_python_inc()
did not work when running from within a build directory. The patch also
fixes the header dependency feature introduced in svn r60287 (it assumed
that builddir == srcdir and also that os.path.sep == '/').

This issue is fixed in the 2.7 and 3.2 trees.

I'm therefore closing this issue.