On Fri, 2004-05-07 at 00:23, W. Borgert wrote:
Quoting Roland Stigge <stigge antcom de>:checking local Python configuration... checking for script directory... NONE/lib/python2.3/site-packages checking for extension module directory... NONE/lib/python2.3/site-packages looks goodThe script says "looks good", but I disagree. Maybe some autoconf stuff replaced '/usr' with 'NONE'?
I think the problem is in acinclude.m4 - I've attached the unified diff
of 0.93 against 0.92.2
Note the following extract:
+ if test -e "${py_config_dir}/libpython${PYTHON_VERSION}.a"; then
+ PYTHON_LIBS="-L${py_config_dir} -lpython${PYTHON_VERSION}
$py_localmodlibs $py_basemodlibs $py_other_libs"
Now py_config_dir is set here:
-py_exec_prefix=`$PYTHON -c "import sys; print sys.exec_prefix"`
+py_config_dir=`$PYTHON -c "import os; print
os.path.abspath('$pyexecdir/../config')"`
So it relies on the value of pyexecdir (note that 9.2.2 does not).
My autoconf and m4 skills are pretty much non-existent, but the
following appears to set pyexecdir:
+ AC_CACHE_CHECK([for $am_display_PYTHON extension module directory],
+ [am_cv_python_pyexecdir],
+ [am_cv_python_pyexecdir=`$PYTHON -c "from distutils import
sysconfig; print
sysconfig.get_python_lib(1,0,prefix='$PYTHON_EXEC_PREFIX')" 2>/dev/null
||
+ echo
"${PYTHON_EXEC_PREFIX}/lib/python${PYTHON_VERSION}/site-packages"`])
+ AC_SUBST([pyexecdir], [$am_cv_python_pyexecdir])
Both OR clauses make use of PYTHON_EXEC_PREFIX, which is set here:
+ AC_SUBST([PYTHON_EXEC_PREFIX], [${exec_prefix}])
Of course, exec_prefix defaults to NONE, but can be passed as an option
to ./configure (--exec-prefix).
I have confirmed that passing --exec-prefix=/usr to ./configure results
in successful detection of python libraries (when --with-python is used
as well, of course).
However, ./configure --help indicates that exec-prefix is used to
specificy the location to "install architecture-dependent files". I
think that using exec-prefix to determine the location of python
libraries is therefore incorrect.
This is probably the point where I'd attach a patch to solve the
problem, but as mentioned previously I've never used autoconf or m4
before - I don't even know if my (possible) explanation is correct.
Hopefully someone with more experience can confirm/deny the contents of
this email :-)
Cheers
Andrew
Attachment:
acinclude.m4.diff
Description: Text Data