Building the modules fails if --enable-shared is used. The linker picks
up an existing library from /usr/local/lib, which has not been compiled
with -fPIC:


gcc -pthread -shared build/temp.linux-x86_64-3.2/home/stefan/svn/py3k/Modules/_struct.o -L/usr/local/lib -L. -lpython3.2m -o build/lib.linux-x86_64-3.2/_struct.cpython-32m.so
/usr/bin/ld: /usr/local/lib/libpython3.2m.a(abstract.o): relocation R_X86_64_32S against `_PyObject_NextNotImplemented' can not be used when making a shared object; recompile with -fPIC
/usr/local/lib/libpython3.2m.a: could not read symbols: Bad value
collect2: ld returned 1 exit status


Linking works if the order of the flags is changed to "-L. L/usr/local/lib".

So, you must have done a 'make install' or 'make altinstall' to get a build into /usr/local, right?  Without that of course it works fine.

You're probably still right about changing the order so it picks up the local copy first.

I'm bumping the status down since this likely won't affect the average user, only developers who do their own source installs.

-L. should appear before -L/usr/local/lib.

Since I'm on Linux, I did a prior install into /usr/local. But I'm pretty
sure that BSD ports (which you might view as the system install) also use
/usr/local.

r86734

BTW, I am not sure it's worth backporting this to 3.1 and 2.7.  It seems like a corner case that will not affect most users.

It is wort to fix regression if all directories are absolute. 

diff --git a/setup.py b/setup.py
--- a/setup.py
+++ b/setup.py
@@ -54,7 +54,8 @@
     for i, path in enumerate(dirlist):
         if not os.path.isabs(path):
             dirlist.insert(i + 1, dir)
-            break
+            return
+    dirlist.insert(0, dir)

Roumen's patch fixes a regression where readline extension would fail to build on Mac OS X 10.6.5.

r86837