when compiling an extension module, the source file
location
is not mirrored into the object file location, i.e.
subdirectories
are ignored (foo/bar/baz.c is compiled to baz.o, not
foo/bar/baz.o)

Logged In: YES 
user_id=3066

Why is this a problem?  Are you running into name clashes?

Logged In: YES 
user_id=764

precisely.

Logged In: YES 
user_id=11375

Hm, this is supposed to work, judging by the docstring of 
Compiler.compile().  Maybe some  subclass of Compiler gets it wrong; what platform are you on?

Logged In: YES 
user_id=764

I experienced it with linux / gcc

Logged In: YES 
user_id=11375

Hm.  Compiling my cryptography modules, it mirrors the location for the final source file (Crypto/Cipher/DES.so), but not for the intermediate 
.o file, which gets written to  build/temp.linux-i686-2.3/DES.o.  Is that 
the problem you're seeing?

(In any case, the temp. directory should use full paths, so this is a bug.)

Logged In: YES 
user_id=764

yes, precisely !

Logged In: YES 
user_id=11375

OK, then try the attached patch.  Does it fix the problem?

Logged In: YES 
user_id=764

I 'applied' the patch to my python2.2 installation, i.e. I
replaced the (single)
call to 'object_filenames' by your version. That works.

Thanks !

Logged In: YES 
user_id=11375

Great!  Checked in as rev 1.52 of ccompiler.py, and I'll
backport it to 
the 2.2 maintenance branch.

Logged In: YES 
user_id=6656

This broke out-of-tree builds of Python.

Any quick ideas on what to do?

Logged In: YES 
user_id=11375

Not sure.  I'm trying to replicate it, but current CVS out-of-tree builds 
are failing much earlier for me, when it tries to link Parser/pgen.

Logged In: YES 
user_id=11375

One "make distclean" later, out-of-the-tree builds work fine for me.  How do they fail?

Logged In: YES 
user_id=6656

OK, they don't actually break, but .o files end up in 
  src/Modules/ 
rather than 
  src/build-temp/build/temp.foo/
so multiple oot builds don't so what you'd expect.

This came up on python-dev in the thread
"extension module .o files wind up in the wrong place"
starting here:
http://mail.python.org/pipermail/python-dev/2002-December/030644.html

Logged In: YES 
user_id=11375

Now I get it. 

In an in-tree build, the source path is something relative, such as 
../Modules/foo.c, but in an out-of-tree build it's absolute.  This is due to a line in setup.py, but I can't remember why this is done.

One fix that's an egregious hack is to modify CCompiler.object_filenames: if strip_dir is false and
the path is absolute, strip off the path to the source directory; see the attached patch.  The patch also modifies setup.py to leave the path names alone.

I'll keep thinking about this.

Logged In: YES 
user_id=11375

A much simpler change is to modify the original patch that caused the problem, and strip the directory prefixes when python_build is true, trusting that there won't be conflicting filenames in the Python distribution.

Logged In: YES 
user_id=6656

That would do, yes.

Logged In: YES 
user_id=11375

Checked in.