When I compile on my OS X 10.3.4 with Fink's readline 4.3-25 I get 
the following two warnings:

/Users/drifty/Code/CVS/python/dist/src/Modules/readline.c: In 
function `flex_complete':
/Users/drifty/Code/CVS/python/dist/src/Modules/readline.c:604: 
warning: implicit declaration of function `completion_matches'
/Users/drifty/Code/CVS/python/dist/src/Modules/readline.c:604: 
warning: return makes pointer from integer without a cast

Now, although I have a version greater than 4.2, this only makes 
sense if HAVE_RL_COMPLETION_MATCHES is undefined and thus 
the completion_matches() macro is not being defined (although I 
checked and the headers have a prototype for 
rl_completion_matches() and my readline seems to compile and 
work well enough).

Would a #else statement that defines completion_matches() fit in 
there?  Problem would be the return type for flex_complete() 
might need to be covered as well unless a clever macro can be 
created that returns a char **.

Initially assigned to Michael since he seems to always deal with the 
readline stuff.

Logged In: YES 
user_id=6656

> Initially assigned to Michael since he seems to always 
> deal with the readline stuff.

Oh, er, "good".

I'm a little confused.  Is this a new thing?  As in, introduced by my 
changes just before 2.4a1?  *Why* is 
HAVE_RL_COMPLETION_MATCHES undefined?  What do config.log 
and pyconfig.h have to say on the matter?

> Would a #else statement that defines completion_matches()
> fit in there?

Not unless we can work out why it's needed.

Logged In: YES 
user_id=357491

> Oh, er, "good".

If I was wrong to assign it to you feel free to unassign it.

I don't know when it was introduced; it might be old.  I just decided to go 
through my make output and look for any warnings before a1 when out 
the door.  I tend to compile with ``--without-toolbox-glue`` and that 
leads to a ton of output from modules wanting Carbon stuff and throwing 
a fit that they can't link so it just didn't catch my eye.

As for RL_COMPLETION_MATCHES, pyconfig.h has it commented out and 
config.log has test code fail at compilation because it can't find code for 
-lreadline::

configure:17625: checking for rl_completion_matches in -lreadline
configure:17656: gcc -o conftest -g -O2   conftest.c -lreadline -ltermcap 
-ldl  >        &5
ld: can't locate file for: -lreadline

so I end up with::

``ac_cv_lib_readline_rl_completion_matches=no``.

in config.log .

Logged In: YES 
user_id=6656

So I'm guessing that the answer to this is to teach the
configure test to look in /sw/lib, then.  Not sure how to do
that...

Logged In: YES 
user_id=357491

Sounds like the right solution.  But unfortunately I don't know how to 
implement the fix either.  Ask on python-dev?

Logged In: YES 
user_id=6656

Hmm.  Does running 

LDFLAGS=-L/sw/lib ./configure

acheive anything interesting?

Speaking for myself, I'd prefer it if python's build process ignored 
/sw/ entirely by default...

Logged In: YES 
user_id=357491

OK, that seemed to solve it.  I will fire off a quick email to python-dev to 
see if anyone feels we should try to deal with this or add a mention in 
the README.

Reassigning to myself.  Thanks, Michael, for the help.

Logged In: YES 
user_id=357491

Closing as won't fix since turned out mostly to be a setup issue on my 
box of not having Configure know where the header file was.