This patch puts the delayed import of linecache back
into warnings.py, but protects it with a check against
the import lock being held by someone else.  This
results in regaining 50% of the launch speed that we
lost moving from 2.2.2 to 2.3.

Logged In: YES 
user_id=764593

You only need to import linecache once; after that, it 
shouldn't matter whether the lock is held.

You can track whether it was loaded with either a global or a 
mutable default value, but I'm not sure how unacceptable the 
style would be.  The warningupdate files shows a 
formatwarning with the mutable default variant.

Logged In: YES 
user_id=764593

It isn't letting me a attach a file just now, so I'll paste.

# Assumes that imp is imported at the top instead of 
# linecache, as in the original patch.

def formatwarning(message, category, filename, lineno,
                  linec = []):
    """Function to format a warning the standard way."""
    s =  "%s:%s: %s: %s\n" % (filename, lineno, category.
__name__, message)
    if not linec:
        if imp.lock_held():
            # Somebody else is holding the import lock.  To avoid  
            # a deadlock we just return the string so in worst 
            # case the user can look it up themselves.  Sorry.
            return s
        else:
            import linecache
            linec.append(linecache.getline)
    line = linec[0](filename, lineno).strip()
    if line:
        s = s + "  " + line + "\n"
    return s

Logged In: YES 
user_id=430343

Yes, Jim's patch fixes the obvious problem...that's what I get for
writing code after a long day at PyCon.. :-)

Logged In: YES 
user_id=430343

I've attached a new patch which corrects the multiple import
lock checks, and uses a global binding instead of the static
mutable default arg hack.. :-)

Logged In: YES 
user_id=430343

Apparently the file didn't stick last time.  Lets try it again.

Logged In: YES 
user_id=430343

Hopefully the last patch - whitespace screwed up in the last one.

Logged In: YES 
user_id=430343

Arg - this only works in cvs.  In the installed version it doesn't make any 
difference.

Logged In: YES 
user_id=764593

Does that still make it worth applying for 2.4?
It isn't really a good 2.3 candidate anyhow, since "slow" isn't a 
bug.