Magnus Hagander <magnus@hagander.net> writes:
> Tom Lane wrote:
>> Surely we must *not* be providing our own definitions of these symbols
>> when using a system version of fnmatch.
> That's the define that I reversed in the second patch I sent. It's
> supposed to be ifndef.
Okay.
>> Also, judging from the comments in the autoconf manual, you'd better
>> use AC_FUNC_FNMATCH not just test whether the function exists.
> Ok, will look at switching to that.
Hmm ... actually there's still possibly an issue there: what if the
system provides a broken version of fnmatch? AC_FUNC_FNMATCH will not
set HAVE_FNMATCH, and then we might end up with #define conflicts
anyway.
Since fnmatch and the #define's are supposed to be provided by
<fnmatch.h>, I think you should probably put the substitute definitions
in a substitute fnmatch.h, not port.h, to avoid that risk.
regards, tom lane