Gnu cp behaves the same way.

when --relative-paths is used.  The code was accidentally skipped starting
in CVS version 1.32 of receiver.c.  Noticed by Craig Barratt.

the Makefile fix and I like the old EXE syntax better.

AI_PASSIVE.

by system headers.  This fixes compile errors on Irix 6.5.

    rsync: stack overflow in function match_address
on openbsd.  Patch from Brian Poole <raj@cerias.purdue.edu>.

multiply defined).

possibly be affecting the IRIX build (but probably not).

error in the Makefile.

names if there were more than 2 identical names in a row.

I've had to go to the man page many times for that information and I
finally got sick of it.

compiler warnings when calling it with a const char *.

the tests that tested it.  As far as I can tell, it was always broken
since the day I put it in years ago.  In my investigation into this I
was unable to figure out what in the world I was thinking back then,
to introduce a global variable with the wrong information in it rather
than using a parameter that was already available in readlink_stat
function.  That still bothers me a bit but I decided to stop worrying
about it.

2.5.6pre2cvs.

changed the test to the common form of prefixing an x to both sides of
an equivalence because it didn't even notice a problem on the cray when
it was instead using test -n.
Also noticed an error in the test for ANSI c, doing a "$xac..." = xno"
when it should have been "x$ac..." = xno, so I fixed that too.

available, as apparently is the case on Cray UNICOS.  The AC_FUNC_ALLOCA
autoconf documentation talks about having an included alloca.c and
periodically calling alloca(0) to garbage collect when C_ALLOCA is defined,
but I don't know where to get the code or if there's anybody that cares
enough about the UNICOS port for it to be worth the trouble.

struct in_addr, which means passing &sin.sin_addr instead of
&sin.sin_addr.s_addr.  Also changed the AF_INET6 version to pass
&sin6.sin6_addr.  Hopefully this will fix UNICOS and not break
anyone else.