gcc, to quieten its worries about parameters that must always be
unused.

don't it on a machine that does have supplementary groups would be a
security hole.  If this breaks anything we'll fix it later.

close fd, sleep, then try again.

rsync functions.

slash.

<http://www.opensource.apple.com/bugs/X/BSD%20Kernel/2734739.html>

an error message.  (Colin Walters)

#100295.

include mtimes.

  #defiine lchown chown

could be bad on systems which have no lchown and where chown
follows symbollic links.  On such systems it might be better not to
try to chown symlinks at all.

we don't get stuck in a recursive loop trying to report a broken pipe
across that same broken pipe.  Debian bug #128632 (Colin Walters)

unexpectedly disconnects.  (Colin Walters) (Debian bug #128632)

message as an FERROR rather than an FINFO.

forgotten for the next release.

than HAVE_SOCK_SIN_LEN around use of sin_len.  Correct usage was already
in place in clientname.c.

system error message when mkstemp fails.

emit_filelist_progress() makes the native compilers on systems
including Solaris and Irix happier.

nothing has yet been malloced in flist_expand(), call malloc instead of
realloc.  Problem introduced in revision 1.106 of flist.c on January 25.

'unsigned int64'.  This should fix some compile problems on machines
where int64 is not a simple integer type, but I'm not convinced it is
the ideal fix.

Add a little doc about potential optimization of stat() calls.

local transfer as root.

user is a member.

about successful tests, you can.