I think open-pipe leaves an extra copy of its input or output fd in
the forked child.  For instance, using showfds.c below and

	(use-modules (ice-9 popen))

	(let ((p (open-input-pipe "./showfds")))
	  (sleep 1)
	  (close-pipe p))

	(let ((p (open-output-pipe "./showfds")))
	  (sleep 1)
	  (close-pipe p))

I take it the intention is to leave the child clean except for the
normal 0, 1 and 2.

        * popen.scm (open-process): Close input-fdes, output-fdes and
        error-fdes after duping them to 0, 1 and 2.

I think this is worth putting in the 1.6 branch too.

The only likely ill effect I could spot from the current code was that
for an output pipe the extra copy of the read side means that if the
spawned program closes stdin, a write from the parent (ie. guile)
won't produce a SIGPIPE.  For instance the following contrivance
currently gets a SIGPIPE only when the sleep finishes and the shell
exits, whereas with the change above it happens immediately.

	(use-modules (ice-9 popen))
	(let ((p (open-output-pipe "exec 0>/dev/null; sleep 10")))
	  (while #t
	    (display "x" p)))