bug#17647: 24.4.50; accept-process-output with process nil can miss output

[Glenn Morris]

It's possible this change did... something.

If I run `make check' in the "current" (before 117400) trunk, I get:

2 unexpected results:
   FAILED  warning-predicate-function-gcc
   FAILED  warning-predicate-rx-gcc

You can see the same failures on hydra.
They seem to start at

http://hydra.nixos.org/build/11864228

which corresponds to

http://lists.gnu.org/archive/html/emacs-diffs/2014-06/msg00165.html


We did not notice these failures because for some reason Emacs still
exits with status 0, despite (kill-emacs 1) being called in
ert-run-tests-batch-and-exit (???).


To reproduce the issue, do (prior to my latest change to flymake-tests.el):

cd test/automated
../../src/emacs -Q -batch -l ./flymake-tests.el \
   -f ert-run-tests-batch-and-exit < /dev/null

It returns far too quickly, as if it did not sleep at all.
I found that (sleep-for 1.0) did nothing, but 
(sleep-for 0.5) (sleep-for 0.5) worked.


Frankly I have no idea what is going on, with the sleeping nor with the
(kill-emacs 1). But it seems to have something to do with processes, and
your change is related to processes, so... over to you! ;)