Tags: patch


Compile the following C program, close_and_sit.c:

#include <stdio.h>
#include <unistd.h>
void main() {
  fclose(stdout);
  fclose(stderr);
  while (1) sleep(1);
}

Then run this snippet of Lisp, updating the executable file name
appropriately:

(make-process
 :name "close_and_sit"
 :command '("./close_and_sit")
 :connection-type 'pipe)
(make-process :name "true" :command '("true"))
(message "accept-process-output")
(accept-process-output nil 1)

This will cause Emacs to hang; accept-process-output will never time
out.

There are two bugs which combine to cause this.

The first bug is in our timeout handling.  The attached patch resolves
that, which causes accept-process-output to time out properly and fixes
the bug.

Paul, this patch partially reverts a change you made to cache
current_timespec.  Does it seem right to you?

The second bug is that when a child process closes stdout/stderr, we
continue watching the pipe, generating spurious events.  (We should
still be timing out despite the spurious events - real events could also
cause this hang, this bug just makes it easier to trigger the hang.)
I'm still working on a patch to fix the second bug.