Re: emacs abort backtrace from C-g in Message mode

[David Hunter]

Richard Stallman wrote:
> Does this fix it?
<patch against process.c 1.450>

I finally reproduced this bug.  By opening a TCP network stream 
(`open-network-stream') to a localhost server that connects and immediately 
closes the connection, then attempting to write to the stream, a reliable 
SIGPIPE is thrown.  strace shows that emacs fails to unblock SIGPIPE or restore 
the signal handler.

Richard fixed the blocked SIGPIPE.  The following patch restores the signal 
handler.  I have not verified that it's possible to crash emacs by killing the 
X server after a SIGPIPE (or that anyone would see the result), but strace 
output validates the change.

-Dave

*** process.c   14 May 2005 14:06:33 -0000      1.451
--- process.c   17 May 2005 04:02:22 -0000
***************
*** 5134,5139 ****
--- 5134,5140 ----
   int rv;
   struct coding_system *coding;
   struct gcpro gcpro1;
+   volatile SIGTYPE (*old_sigpipe)();
 
   GCPRO1 (object);
 
***************
*** 5258,5264 ****
       while (len > 0)
        {
          int this = len;
-         SIGTYPE (*old_sigpipe)();
 
 	  /* Decide how much data we can send in one batch.
             Long lines need to be split into multiple batches.  */
--- 5259,5264 ----
***************
*** 5401,5406 ****
--- 5401,5407 ----
 #endif /* not VMS */
   else
     {
+       signal (SIGPIPE, old_sigpipe);
 #ifndef VMS
       proc = process_sent_to;
       p = XPROCESS (proc);