ojr@itk.unit.no (04/13/89)
Description:
We are running SUN OS4.0.1 and emacs 18.53.3. When a broadcast message
arrives at a terminal/window where emacs is running, emacs stops
accepting input and hangs with 100% cpu being used. ^G does not work.
It is possible to restart emacs with an INT signal to the emacs
process from another window.
What I found:
Something strange happens with the SUN i/o system when a broadcast is
delivered. I looks as if no signals are delivered to the emacs process
when the signal originates in keyboard events (neither SIGINT or
SIGIO). The function 'wait_reading_process' loops forever because
select (line 1373) returns with bit 0 in 'Available' set, instead of
delivering the SIGIO signal. As this bit is ignored in the function
the result is an eternal active wait.
Solution (hack?):
What I did was checking on bit 0 in 'wait_reading_process' and sending
the missing SIGIO to emacs if the bit is set. This looks as if it
fixes the problem. The diff is listed below:
lisa% diff -c2 process.c~ process.c
*** process.c~ Sat Oct 8 17:24:20 1988
--- process.c Thu Apr 13 11:53:26 1989
***************
*** 1399,1402 ****
--- 1399,1413 ----
error("select error: %s", sys_errlist[xerrno]);
}
+ #ifdef SUN_REGTEK
+ else if (nfds > 0 && (Available & 1))
+ {
+ /*
+ * Input from standard input should go to
+ * interrupthandler.
+ */
+
+ kill(getpid(), SIGIO);
+ }
+ #endif
/* Check for keyboard input */
You have to have -DSUN_REGTEK in the compiler command when compiling.
process.c.
Best regards
ojr@itk.unit.no Ornulf Jan Rodseth
SINTEF Dep. of Automatic Control, N-7034 TRONDHEIM
NORWAYsdry@vax5.CIT.CORNELL.EDU (04/20/89)
Ornulf Rodseth <ojr@itk.unit.no> recently submitted an analysis of a problem
involving emacs and [r]wall, together with a fix.
We had encountered the same problem here, and I agree with the analysis given.
(Actually, we first detected the problem with 18.52. The same fix applies
to both 18.52 and 18.53.)
I don't see the need for -DSUN_REGTEK, however - my own fix doesn't use it.
So here is an alternate solution (pick your favorite):
*** process.c.old1853 Sat Aug 20 13:27:41 1988
--- process.c Wed Apr 12 17:00:56 1989
***************
*** 1400,1403 ****
--- 1400,1415 ----
}
+ /* Bug Fix on 14 March 1989 by S.Gelato, Cornell Astronomy Dept.
+ * --
+ * select(2) keeps returning "stdin ready for read|exception" until
+ * something is done with the descriptor (at the very least, an
+ * ioctl(0, FIONREAD) should be attempted).
+ * get_input_pending() is too lazy, and won't do this unless there
+ * has been an interrupt.
+ * wall(1) apparently causes this condition to be set without any
+ * SIGIO, so we need to do something about it... */
+
+ if ((Available | Exception) & input_wait_mask) gobble_input();
+
/* Check for keyboard input */
/* If there is any, return immediately
-----------------------------------------------------------------------
Sergio Gelato gelato@astrosun.tn.cornell.edu
or sdry@crnlvax5.bitnet