Re: Incorrect signal handling under SYSV machines

[Chris R Dalton <crd@hplb.hpl.hp.com>]

Hi,

> I have used es-0.9-alpha1 for many moon under SunOS 4.1.[34] and DEC
> Alpha UNIX V4.0.  [Note: I always have `interrupt' set to Control C
> under the terminal driver.]  Whenever I hit Control C, most normal
> programs interactively activated under the shell are killed and I get
> a fresh command prompt.  If I was in the middle of forming a command
> line, it is killed and I get a fresh prompt.  Even if I had typed
> nothing, I get a fresh prompt. [...]
> 
> Under HP-UX 10.20, when I hit Control C twice at any points during a
> session under the es shell, the shell ``terminate[s] on signal 2.''

Hmm, yes...

I came across this a while ago, and I think the following fixed it for
me, pending further investigation (which is still pending 8^/ ):

*** signal.c.orig	Tue May 30 12:13:51 1995
--- signal.c	Fri Jul  7 10:36:44 1995
***************
*** 65,70 ****
--- 65,71 ----
  /*
   * the signal handler
   */
+ static Sighandler setsignal(int sig, Sighandler handler);
  
  /* catcher -- catch (and defer) a signal from the kernel */
  static void catcher(int sig) {
***************
*** 79,86 ****
  		++sigcount;
  	}
  	interrupted = TRUE;
! 	if (slow)
  		longjmp(slowlabel, 1);
  }
  
  
--- 80,91 ----
  		++sigcount;
  	}
  	interrupted = TRUE;
! 	if (slow) {
! #if USE_SIGACTION
! 		setsignal(sig, catcher);
! #endif
  		longjmp(slowlabel, 1);
+ 	}
  }
  
  
It looks like HP-UX is rather more strict, or something, about the
circumstances under which it restores the original signal mask in
a signal handler, using sigaction.

The HP-UX sigaction(2) manual entry says:

    When a signal is caught by a signal-catching function installed
    by sigaction, a new mask is calculated and installed for the
    duration of the signal-catching function, or until a call is
    made to sigprocmask() or sigsuspend() (see sigprocmask(2) and
    sigsuspend(2)).  [...] If and when the signal-catching function
    returns normally, the original signal mask is restored.

The key is in that last sentence.  When you press the INTR key during
command entry, the `slow' flag is true, and the code takes the longjmp
out of the signal handler.  The longjmp doesn't count as `returning
normally' from the signal handler, so the signal mask isn't restored.


Hope this helps,

Chris
-- 
Chris R Dalton  (crd@hplb.hpl.hp.com)
Hewlett-Packard Laboratories, Stoke Gifford, Bristol, UK  BS12 6QZ
Phone: +44 117 922 8744 or +44 117 979 9910 x28744  FAX: +44 117 922 8924