[kzeitler@lucent.com: Re: dabbrev in minibuffer]

[kzeitler@lucent.com: Re: dabbrev in minibuffer]

[Richard Stallman]

Are there any people on HPUX and Solaris that can try to debug these
problems with C-g?  The HPUX case may have been fixed by the recent
change regarding SA_RESTART; if so, it is conceivable that adding

    #define BROKEN_SA_RESTART

in the s/*.h file for Solaris might fix the problem.  Does it?

------- Start of forwarded message -------
X-Authentication-Warning: sfsw51.de.lucent.com: kzeitler set sender to kzeitler@lucent.com using -f
To: rms@gnu.org
Subject: Re: dabbrev in minibuffer
Emacs: Lovecraft was an optimist.
From: Klaus Zeitler <kzeitler@lucent.com>
Date: 17 Jul 2002 16:26:49 +0200
In-Reply-To: <200207141522.g6EFMgM20517@aztec.santafe.edu>

>>>>> "Richard" == Richard Stallman <rms@gnu.org> writes:
    Richard> 
    Richard> One question is, does interrupting a loop with C-g ever work on
    Richard> those systems?  Please try executing (while t nil) and see if you
    Richard> can quit out of that.
    Richard> 
ok tried this and I can't quit on HP at all and on SUN not always.

I've taken over emacs production here in my department and I've got emacs 21.2
now running on the following 4 systems:

HP-UX-B.10.20
HP-UX-B.11.11
SunOS-5.7
SunOS-5.8

On Monday I additionally installed the CVS version, but production on
HP 10.20 and 11.11 failed (in process.c, I've sent a bug report to pretest)

The SunOS CVS version behaves exactly like 21.2. Here's what I've found out
so far (not much):

1. start emacs -q --no-site-file
2. in scratch buffer execute (while t nil)
3. C-g works
4. make a new frame (C-x 5 2)
5. in scratch buffer execute (while t nil)
6. C-g doesn't work anymore

On both HP versions emacs wont quit that while loop

Klaus

- -- 
 ------------------------------------------
|  Klaus Zeitler      Lucent Technologies  |
|  Email:             kzeitler@lucent.com  |
 ------------------------------------------
- ---
There exist tasks which cannot be done by more
than 10 men or fewer than 100. -- Steele's Law
------- End of forwarded message -------

Re: [kzeitler@lucent.com: Re: dabbrev in minibuffer]

[Klaus Zeitler]

>>>>> "Richard" == Richard Stallman <rms@gnu.org> writes:
    Richard> 
    Richard> Are there any people on HPUX and Solaris that can try to debug
    Richard> these problems with C-g?  The HPUX case may have been fixed by
    Richard> the recent change regarding SA_RESTART; if so, it is conceivable
    Richard> that adding
    Richard> 
    Richard>     #define BROKEN_SA_RESTART
    Richard> 
    Richard> in the s/*.h file for Solaris might fix the problem.  Does it?
    Richard> 
cvs version contains BROKEN_SA_RESTART, but this doesn't help
neither on HP-UX 10.20 nor on 11.11.
So I guess that wont do for Solaris either (but haven't tried)

-- 
 ------------------------------------------
|  Klaus Zeitler      Lucent Technologies  |
|  Email:             kzeitler@lucent.com  |
 ------------------------------------------
---
Steinbach's Guideline for Systems Programming
 Never test for an error condition you don't know how to handle.

21.2 C-g problems on HP-UX-B.10.20, 11.11, Solaris 5.7 and 5.8

[Klaus Zeitler]

>>>>> "Richard" == Richard Stallman <rms@gnu.org> writes:
    Richard> 
    Richard> Are there any people on HPUX and Solaris that can try to debug
    Richard> these problems with C-g?  The HPUX case may have been fixed by
    Richard> the recent change regarding SA_RESTART; if so, it is conceivable
    Richard> that adding
    Richard> 
    Richard>     #define BROKEN_SA_RESTART
    Richard> 
    Richard> in the s/*.h file for Solaris might fix the problem.  Does it?
    Richard> 

I've started looking into this on Solaris 5.8 in cvs version.

The first question that came up was, is it really necessary to set
BROKEN_SIGINFO and use polling instead?
Since I don't know I made sure that BROKEN_SIGINFO is not set, i.e.
I've added to sol2-5.h the line:
#undef BROKEN_SIGIO
this alone didn't do it cause then the function unrequest_sigio
isn't defined (properly) in sysdep.c.
Thus I made sure that FASYNC is defined (from /usr/include/sys/file.h) by
adding to sysdep.c (seems that there's no define for SOLARIS2_5)
#ifdef SOLARIS2_4
#include <sys/file.h>
#endif

with these changes emacs appears to work (i.e. C-g works, but I haven't
thoroughly tested yet, especially since I don't know the developers
intentions to disable SIGIO)

Still I think that the problem why C-g doesn't work with polling instead
of SIGIO may need further investigation.

If I'm on the right track I can investigate Solaris 5.7 and HP-UX next week.

Klaus

-- 
 ------------------------------------------
|  Klaus Zeitler      Lucent Technologies  |
|  Email:             kzeitler@lucent.com  |
 ------------------------------------------
---
I didn't know it was impossible when I did it.

Re: 21.2 C-g problems on HP-UX-B.10.20, 11.11, Solaris 5.7 and 5.8

[Richard Stallman]

It is definitely good to make interrupt input work and avoid the need
for polling.  Did you test it with -nw as well as talking to X?
Those modes work differently so it is important to test both.

    Thus I made sure that FASYNC is defined (from /usr/include/sys/file.h) by
    adding to sysdep.c (seems that there's no define for SOLARIS2_5)
    #ifdef SOLARIS2_4
    #include <sys/file.h>
    #endif

Other files include sys/file.h unconditionally; I'd rather
handle this file like the others and avoid the conditional.

If it works, please send the patch that you recommend.

    Still I think that the problem why C-g doesn't work with polling instead
    of SIGIO may need further investigation.

It would be useful to fix that too if you can.