Core dump takes forever, Traceback Command?

[Tom Allen]

I have a program in the "BETA TEST" stage and alas it sometimes
crashes with segmentation faults, etc. HP-UX seems to take 2-3
minutes doing something (making a core file?) after the fault
occurs. Why does it take so long? I'm running on a 735 with a
fast and wide disk. The program is a X/Motif application which
takes about 20mbyte of memory

Also does HP-UX have any sort of traceback command that will
provide a list of called routines with line numbers leading to
the point at which the program crashed. I know the debugger has
this but it hard to get non programming users to use xdb.

---
Tom Allen
Ford Research Laboratory
t...@smail.srl.ford.com

[Wuyts Jan]

Your application is about 20MB, so you must have noticed the size of the core file.
It takes a while to create such a huge file.

When you are not interrested in the core file you could do

rm core ; touch core ; chmod 400 core

This creates a read-only and empty core file. This prevents the creation of the
real core file. So the prompt returns immediately after the crash.

Can't help you on the traceback. I've always wondered why that isn't the default.
It is not really that difficult. I've implemented one myself once but that wasn't on HP-UX.

---
---
Jan Wuyts (RS66), Internet: wuy...@rsd.bel.alcatel.be
Alcanet: btmx.wuytsj
Alcatel Bell Telephone
Berkenrodelei 33 Phone: (+32) 3/829.5479
B-2660 Hoboken Fax: (+32) 3/829.5502
Belgium.

[Dennis Handly]

/ t...@holo5.srl.ford.com (Tom Allen) / 10:56 am Jun 21, 1994 /

>Also does HP-UX have any sort of traceback command that will
>provide a list of called routines with line numbers leading to
>the point at which the program crashed.

What language is it written in? Pascal will automatically give you
a traceback (no line numbers, just hex offsets from procedure start).

You can call the Trap Library's handler in C by:

extern void U_INIT_TRAPS(void);

U_INIT_TRAPS();

You'll need to link with -lcl.
You also won't get a core file.

[Roger Knight]

It would be an awful lot better if HP-UX did this by default itself (like
Domain/OS does) instead of relying on the application programmer to do
it manually.

We use Mentor s/w and they have trapped a lot of fault conditions to
provide tracebacks but there are still some things (like an obscure signal
being received at the wrong time, with a following wind, when there's
a full moon.....) which cause 64MB core dumps. Sod's law dictates these
always happen to an NFS mounted directory with only 60MB of free space
so the NFS host crashes after locking up 2 machines and the ethernet
for 15 minutes!!!!

When I complained about this I was told to use a workaround - create a
root owned file called core in the affected directory ( i.e. every directory
under every user's $HOME !) with all the rwx rights removed. This means
users that just use the applications and don't know UN*X get very confused
when they can't remove directories and come crying to me for help. Needless
to say, we don't use the workaround.

HP said they may consider implementing "set coresize = 0" (like SUN)
some time in the future (don't hold your breath).

-Roger Knight
DISCLAIMER: I said it, not my company.