topic Re: traceback information in Operating System - OpenVMS

traceback information

John Carlier — Thu, 16 Mar 2006 09:16:08 GMT

For a general error routine we wish to extract the traceback information and write this to a central log file. This is for an ACMS application.
I can get the stack information via LIB$GET_CURR_INVO_CONTEXT and LIB$GET_PREV_INVO_CONTEXT. From this I can get to the PDSC but how do I now get the NAME of each module? Is this the correct approach or is there a simple way to get the traceback information?

Re: traceback information

Ian Miller. — Thu, 16 Mar 2006 09:43:57 GMT

On which platform and which version of VMS?
The calling standard and related things are somewhat different on I64, Alpha and VAX.
There was talk of a callable interface to trace. The symbol table information is in the .EXE but its format is not documented.

Re: traceback information

Hein van den Heuvel — Thu, 16 Mar 2006 10:03:07 GMT

John Carlier!
What a blast from the past!

Happened to mention your name recently (17-Jan) on the www.decnet.nl organized 'Digital' reunion/newyears drink.

Met vriendelijke groetjes uit Nashua, NH,
Hein.

Re: traceback information

John Carlier — Thu, 16 Mar 2006 10:16:13 GMT

Sorry should have put it in the original post.

OpenVMS V7.2-1H1 on Alpha

Hein,

Good to hear from you after all this time.
Digital right-sized me 11 year ago, now with VXcompany.

Regards

John.

Re: traceback information

Volker Halle — Thu, 16 Mar 2006 11:00:50 GMT

John,

there is an undocumented interface to TBK$SHOW_TRACEBACK, see ATW 5175:

http://h71000.www7.hp.com/wizard/wiz_5175.html

It may be possible to use the user action routine (USER_ACTION_PROCEDURE) to write the traceback info to some user specific log file instead of SYS$OUTPUT.

Volker.

Re: traceback information

John Gillings — Thu, 16 Mar 2006 21:24:15 GMT

John,

The TBK$ routines can work, but aren't necessarily architecture independent and subject to change.

It may be simpler and more reliable just to use the LIB$*INVO_CONTEXT routines and dump out non-symbolic information. Worry about correlating it to routine names only if actually necessary. How often do you need to do it? It's fairly straightforward - use the MAP file for the image.

My preference is to keep error handing routines as simple as possible, to avoid the issue of trying to debug "meta" problems, or having logic errors in the handling code mask or misrepresent the real problem.