RE: AW: AW: Patch to DorothyLocker

I need to grab a copy of all the frames in the stack because frames are in
constant use. In earlier versions of DorothyLocker, the printed stack trace
would reflect the line numbers for each frame as was current when the trace
was printed, *not* when the lock was taken or when the problem was detected.
That made things kinda hard to debug. Would some option to limit the copied
stack depth be useful? 

I could also stash a copy of the calling context when the lock is created,
and trim from the stack any common frames.

context=0 is cute! 

-----Original Message-----
From: pyds-dev-admin-iYtK5bfT9M//Ad8WF/[email protected]
[mailto:pyds-dev-admin-iYtK5bfT9M//Ad8WF/[email protected]] On Behalf Of Thomas Klaeger
Sent: Saturday, 25 September 2004 9:59 PM
To: Bauer, Georg
Cc: Garth T Kidd; pyds-dev-iYtK5bfT9M//Ad8WF/[email protected]
Subject: Re: AW: AW: [Pyds-dev] Patch to DorothyLocker

Hello!

One problem is that DorothyLocker gets the full stack trace every time - and
this needs some additional time even if the locking succeeds without
waiting.

A small improvement provides the following patch. With this patch the stack
trace is built without reading the corresponding source lines.

Garth, do you see a possibility to do the work without getting the stack
trace every time? I envision something like getting just the callers frame
on acquire/release and getting the full stack trace only if checking for
re-acquire within the same thread.

Regards,
Thomas

> Even though it does work now, it is horribly slow. Any idea what slows 
> down the system that  much? Even if I don't get that much lock 
> messages, the system is crawling with every access. I noticed this 
> mostly in the aggregator, but other pages are much slower, too. For 
> example the shutdown takes forever. And the system takes far too much 
> CPU resources. I actually kill -9 the system because the shutdown took
forever.
> 
> Without -v it works as expected. So there must still be a bug in the 
> DorothyLocker stuff.
> 
> bye, Georg
>