Re: create a watchpoint with trace functionality

[Michael Snyder <msnyder at vmware dot com>]

Nicholas Mc Guire wrote:
> On Wed, 21 Oct 2009, Michael Snyder wrote:
>
> > martin mangard wrote:
> > > Hello
> > >
> > > I'm trying to trace the value of a global status variable of an
> > > application. This variable is changed from various code positions. I
> > > would like to log/trace each change of this variable without stopping
> > > the application (at least for a long time) in order to perform a
> > > longterm test. I'm currently using a hardware-watchpoint
> > >
> > > Is there a possibility to configure a "autocontinue"  after a
> > > watchpoint-event? Which means that the changed value is printed and
> > > the execution continues without any user interaction (pressing "c")?
> > Yes, sure.  See below.
> >
> > > Is ist possible to set a tracepoint on a change of a memory area?
> > A tracepoint is different from a watchpoint, and no, I don't
> > remember that you can do that with a tracepoint (I could be
> > wrong).  
>
> you are right - no support for that in tracepoints - tracepoints are just 
> like breakpoints just that they execute the bytecode provided 
> rather than stoping the process and calling the gdb-frontend. But
> I guess that due to the invasive nature of watchpoints (execution timing wise)
> it would not make that much sense to have a that behavior in tracepoints as
> the prime intention of tracepoints is to allow debugging without breaking
> the temporal behoavior too badly (the impact is of course still significant).

Hardware watchpoints associated with tracepoints would make sense.
Software watchpoints wouldn't.

And of course, there are resource limits associated with hardware
watchpoints.