Re: Display of read/access watchpoints when HAVE_NONSTEPPABLE_WATCHPOINT

[Daniel Jacobowitz <drow@false.org>]

On Tue, Apr 06, 2004 at 12:14:12PM +0200, Orjan Friberg wrote:
> Eli Zaretskii wrote:
> >
> >Anyway, from your description, it is quite clear that if a target
> >defines HAVE_NONSTEPPABLE_WATCHPOINT, GDB must call
> >target_stopped_data_address before it disables the watchpoint and
> >steps over it, or else the target end should store the necessary info
> >somewhere and deliver it when target_stopped_data_address is called.
> >
> >In other words, it sounds like a bug.
> 
> I had missed a previous thread on the subject, initiated by Paul Koning 
> (starting at 
> http://sources.redhat.com/ml/gdb-patches/2003-05/msg00506.html).  Paul 
> sent me a patch for 5.3 which no longer applies cleanly, and with his 
> blessings I'm trying to update his patch to the current CVS.
> 
> Doing that, I encountered a problem with read watchpoints (rwatch 
> command): in insert_breakpoints, before calling insert_bp_location, the 
> current value of the watched expression is read.  Then in 
> bpstat_stop_status, when the watchpoint has hit, there's the following 
> code snippet:
> 
> case WP_VALUE_CHANGED:
>   if (b->type == bp_read_watchpoint)
>     {
>       /* Don't stop: read watchpoints shouldn't fire if
>          the value has changed.  This is for targets
>          which cannot set read-only watchpoints.  */
>       bs->print_it = print_it_noop;
>       bs->stop = 0;
>       continue;
>     }
>   ++(b->hit_count);
>   break;
> 
> So, if we're "read watching" a variable that indeed has changed since we 
> read it before inserting it we decide not to stop.  Here's another 
> example (foo is rwatched):
> 
>   volatile int foo = 0;
>   int a;
> 
>   a = foo;   /* foo read; stops.  */
>   foo = 1;
>   a = foo;   /* foo read; doesn't stop because value changed.  */
> 
> 
> The "Don't stop" comment is consistent with the behaviour, so it seems 
> like this is intended.  What am I missing here?

Presumably, the relevant part of the comment is the second half: "this
is for targets which cannot set read-only watchpoints".  On the other
hand, it does not sound like that code will work for targets which
_can_ set read-only watchpoints.  The patch was:

2000-03-21  Eli Zaretskii  <eliz@is.elta.co.il>

        * breakpoint.c (bpstat_stop_status): Don't stop if a read
        watchpoint appears to break, but the watched value changed.

And since Eli demonstrably knows more than I do about how watchpoints
work, I'm going to defer further comments at this point :)

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer