Pedro Alves wrote:
> Thanks for the update.
> 
> On Thursday 16 April 2009 18:08:07, David Daney wrote:
>>>> Reported instruction location of the watchpoint trigger is one 
>>>> instruction later and one line later.
>>> Why is that?
>> I was mistaken, it is earlier, not later.  With hardware watch points, 
>> $pc points to the faulting instruction, with software watch points $pc 
>> points to the following instruction.  In a couple of tests, this results 
>> in the reported line number being different than the expected value.
> 
> I'm missing something and am still confused.  PC points at the
> faulting instruction when the target reports the watchpoint hit to
> infrun --- .  That step-over-watchpoint dance that patch 1/2 took care of comes
> into play.  That should move the inferior to the following instruction, evaluate
> the watchpoint expression, notice the value changed, and report to the
> user.  Why does the PC still point at the faulting instruction?
> 

This patch had a small problem:

http://sourceware.org/ml/gdb-patches/2009-04/msg00245.html

I was calling registers_changed() to clear the register cache, but then 
immediately calling read_pc() which caused it to be reloaded.  After the 
single step to move past the watchpoint, the old cached register values 
were used instead of the current values.

The fix: Move the registers_changed() call after read_pc().

Tested on x86_64-unknown-linux-gnu and mips64-unknown-linux-gnu with no 
regressions.

OK to commit?