Re: [rfc] Fix register cache invalidation

[Pedro Alves <pedro at codesourcery dot com>]

On Tuesday 14 July 2009 15:59:28, Ulrich Weigand wrote:

> when testing the Cell/B.E. combined debugger patches I ran into a situation
> where registers_changed was not called soon enough.
> 
> A while ago I proposed a patch to simplify register invalidation (as part of
> a infrun cleanup series):
> http://sourceware.org/ml/gdb-patches/2008-12/msg00120.html
> This cleanup actually fixes the combined debugger problem as well.

Curious.  Are we sure we're not hidding some other bug then?

> At the time, Pedro pointed out that parts of that patch weren't safe in the
> non-stop case.  The version below omits those parts, and only changes code
> to simplify calling registers_changed and resetting waiton_ptid.

> Tested on powerpc64-linux with no regressions.
> Any comments?  I'd like to commit this in a couple of days.

I like it!

> 	> 	(fetch_inferior_event): Remove check for always-true condition.   
> 

> -  /* We can only rely on wait_for_more being correct before handling
> -     the event in all-stop, but previous_inferior_ptid isn't used in
> -     non-stop.  */
> -  if (!ecs->wait_some_more)
> -    /* We'll update this if & when we switch to a new thread.  */
> -    previous_inferior_ptid = inferior_ptid;
> +  /* We'll update this if & when we switch to a new thread.  */
> +  previous_inferior_ptid = inferior_ptid;

This is actually a bug in the async/all-stop (you're not introducing
it, it's already there).  We shouldn't be clobbering previous_inferior_ptid
on every internaly handled event.  Just noting it --- don't worry about it.

-- 
Pedro Alves