Re: [rfa/6.0] Fix a thinko in dwarf2loc

[Daniel Jacobowitz <drow at mvista dot com>]

On Tue, Sep 09, 2003 at 05:27:10PM -0400, Elena Zannoni wrote:
> Daniel Jacobowitz writes:
>  > On Sun, Aug 24, 2003 at 12:48:55PM -0400, Daniel Jacobowitz wrote:
>  > > On Sun, Aug 17, 2003 at 04:58:22PM -0400, Daniel Jacobowitz wrote:
>  > > > On Fri, Aug 01, 2003 at 03:32:45PM -0400, Daniel Jacobowitz wrote:
>  > > > > Any variable with DW_OP_reg* requires a frame to read it - what
>  > > > > symbol_read_needs_frame () really means is "is this variable meaningful when
>  > > > > the program is not running".
>  > > > > 
>  > > > > This would show up in the testsuite, but GCC doesn't tend to put locals in
>  > > > > registers without optimization or register keywords everywhere.  Symptom is
>  > > > > that a watchpoint was not deleted when we left its scope, since we thought
>  > > > > it was global.
>  > > > > 
>  > > > > OK?
>  > > > 
>  > > > Ping...
>  > > > 
>  > > > [Does dwarf2loc fall under the maintenance umbrella of the dwarf2
>  > > > reader?]
>  > > 
>  > > Ping?  I would like to fix this bug for 6.0.
>  > 
>  > Jim, Elena?  I hit this bug again in testing the
>  > deprecated_safe_get_selected_frame () patch.
>  > 
>  > > > > 2003-08-01  Daniel Jacobowitz  <drow@mvista.com>
>  > > > > 
>  > > > > 	* dwarf2loc.c (dwarf2_loc_desc_needs_frame): Variables in a
>  > > > > 	register do need a frame.
>  > > > > 
>  > > > > --- gdb-5.3.20030801/gdb/dwarf2loc.c.orig	2003-08-01 14:58:57.000000000 -0400
>  > > > > +++ gdb-5.3.20030801/gdb/dwarf2loc.c	2003-08-01 15:00:05.000000000 -0400
>  > > > > @@ -318,7 +318,7 @@ dwarf2_loc_desc_needs_frame (unsigned ch
>  > > > >  
>  > > > >    free_dwarf_expr_context (ctx);
>  > > > >  
>  > > > > -  return baton.needs_frame;
>  > > > > +  return baton.needs_frame || ctx->in_reg;
>  > > > >  }
>  > > > >  
> 
> Hmmm, the ctx just got freed, you cannot dereference it. Need to use a

Eek, I'm embarassed.  You are so right.

> local variable.  What you are really saying here is that the
> needs_frame variable is not sufficient to describe what's needed. Can
> the case be added, so that needs_frame becomes OK? Proabably you
> need another need_frame_* function.

No, I can't.  The needs_frame_* functions are hooks for things which
are external to the expression evaluator - concepts like "object" and
"thread local storage address" which require information.  But the
evaluator won't actually try to read from the register when parsing a
DW_OP_reg1; it just sets in_reg.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer