[m-dev.] for review: new debugger command set (part 3 of 4)

[Zoltan Somogyi]

> > @@ -61,6 +65,7 @@
> >  MR_agc_dump_stack_frames(MR_Internal *label, MemoryZone *heap_zone,
> >  	Word *stack_pointer, Word *current_frame)
> >  {
> > +	Word saved_regs[MAX_FAKE_REG];
> >  	int i, var_count;
> >  	const MR_Stack_Layout_Vars *vars;
> >  	Word *type_params, type_info, value;
> > @@ -87,8 +92,8 @@
> >  		var_count = layout->MR_sll_var_count;
> >  		vars = &(layout->MR_sll_var_info);
> >  
> > -		type_params = MR_trace_materialize_typeinfos_base(vars,
> > -			top_frame, stack_pointer, current_frame);
> > +		type_params = MR_materialize_typeinfos_base(vars,
> > +			NULL, stack_pointer, current_frame);
> 
> Isn't this the same thing?  Aren't the registers actually saved in
> this case.  Either the NULL should be "saved_regs" or you need an
> XXX here (or both).

I have added the following comment:

/*
** XXX For the top stack frame, we should pass a pointer to
** a filled-in saved_regs instead of NULL. For other stack
** frames, passing NULL is fine, since output arguments are
** not live yet for any call except the top one.
*/

> > +MR_get_type_and_value_filtered(const MR_Stack_Layout_Var *var, 
> > +	Word *saved_regs, const char *name, Word *type_info, Word *value)
> >  {
> >  	return ((strncmp(name, "TypeInfo", 8) != 0)
> >  	       && (strncmp(name, "ModuleInfo", 10) != 0)
> >  	       && (strncmp(name, "HLDS", 4) != 0)
> > -	       && MR_trace_get_type_and_value(var, NULL, type_info, value));
> > +	       && MR_get_type_and_value(var, saved_regs, NULL,
> > +			type_info, value));
> >  }
> 
> You probably want to filter on TypeClassInfo as well (I think we now
> generate TypeClassInfo variables instead of just TypeInfo).

The big change has nothing to do with type classes.

However, in another workspace I just got the debugger working with type
classes, and that did indeed require this change (not just here, but in some
other places as well).

Zoltan.