Re: [RFC] better dwarf checking for values on the stack

[Doug Evans <dje at google dot com>]

On Fri, Sep 11, 2009 at 11:48 AM, Tom Tromey <tromey@redhat.com> wrote:
>>>>>> "Doug" == Doug Evans <dje@google.com> writes:
>
> Doug> 2009-09-10 ?Doug Evans ?<dje@google.com>
> Doug> ? Add better checking for values on stack.
> Doug> ? * dwarf2expr.h (dwarf_value_location): Rename DWARF_VALUE_STACK to
> Doug> ? DWARF_VALUE_DWARF_STACK, all uses updated.
> Doug> ? New enum DWARF_VALUE_MEMORY_STACK.
> Doug> ? * dwarf2expr.c (execute_stack_op, case DW_OP_fbreg): Mark location
> Doug> ? as DWARF_VALUE_MEMORY_STACK.
> Doug> ? (execute_stack_op, case DW_OP_call_frame_cfa): Ditto.
> Doug> ? (execute_stack_op, cases DW_OP_deref, DW_OP_deref_size): Mark
> Doug> ? location as DWARF_VALUE_MEMORY.
> Doug> ? (execute_stack_op, case DW_OP_piece): Remove unused addr_or_regnum.
>
> I think the tracking idea is sound, but I think this implementation has
> a flaw.
>
> It seems to me that the inferior-stack-ness of a value must be an
> attribute carried alongside the value on the dwarf stack.
>
> Here's a really bogus example that I hope still shows what I mean:
>
> ?DW_OP_call_frame_cfa
> ?DW_OP_deref
> ?DW_OP_call_frame_cfa
> ?DW_OP_drop
>
> After the drop, ctx->location will be DWARF_VALUE_MEMORY_STACK.
> However, this is incorrect.
>
> Tom
>

Yeah, I was wondering about this.  [It's what prompted my question on IRC.]