On 07/30, Luc Van Oostenryck wrote:
>
> On Thu, Jul 30, 2020 at 05:08:37PM +0200, Oleg Nesterov wrote:
> > On 07/29, Luc Van Oostenryck wrote:
> > > The returned type will just be
> > > quite arbitrary, but I don't know how much it matters.
> >
> > Of course. And this is not good. For example:
> >
> > void func(void)
> > {
> > struct B *b; struct C *c; struct D *d;
> > _Generic(a,
> > int: b,
> > void*: c,
> > default: d
> > ) ->mem++;
> > }
> >
> > output:
> >
> > 1:6 def f func void ( ... )
> > 3:18 func def . v b struct B *
> > 3:31 func def . v c struct C *
> > 3:44 func def . v d struct D *
> > 4:18 func --- v a bad type
> > 5:33 func --m . v b struct B *
> > 6:33 func --m . v c struct C *
> > 7:33 func --m . v d struct D *
> > 8:11 func -m- m D.mem bad type
> >
> > But I do not know how to improve it without serious complications, and
>
> Are you thinking about calling evaluate_symbol_list()
I meant, it is not simple to teach dissect() to handle this case correctly.
It understand the types, but for example it doesn't even try to distinguish
"int" and "float".
And I would like to avoid evaluate_expression/etc.
> or about
> something else?
And something else. See the example above, this code is incomplete and in this
case evaluate can't help. Ideally dissect should also report the (possible) usage
of B.mem and C.mem.
> > (so far) I think it doesn't worth the effort.
>
> Yes, _Generic() clearly makes things a bit more complicated here.
> Same for __auto_type,
Yes, but hopefully dissect needs much more simple changes to handle __auto_type.
Thanks,
Oleg.
