On Wed, Jul 29, 2020 at 01:28:02PM +0200, Oleg Nesterov wrote:
> On 07/29, Luc Van Oostenryck wrote:
> >
> > OTOH, I wonder what can be done without first evaluating
> > (the type of) the controlling expression and the types of the map
> > (if I understand correctly, evaluation is avoided in dissect).
>
> Yes. I'll try to think a bit more, but so far I think I'll simply
> send the patch below.
...
> Of course, technically this is wrong, it looks as if all 3 variables are
> modified. But not that bad imo, dissect doesn't even try to be "precise",
> and this output still looks useful for the indexing/etc.
>
> --- a/dissect.c
> +++ b/dissect.c
> @@ -342,7 +342,6 @@ again:
> case EXPR_TYPE: // [struct T]; Why ???
> case EXPR_VALUE:
> case EXPR_FVALUE:
> - case EXPR_GENERIC:
>
> break; case EXPR_LABEL:
> ret = &label_ctype;
> @@ -472,6 +471,17 @@ again:
> } while ((expr = expr->down));
> }
>
> + break; case EXPR_GENERIC: {
> + struct type_expression *map;
> +
> + do_expression(U_VOID, expr->control);
> +
> + for (map = expr->map; map; map = map->next)
> + ret = do_expression(mode, map->expr);
> + if (expr->def)
> + ret = do_expression(mode, expr->def);
> + }
> +
> break; case EXPR_SYMBOL:
> ret = report_symbol(mode, expr);
> }
Yes, that should do the 'walking'. The returned type will just be
quite arbitrary, but I don't know how much it matters.
-- Luc
