Re: [PATCH 0/3] MR8507: widl: Drop an incorrect check on runtimeclass definitions.

[Rémi Bernon (＠rbernon)]

Rémi Bernon (@rbernon) commented about tools/widl/typetree.c:
>  return type;
>
> }
>
> +static void set_constructor_runtimeclass(type_t *runtimeclass, attr_list_t *attrs)
> +{
> - const attr_t *attr;
> - if (!attrs) return;
> - LIST_FOR_EACH_ENTRY(attr, attrs, const attr_t, entry)
> - {
> -
>    if (attr->type == ATTR_ACTIVATABLE || attr->type == ATTR_COMPOSABLE)
> -
>    {
> -
>        const expr_t *value = attr->u.pval;
> -
>        if (value->u.tref.type->details.iface)
> -
>            value->u.tref.type->details.iface->runtime_class = runtimeclass;

It doesn't seem safe to access expression like that. The `activatable` attribute may have an interface or not. When it has, the expression is an `EXPR_MEMBER (decl_spec, contract_req)` but when not, the expression is a `contract_req` expression only `EXPR_GTREQL (decl_spec, version)` and its decl_spec type is an apicontract, not an interface.

This was enough until now, but maybe we should start using dedicated attribute values instead of abusing expressions.