On Tue, Feb 5, 2019 at 1:37 AM Alexandre Oliva <aol...@redhat.com> wrote:
> On Jan 31, 2019, Jason Merrill <ja...@redhat.com> wrote:
>
> > Let's use strip_using_decl instead
>
> Aah, nice! Thanks, I'll make the changes, test them, and post a new patch.
>
>
> >> @@ -13288,7 +13295,8 @@ grok_special_member_properties (tree decl)
> >> {
> >> tree class_type;
> >> - if (!DECL_NONSTATIC_MEMBER_FUNCTION_P (decl))
> >> + if (TREE_CODE (decl) != USING_DECL
> >> + && !DECL_NONSTATIC_MEMBER_FUNCTION_P (decl))
> >> return;
>
> > Is there a reason not to use it here, as well?
>
> The using decl will take us to a member of a different class, and this
> function takes the DECL_CONTEXT of decl and adjusts the properties of
> that class. If we followed USING_DECLs in decl that early, we'd adjust
> (again?) the member properties of USING_DECL_SCOPE(original using decl),
> rather than of DECL_CONTEXT (original using decl) as intended.
But a little further in, copy_fn_p will also check
DECL_NONSTATIC_MEMBER_FUNCTION_P and crash.
This function used to do nothing for USING_DECL (since its TREE_TYPE
wasn't METHOD_TYPE), right? It should continue to do nothing.
Come to think of it, how about fixing DECL_NONSTATIC_MEMBER_FUNCTION_P
to be false for non-functions rather than messing with all these
places that use it?
Jason
