Re: PATCH PR c++/13989 and PR c++/9844

[Mark Mitchell <mark at codesourcery dot com>]

Fariborz Jahanian wrote:

> ChangeLog:
>
> This patch addresses PRs 13989 and 9844 and more.
> Patch adds support to allow specification of attributes in c++'s function
> return type. This includes, member functions, friend functions and type
> conversion functions. This patch has been bootstrapped and dejagnu
> tested on apple-ppc-darwin. Several tests is included which shows the
> extend of support that this patch provides.
>
>
> OK for mainline?
>
> - Thanks, fariborz
>
> 2004-09-21 Fariborz Jahanian <fjahanian@apple.com>
>         PR c++/13989
>         PR c++/9844
>
>         * tree.c (reconstruct_complex_type): Remove extra "this".
>         * cp/cp-tree.h (check_classfn): Add new argument to declaration.
>         * cp/pt.c (tsubst_friend_function): Add extra argument to call
>         to check_classfn.
>         * cp/cvt.c (ocp_convert): Add support to use type conversion
>         function to vector type.
>         * cp/decl.c (grokfndecl): Add new argument "attrlist".
>         * cp/decl2.c (check_classfn): Added new argument "attrlist"and
>         use the new "attrlist" parameter to build the vector type.
>         * cp/friend.c (do_friend): Pass attrlist to check_classfn call.
>         * cp/parser.c (cp_parser_conversion_type_id): Add attributes, 
> if any,
>          to the parsed type.
>         * config/rs6000/rs6000.c (rs6000_handle_altivec_attribute):
>         Add V4SFmode to case statement.

I think that you should do the attribute processing in grokfndecl, not 
in check_classfn.  There's nothing special about member functions from 
this point of view; we should be processing attributes for non-member 
functions and member functions in the same way.  Where do non-member 
functions get their attributes attached?  Is that too late for member 
functions?  If so, we should maybe move the processing earlier.

Thanks,

--
Mark Mitchell
CodeSourcery, LLC
(916) 791-8304
mark@codesourcery.com