From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <gcc-patches-return-305791-listarch-gcc-patches=gcc.gnu.org@gcc.gnu.org>
Received: (qmail 2775 invoked by alias); 28 Oct 2011 00:01:56 -0000
Received: (qmail 2758 invoked by uid 22791); 28 Oct 2011 00:01:55 -0000
X-SWARE-Spam-Status: No, hits=-2.4 required=5.0	tests=AWL,BAYES_00,RP_MATCHES_RCVD
X-Spam-Check-By: sourceware.org
Received: from acsinet15.oracle.com (HELO acsinet15.oracle.com) (141.146.126.227)    by sourceware.org (qpsmtpd/0.43rc1) with ESMTP; Fri, 28 Oct 2011 00:01:37 +0000
Received: from ucsinet21.oracle.com (ucsinet21.oracle.com [156.151.31.93])	by acsinet15.oracle.com (Switch-3.4.4/Switch-3.4.4) with ESMTP id p9S01ZaJ029698	(version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=OK);	Fri, 28 Oct 2011 00:01:36 GMT
Received: from acsmt357.oracle.com (acsmt357.oracle.com [141.146.40.157])	by ucsinet21.oracle.com (8.14.4+Sun/8.14.4) with ESMTP id p9S01Ybg003991	(version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=NO);	Fri, 28 Oct 2011 00:01:34 GMT
Received: from abhmt112.oracle.com (abhmt112.oracle.com [141.146.116.64])	by acsmt357.oracle.com (8.12.11.20060308/8.12.11) with ESMTP id p9S01TWL013623;	Thu, 27 Oct 2011 19:01:29 -0500
Received: from [192.168.1.4] (/79.52.193.167)	by default (Oracle Beehive Gateway v4.0)	with ESMTP ; Thu, 27 Oct 2011 17:01:28 -0700
Message-ID: <4EA9F0D3.4040800@oracle.com>
Date: Fri, 28 Oct 2011 01:44:00 -0000
From: Paolo Carlini <paolo.carlini@oracle.com>
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:7.0.1) Gecko/20110929 Thunderbird/7.0.1
MIME-Version: 1.0
To: Paolo Carlini <pcarlini@gmail.com>
CC: Jason Merrill <jason@redhat.com>,        "gcc-patches@gcc.gnu.org" <gcc-patches@gcc.gnu.org>
Subject: Re: [C++ Patch / RFC] PR 50870
References: <4EA81BB4.8090900@oracle.com> <4EA826EA.4040900@redhat.com> <4EA82C27.3020700@oracle.com> <4EA8350D.8060907@redhat.com> <4EA8879F.40906@oracle.com> <4EA88E3A.4010808@redhat.com> <4EA88F7B.3060706@oracle.com> <4EA89101.6050001@oracle.com> <4EA8D9D8.3050307@redhat.com> <4EA940B1.9050905@oracle.com> <4EA951F2.70001@redhat.com> <11793E27-1CB6-4DD4-8CE4-3E093859A870@oracle.com> <4EA956C7.6070700@redhat.com> <2DB482BE-B822-4FEF-80CC-2BF05FF4B9DB@oracle.com> <4EA9E469.2070701@oracle.com>
In-Reply-To: <4EA9E469.2070701@oracle.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit
X-IsSubscribed: yes
Mailing-List: contact gcc-patches-help@gcc.gnu.org; run by ezmlm
Precedence: bulk
List-Id: <gcc-patches.gcc.gnu.org>
List-Archive: <http://gcc.gnu.org/ml/gcc-patches/>
List-Post: <mailto:gcc-patches@gcc.gnu.org>
List-Help: <mailto:gcc-patches-help@gcc.gnu.org>
Sender: gcc-patches-owner@gcc.gnu.org
X-SW-Source: 2011-10/txt/msg02589.txt.bz2

.. thus now I'm investigating, comparing 4.5 to mainline, where the 
former gets the TYPE_BINFO of the TREE_OPERAND (member, 0). And the 
answer is "simple": at the beginning of tsubst_copy_and_build case 
COMPONENT_REF itself! Where:

     member = TREE_OPERAND (t, 1);
     if (BASELINK_P (member))
       member = tsubst_baselink (member,
                     non_reference (TREE_TYPE (object)),
                     args, complain, in_decl);
     else
       member = tsubst_copy (member, args, complain, in_decl);

the tsubst_copy call changes very little member in mainline. From this:

<scope_ref 0x7ffff563db70 tree_0 tree_1
     arg 0 <record_type 0x7ffff577f540 impl type_0 type_5 type_6 VOID
         align 8 symtab 0 alias set -1 canonical type 0x7ffff577f540 
context <translation_unit_decl 0x7ffff5646170 D.1>
         full-name "struct impl<V>"
         no-binfo use_template=1 interface-unknown
         chain <type_decl 0x7ffff5780170 impl>>
     arg 1 <template_id_expr 0x7ffff563dae0
         arg 0 <identifier_node 0x7ffff577e790 create
             bindings <(nil)>
             local bindings <(nil)>>
         arg 1 <tree_vec 0x7ffff5769ed8 elt 0 <template_type_parm 
0x7ffff577f3f0 U>>>>

to this:

<scope_ref 0x7ffff563dd20 tree_0 tree_1
     arg 0 <record_type 0x7ffff577f9d8 impl type_0 type_5 type_6 VOID
         align 8 symtab 0 alias set -1 canonical type 0x7ffff577f9d8 
context <translation_unit_decl 0x7ffff5646170 D.1>
         full-name "struct impl<V>"
         no-binfo use_template=1 interface-unknown
         chain <type_decl 0x7ffff57802e0 impl>>
     arg 1 <template_id_expr 0x7ffff563dcf0
         type <lang_type 0x7ffff5765888 unknown type type <lang_type 
0x7ffff5765888 unknown type>
             VOID
             align 1 symtab 0 alias set -1 canonical type 0x7ffff5765888
             pointer_to_this <lang_type 0x7ffff5765888 unknown type> 
reference_to_this <lang_type 0x7ffff5765888 unknown type>>

         arg 0 <identifier_node 0x7ffff577e790 create
             bindings <(nil)>
             local bindings <(nil)>>
         arg 1 <tree_vec 0x7ffff5783140 elt 0 <template_type_parm 
0x7ffff577f7e0 U>>>>

note: no changes for arg0, no-binfo.

Instead, in the 4.5 case, from:

<scope_ref 0x7ffff7f8d188 tree_0
     arg 0 <record_type 0x7ffff5a15348 impl type_0 type_5 type_6 VOID
         align 8 symtab 0 alias set -1 canonical type 0x7ffff5a15348
         full-name "struct impl<V>"
         no-binfo use_template=1 interface-unknown
         chain <type_decl 0x7ffff5a01c38 impl>>
     arg 1 <template_id_expr 0x7ffff7f8d0e0
         arg 0 <identifier_node 0x7ffff5a11840 create
         bindings <(nil)>
         local bindings <(nil)>>
         arg 1 <tree_vec 0x7ffff5a164d8
             elt 0 <template_type_parm 0x7ffff5a151f8 U>>>>

To:

<scope_ref 0x7ffff7f8d268 tree_0
     arg 0 <record_type 0x7ffff5a15690 impl type_5 type_6 QI
         size <integer_cst 0x7ffff7e6d730 constant 8>
         unit size <integer_cst 0x7ffff7e6d758 constant 1>
         align 8 symtab 0 alias set -1 canonical type 0x7ffff5a15690
         fields <type_decl 0x7ffff5a17170 impl type <record_type 
0x7ffff5a157e0 impl>
             external nonlocal suppress-debug decl_4 VOID file 50870.C 
line 6 col 3
             align 8 context <record_type 0x7ffff5a15690 impl> result 
<record_type 0x7ffff5a15690 impl>
 >
         full-name "struct impl<float>"
         X() X(constX&) this=(X&) n_parents=0 use_template=1 
interface-unknown
         pointer_to_this <pointer_type 0x7ffff5a15738> chain <type_decl 
0x7ffff5a170b8 impl>>
     arg 1 <template_id_expr 0x7ffff7f8d230
         type <lang_type 0x7ffff59f8bd0 unknown type type <lang_type 
0x7ffff59f8bd0 unknown type>
             VOID
             align 1 symtab 0 alias set -1 canonical type 0x7ffff59f8bd0
             pointer_to_this <lang_type 0x7ffff59f8bd0 unknown type> 
reference_to_this <lang_type 0x7ffff59f8bd0 unknown type>>

         arg 0 <identifier_node 0x7ffff5a11840 create
         bindings <(nil)>
         local bindings <(nil)>>
         arg 1 <tree_vec 0x7ffff5a16b18
             elt 0 <integer_type 0x7ffff7e82498 int>>>>

arg0 gets its fields. Now, a big difference between the two - 4.5 vs 
mainline - tsubst_copy calls, is the args argument.

In 4.5, a very simple and understandable:

<tree_vec 0x7ffff59f67c0
     elt 0 <pointer_type 0x7ffff5a15738>
     elt 1 <integer_type 0x7ffff7e82498 int>
     elt 2 <real_type 0x7ffff7e96150 float>>

in mainline:

<tree_vec 0x7ffff5758fc0
     elt 0 <template_type_parm 0x7ffff577f738 T VOID
         align 8 symtab 0 alias set -1 canonical type 0x7ffff577f738
        index 0 level 1 orig_level 1
         chain <type_decl 0x7ffff576df18 T>>
     elt 1 <template_type_parm 0x7ffff577f7e0 U VOID
         align 8 symtab 0 alias set -1 canonical type 0x7ffff577f7e0
        index 1 level 1 orig_level 1
         chain <type_decl 0x7ffff5780000 U>>
     elt 2 <template_type_parm 0x7ffff577f888 V type_0 type_6 VOID
         align 8 symtab 0 alias set -1 canonical type 0x7ffff577f888
        index 2 level 1 orig_level 1
         chain <type_decl 0x7ffff57800b8 V>>
     elt 3 <template_type_parm 0x7ffff577f930 VOID
         align 8 symtab 0 alias set -1 canonical type 0x7ffff577f930
        index 3 level 1 orig_level 1
         chain <type_decl 0x7ffff5780228 D.1853>>>

args comes directly from the tsubst_copy_and_build arguments, looks like 
we reach case COMPONENT_REF with the template arguments still 
unsubstituted, something is simply not done earlier in mainline.

I guess I had better stopping here in terms of details on the mailing 
list today. If you have some further hints...

Thanks!
Paolo.