[MSVC 7.1] Bad code generation of the std::moneypunct ctor (and possibly of the std::messages ctor)

• Key: STDCXX-554
• Type: Bug
• Status: Closed
• Resolution: Fixed
• Priority: Major
• Assignee: Farid Zaripov
• Reporter: Farid Zaripov
• Component/s: 22. Localization
• Affects Version/s: 4.1.3
• Fix Version/s: 4.2.0

Environment

MSVC 7.1 with Service Pack 1

Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 13.10.6030 for 80x86
Copyright (C) Microsoft Corporation 1984-2002. All rights reserved.

Description

The 22.locale.money.put.cpp test fails on MSVC 7.1 (15s build type) with buffer overrun error due to bad code generation.

Here the assembly code for moneypunct ctor:
-------------
_EXPLICIT moneypunct (_RWSTD_SIZE_T __refs = 0)
: RW::rw_facet (_refs), money_base () { }
004018C0 push ebp
004018C1 mov ebp,esp
004018C3 push ecx
004018C4 mov dword ptr [ebp-4],ecx
004018C7 mov eax,dword ptr [__refs]
004018CA push eax
004018CB mov ecx,dword ptr [this]
004018CE call _rw::rw_facet::_rw_facet (412E20h)

004018D3 xor ecx,ecx
004018D5 mov edx,dword ptr [this]
004018D8 add edx,38h // the sizeof (moneypunct) == 0x38
004018DB mov byte ptr [edx],cl // here the place of the buffer overrun

004018DD mov eax,dword ptr [this]
004018E0 mov dword ptr [eax],offset std::moneypunct<char,0>::`vftable' (488838h)
004018E6 mov eax,dword ptr [this]
004018E9 mov esp,ebp
004018EB pop ebp
004018EC ret 4
-------------

When I commented the money_base () call the test succeeded and assembly code has changed to:
-------------
_EXPLICIT moneypunct (_RWSTD_SIZE_T __refs = 0)
: RW::rw_facet (_refs)/, money_base ()/ { }
004018C0 push ebp
004018C1 mov ebp,esp
004018C3 push ecx
004018C4 mov dword ptr [ebp-4],ecx
004018C7 mov eax,dword ptr [__refs]
004018CA push eax
004018CB mov ecx,dword ptr [this]
004018CE call _rw::rw_facet::_rw_facet (412E20h)
004018D3 mov ecx,dword ptr [this]
004018D6 mov dword ptr [ecx],offset std::moneypunct<char,0>::`vftable' (488838h)
004018DC mov eax,dword ptr [this]
004018DF mov esp,ebp
004018E1 pop ebp
004018E2 ret 4
-------------

Here the same assembly, but in 12s configuration:

before change:
-------------
const PunctT pun;
004018B1 push 1
004018B3 lea ecx,[esp+0B4h]
004018BA call _rw::rw_facet::_rw_facet (40A770h)

004018BF mov byte ptr [esp+0E8h],bl // 0xE8 - 0xB4 == 0x34, so here not buffer overrun,
// but maybe changed last 4-byte member of the __rw_facet
// (I suppose is _C_pid)

004018C6 mov dword ptr [esp+0B0h],offset Punct<char,0>::`vftable' (43A258h)
-------------

after change:
-------------
const PunctT pun;
00401891 push 1
00401893 lea ecx,[esp+0B4h]
0040189A call _rw::rw_facet::_rw_facet (40A720h)
0040189F mov dword ptr [esp+0B0h],offset Punct<char,0>::`vftable' (43A258h)
-------------

I have not verified, but I suppose that the same problem might be with messages class.

Comments

[Farid Zaripov]

The reference thread in development mailing-list: http://www.mail-archive.com/[e-mail]/msg04931.html

[Farid Zaripov]

Fixed thus: http://svn.apache.org/viewvc?rev=575613&view=rev

[Farid Zaripov]

The regression test added thus: http://svn.apache.org/viewvc?rev=575614&view=rev