Farid Zaripov wrote:
-----Original Message-----
From: Martin Sebor [mailto:[EMAIL PROTECTED]
Sent: Friday, June 27, 2008 9:39 PM
To: [email protected]
Subject: Re: svn commit: r671547 -
/stdcxx/branches/4.2.x/include/rw/_mutex.h
[EMAIL PROTECTED] wrote:
Author: faridz
Date: Wed Jun 25 06:28:27 2008
New Revision: 671547
URL: http://svn.apache.org/viewvc?rev=671547&view=rev
Log:
2008-06-25 Farid Zaripov <[EMAIL PROTECTED]>
* include/rw/_mutex.h: #define WIN32_LEAN_AND_MEAN
before #including <windows.h>
to prevent #including unneeded headers, especially
<winsock.h>, that causes an
"WinSock.h has already been included" errors when
<winsock2.h> has been #included
after <rw/_mutex.h>.
I don't think we want to do this. Users might want to
#include the fat and flabby <windows.h> after they #include
one of our headers.
But including <windows.h> in <rw/_mutex.h> without WIN32_LEAN_AND_MEAN
defined
causes an errors when user doesn't includes <windows.h>, but includes
<winsock2.h> after
our library header, that including <rw/_mutex.h> (i.e. <istream>).
Due to this issue the most of boost/asio regression tests are failed
to compile on 64-bit MSVC:
We had problems caused by <windows.h> in the past. IIRC,
we solved them by forward declaring some of the functions
that we needed. Is this not being done anymore? Or if we
cannot do this for some reason, is there some saner
header than <windows.h> that we could #include instead?
If not, we might need to insulate rw/_mutex.h from all
this windows garbage by defining all the functions we
need out of line and #including <windows.h> in some
.cpp file.
Martin
http://beta.boost.org/development/tests/trunk/developer/output/msvc-9-0-
x64-stdcxx-4-2-2-boost-bin-v2-libs-asio-test-basic_datagram_socket-test-
msvc-9-0_stdcxx_422-debug-address-model-64-architecture-x86-stdlib-stdcx
x-4-2-2-threading-multi.html
If #undef-ining WIN32_LEAN_AND_MEAN after
#including <windows.h> here users can #include <windows.h>
again and get the rest of the junk defined there that would
be a possible solution.
Unfortunately this doesn't works.
Otherwise we need to revert.
Ok, I'll revert this change along with deleting of the #define
_RWSTD_NO_FWD_DECLARATIONS
in <rw/_config-msvcrt.h>.
Farid.
