RE: asyn driver on Windows XP

["Mark Rivers" <rivers@cars.uchicago.edu>]

Hi Jeff,

> I confess to not being aware that _not_ setting this parameter on the command line might 
> imply that the executable was restricted to only running on the build host's platform.

No, that is NOT the problem.  It's the opposite: NOT setting this parameter can lead to situations where the application will not even run on the host platform!  

That is exactly what happened to me.  My system is XP SP3 (=0x0503), but the SDK include file sets _WIN32_WINNT to 0x0600 by default.  Eric's code in asyn checks the value of POLLIN, and uses WASPoll() if it is defined.  POLLIN is defined if _WIN32_WINNT = 0x0600, so asyn uses WASPoll(), which then results in an application that won't run on my host machine that built it!

I am suggesting that we should define _WIN32_WINNT in a CONFIG_SITE file to specify the minimum host archtecture we want our application to be able to run on.

Mark



-----Original Message-----
From: Jeff Hill [mailto:johill@lanl.gov] 
Sent: Tuesday, May 31, 2011 12:18 PM
To: Mark Rivers; 'Dirk Zimoch'; 'Eric Norum'; 'EPICS'
Cc: 'Janet Anderson'
Subject: RE: asyn driver on Windows XP

Hi Mark,

Searching in the source I can see that _WIN32_WINNT is being set in certain
.c/.cpp src\libCom\osi\os\WIN32 in order to enable use of some of the post
W95 features of windows. 

 
C:\hill\epicsInBazaar\cvs-trunk\src\libCom\osi\os\WIN32\osdMutex.c(32):#defi
ne _WIN32_WINNT 0x0400 
  C:\hill\epicsInBazaar\cvs-trunk\src\libCom\osi\os\WIN32\osdThread.c(27):#
define _WIN32_WINNT 0x400 /* defining this drops support for W95 */
 
C:\hill\epicsInBazaar\cvs-trunk\src\libCom\osi\os\WIN32\osdThread.c(124):#if
_WIN32_WINNT >= 0x0501 
  C:\hill\epicsInBazaar\cvs-trunk\src\libCom\osi\os\WIN32\osdThread.c(213):
* if defined ( NTDDI_VISTA ) && _WIN32_WINNT >= NTDDI_VISTA
 
C:\hill\epicsInBazaar\cvs-trunk\src\libCom\osi\os\WIN32\setThreadName.cpp(15
):#   define _WIN32_WINNT 0x400 /* defining this drops support for W95 */

I have been assuming that setting this parameter too high would keep the
executable from running on older versions of windows where a certain feature
in the WIN32 API is lacking. I confess to not being aware that _not_ setting
this parameter on the command line might imply that the executable was
restricted to only running on the build host's platform. It seems that many
people would be complaining if that were the case? Is this really true? I
found this page, but a quick read doesn't seem to imply that the executable
is restricted to only running on the build host's platform if _WIN32_WINNT
isn't set, but maybe that isn't spelled out by MS.

http://msdn.microsoft.com/en-us/library/aa383745%28v=vs.85%29.aspx

I suppose that we could set _WIN32_WINNT to 0x400 in
CONFIG.win32-x86.win32-x86 in order to document at compile time that EPICS
base will not run on windows 95, but raising it globally above that level
might restrict deployment of EPICS base to some of the older versions of
windows that might still in use?

Perhaps the solution is to only selectively set this _WIN32_WINNT in certain
asyn source file (.c/.cpp files) that require a new feature of the windows
API, and then only these asyn dll will be restricted to certain newer
versions of windows. Does that work?

Jeff
______________________________________________________
Jeffrey O. Hill           Email        johill@lanl.gov
LANL MS H820              Voice        505 665 1831
Los Alamos NM 87545 USA   FAX          505 665 5107

Message content: TSPA

With sufficient thrust, pigs fly just fine. However, this is
not necessarily a good idea. It is hard to be sure where they
are going to land, and it could be dangerous sitting under them
as they fly overhead. -- RFC 1925


> -----Original Message-----
> From: Mark Rivers [mailto:rivers@cars.uchicago.edu]
> Sent: Tuesday, May 31, 2011 10:15 AM
> To: Dirk Zimoch; Eric Norum; EPICS
> Cc: Janet Anderson; Jeff Hill
> Subject: RE: asyn driver on Windows XP
> 
> Folks,
> 
> I think we need to revisit this problem.  The problem is that when
> building on win32-x86 there is confusion about what OS version is being
> built for.  This most often leads to an error when running applications
> built with asyn.  asyn contains the following code to determine if
> WSAPoll should be used:
> 
> # elif defined(_WIN32)
> #  if defined(POLLIN)
> #   define poll(fd,nfd,t) WSAPoll(fd,nfd,t)
> #  else
> #   define FAKE_POLL
> #  endif
> 
> POLLIN should only be defined on more recent (e.g. Windows Vista and
> Windows 7) systems, not on XP, where WSAPoll() does not exist.
> 
> Previously users who were building with Microsoft Visual C++ 2008
> Express Edition were finding that the file
> 
> C:\Program Files\Microsoft SDKs\Windows\v7.0\Include\sdkddkver.h
> 
> Contained the following:
> 
> #if !defined(_WIN32_WINNT) && !defined(_CHICAGO_)
> #define  _WIN32_WINNT   0x0601
> #endif
> 
> The definitions of _WIN32_WINNT are:
> Windows 7                                         _WIN32_WINNT_WIN7
> (0x0601)
> Windows Server 2008                               _WIN32_WINNT_WS08
> (0x0600)
> Windows Vista                                     _WIN32_WINNT_VISTA
> (0x0600)
> Windows Server 2003 with SP1, Windows XP with SP2 _WIN32_WINNT_WS03
> (0x0502)
> Windows Server 2003, Windows XP                   _WIN32_WINNT_WINXP
> (0x0501)
> Windows 2000                                      _WIN32_WINNT_WIN2K
> (0x0500)
> 
> I recently upgraded my Windows XP SP3 system from Visual Studio 2005 to
> Visual Studio 2008, and in doing that it added
> 
> C:\Program Files\Microsoft SDKs\Windows\v6.0A
> 
> In place of
> C:\Program Files\Microsoft SDKs\Windows\v5.0
> 
> 
> C:\Program Files\Microsoft SDKs\Windows\v6.0A\Include\sdkddkver.h
> contains:
> #if !defined(_WIN32_WINNT) && !defined(_CHICAGO_)
> #define  _WIN32_WINNT   0x0600
> #endif
> 
> Note that although this is a Windows XP SP3 system, _WIN32_WINNT is
> defined to be 0x0600, which is Windows Vista.
> 
> When I build and run win32-x86 applications build with asyn they link
> OK, but they give a run-time error that WSAPoll cannot be found in the
> Winsock DLL.
> 
> I worked around the problem by modifying sdkddkver.h to set the value to
> 0x0503 (XP SP3), but this is clearly not the correct solution.
> 
> I has originally blamed Microsoft for doing something stupid with VS
> Express, but I am coming to the conclusion that we are doing something
> wrong in the EPICS build system.
> 
> We are assuming that the default build with Visual Studio with the flags
> we use will build an application that can run on the host machine.  This
> is clearly incorrect.  Microsoft is letting one build an application on
> a host OS that can run on other target OS, for example I can build an
> application on Windows 7 that will run on XP, etc.
> 
> I believe we need to define _WIN32_WINNT in
> CONFIG_SITE.win32-x86.win32-x86 or a similar file to declare what
> Windows OS we are targeting with our EPICS build.  One may well want to
> set this to 0x0503 so that the executable can run on XP, even if it is
> being built on a Windows 7 system, for example.
> 
> Does this seem correct?
> 
> Janet, Jeff?
> 
> Thanks,
> Mark
> 
> -----Original Message-----
> From: Mark Rivers
> Sent: Monday, November 15, 2010 11:07 AM
> To: 'Dirk Zimoch'; EPICS
> Subject: RE: asyn driver on Windows XP
> 
> Hi Dirk,
> 
> A similar problem has been seen by other users when using Microsoft
> Visual C++ 2008 Express Edition.
> 
> The problem turned out to be apparently an error in Microsoft's header
> file.  Here is a message from Anthony Owen at the Australian Synchrotron
> with his fix.
> 
> *******************************************************************
> Hi Mark,
> 
> It turns out that the issue was with the Microsoft SDK. The SDK is
> installed as part of the Visual Studio Package.
> 
> I tried SDK 6.0V and SDK 7.0 and got the same problem. It was narrowed
> down to be the version being incorrectly detected in the file:
> 
> C:\Program Files\Microsoft SDKs\Windows\v7.0\Include\sdkddkver.h
> 
> <snip>
> 
> #if !defined(_WIN32_WINNT) && !defined(_CHICAGO_)
> #define  _WIN32_WINNT   0x0601
> #endif
> 
> <snip>
> 
> It was defining 0x0601 which appears to be Windows 7.
> With this changed to 0x0503 (Windows XP SP3) and ASYN reverted to its
> original form, POLLIN is not defined and ASYN uses FAKE_POLL.
> It compiled and executed successfully.
> 
> Hope this helps if you have to use Visual Express!
> 
> Best Regards,
> 
> Anthony Owen
> *******************************************************************
> 
> So the header file was incorrectly setting the system version to be
> 0x0601, which is Windows 7, even though it was an XP system.  That then
> caused POLLIN to be defined when it should not have been.
> 
> Is there a similar problem in Visual Studio 10?
> 
> Mark
> 
> 
> -----Original Message-----
> From: tech-talk-bounces@aps.anl.gov
> [mailto:tech-talk-bounces@aps.anl.gov] On Behalf Of Dirk Zimoch
> Sent: Monday, November 15, 2010 10:58 AM
> To: EPICS
> Subject: asyn driver on Windows XP
> 
> Hi all,
> 
> I have a problem with the latest version of asyn driver on Windows XP.
> It tries to use a function WSAPoll() that does not exist in XP, only in
> Vista and newer. However I do not want use an old version of asyn driver
> 
> without that function because of all the new features and bugfixes.
> 
> The code in question is:
> 
> #if defined(__rtems__)
> # define USE_SOCKTIMEOUT
> #else
> # define USE_POLL
> # if defined(vxWorks)
> #  define FAKE_POLL
> # elif defined(_WIN32)
> #  if defined(POLLIN)
> #   define poll(fd,nfd,t) WSAPoll(fd,nfd,t)
> #  else
> #   define FAKE_POLL
> #  endif
> # else
> #  include <sys/poll.h>
> # endif
> #endif
> 
> It seems I have POLLIN defined but no WSAPoll.
> 
> I am using the compiler from Visual Studio 10.
> 
> Any idea?
> 
> Dirk