Re: [xslt] win32 build (mingw-cross compilation)

[Daniel Veillard <veillard redhat com>]

On Fri, Aug 24, 2007 at 12:57:19AM +0300, Roumen Petrov wrote:
> Hello list members,
> 
> Before more than one month I post a patch (bug#454392 libxslt bug track 
> system).
> Daniel is not convinced that patch is correct and think that patch will 
> affect other builds.

  yes definitely. I looks to me that the patch was done with mingw-cross
compilation in mind without taking care to not affects other builds. Since
there is a huge majority of people using and building natively on Windows
and so far you're the only one I have heard asking for cross-compilation
support, you will understand that I ask you to be extremely careful in your
patch. I don't want to break thousands of people build just to fix your
problem. This is so obvious I should have to say this but somehow it seems
I need to !

> I attach patch "libxslt-1.1.20.patch-0" as is.
> Notes:
> 
> - Without changes in makefiles mingw build don't create shared 
> libraries(dlls):
>  WIN32_EXTRA_LDFLAGS += -no-undefined : for details see libtool 
>  documentation.
>    configure set this flag but don't use in makefiles.
> - Changes in ./configure.in:
>  remove unused winsock. AC_DEFINE([_WINSOCKAPI_]... seems to me as cut and 
>  paste from libxml.
>  WIN32_EXTRA_LDFLAGS += -avoid-version (new - only for mingw): create dll 
>  with name libXXX.dll instead of
>    libXXX[versufix].dll, where [versufix] is in format "-N" where N is a 
>    number computed from version info.
>   As result created by cross compilation libraries match names from native 
>   builds.
> - Modification in extensions.c is to suppress warning.
> - Now I'm sure that AC_LIBTOOL_WIN32_DLL is not necessary.
> 
> 
> I attach a new more general patch ("libxslt-1.1.20.patch").
> This patch remove completely winsock from libxslt and affect now native 
> builds.
> Note about changes in xsltutils.c:
> - In mingw HAVE_GETTIMEOFDAY should be defined.
>  After patch mingw compile fine in case if XSLT_WIN32_PERFORMANCE_COUNTER 
>  is defined too.
> - for winsock2.h library is ws2_32 but for wsock32 - winsock.h (included be 
> default from windows.h if winsock2.h is not included before I think
> ).
> Please comment on it (I don't have native build environment).
[...]
>  
> -/* gettimeofday on Windows ??? */
> -#if defined(WIN32) && !defined(__CYGWIN__)
> +#ifndef HAVE_GETTIMEOFDAY
>  #ifdef _MSC_VER
> -#include <winsock2.h>
> -#pragma comment(lib, "ws2_32.lib")
> -#define gettimeofday(p1,p2)
> -#define HAVE_GETTIMEOFDAY
> +#include <windows.h>
>  #define XSLT_WIN32_PERFORMANCE_COUNTER
>  #endif /* _MS_VER */
> -#endif /* WIN32 */
> +#endif /* HAVE_GETTIMEOFDAY*/
[...]
>  #if defined(WIN32) && !defined (__CYGWIN__)
> -#if defined(_MSC_VER) || defined(__MINGW32__)
> -#include <winsock2.h>
> +#if defined(_MSC_VER)
>  #define gettimeofday(p1,p2)
>  #define snprintf _snprintf
>  #endif /* _MS_VER */

So you change the code if WIN32 and _MSC_VER is defined, removing winsock2.h.
or the gettimeofday support for such environment. Why ?

  I don't understand how you can suggest making change to non-MINGW build 
and related code when you seems to not be able to test those !
  Please make your patches so that they affect *only* the environment
you can test, that would be a first good point to increase the value of
your patch.

Daniel

-- 
Red Hat Virtualization group http://redhat.com/virtualization/
Daniel Veillard      | virtualization library  http://libvirt.org/
veillard redhat com  | libxml GNOME XML XSLT toolkit  http://xmlsoft.org/
http://veillard.com/ | Rpmfind RPM search engine  http://rpmfind.net/