[Qemu-devel] Recent patches for qemu block-cow.c etc breaks Win32 build becaus e of fsync()

[Qemu-devel] Recent patches for qemu block-cow.c etc breaks Win32 build becaus e of fsync()

[Armistead, Jason]

Hi list

Looking at the code that was recently added

via http://lists.gnu.org/archive/html/qemu-devel/2006-06/msg00025.html

There were a lot of fsync() calls made, but this breaks the QEMU build
process on Win32 because fsync() is not available.

Suggest something like

#ifndef _WIN32
   fsync(blah);
#else
   _commit(blah);
#endif

which seems to work OK for me under MinGW / MSYS environment.

The _commit() function seems to be what Microsoft recommend to use, and it
compiles and runs OK with that in there, so I guess that MinGW is happy.

I'm still trying to get CVS up and running properly so I can contribute
patches in the right format.

Regards

Jason

Re: [Qemu-devel] Recent patches for qemu block-cow.c etc breaks Win32 build becaus e of fsync()

[Johannes Schindelin]

Hi,

On Thu, 8 Jun 2006, Armistead, Jason wrote:

> #ifndef _WIN32
>    fsync(blah);
> #else
>    _commit(blah);
> #endif

Why not define it in config-host.mak like this:

--- configure.old	2006-06-08 23:06:48.000000000 +0200
+++ configure	2006-06-08 23:07:10.000000000 +0200
@@ -102,6 +102,7 @@
 ;;
 MINGW32*)
 mingw32="yes"
+CFLAGS="-Dfsync=_commit"
 ;;
 FreeBSD)
 bsd="yes"


Ciao,
Dscho

Re: [Qemu-devel] Recent patches for qemu block-cow.c etc breaks Win32 build becaus e of fsync()

[Christian MICHON]

neat and concise!
bravos & kudos :)

On 6/8/06, Johannes Schindelin <Johannes.Schindelin@gmx.de> wrote:
> Why not define it in config-host.mak like this:
> +CFLAGS="-Dfsync=_commit"

-- 
Christian

Re: [Qemu-devel] Recent patches for qemu block-cow.c etc breaks Win32 build becaus e of fsync()

[Paul Brook]

> Why not define it in config-host.mak like this:
>
> --- configure.old	2006-06-08 23:06:48.000000000 +0200
> +++ configure	2006-06-08 23:07:10.000000000 +0200
> @@ -102,6 +102,7 @@
>  ;;
>  MINGW32*)
>  mingw32="yes"
> +CFLAGS="-Dfsync=_commit"
>  ;;
>  FreeBSD)
>  bsd="yes"

No. This is what header files are for.

Paul

Re: [Qemu-devel] Recent patches for qemu block-cow.c etc breaks Win32 build becaus e of fsync()

[Johannes Schindelin]

Hi,

On Thu, 8 Jun 2006, Paul Brook wrote:

> > Why not define it in config-host.mak like this:
> >
> > --- configure.old	2006-06-08 23:06:48.000000000 +0200
> > +++ configure	2006-06-08 23:07:10.000000000 +0200
> > @@ -102,6 +102,7 @@
> >  ;;
> >  MINGW32*)
> >  mingw32="yes"
> > +CFLAGS="-Dfsync=_commit"
> >  ;;
> >  FreeBSD)
> >  bsd="yes"
> 
> No. This is what header files are for.

Yes, but MS did not provide fsync in header files ;-)

Ciao,
Dscho

Re: [Qemu-devel] Recent patches for qemu block-cow.c etc breaks Win32 build becaus e of fsync()

[Paul Brook]

> > > Why not define it in config-host.mak like this:
> > >
> > > --- configure.old	2006-06-08 23:06:48.000000000 +0200
> > > +++ configure	2006-06-08 23:07:10.000000000 +0200
> > > @@ -102,6 +102,7 @@
> > >  ;;
> > >  MINGW32*)
> > >  mingw32="yes"
> > > +CFLAGS="-Dfsync=_commit"
> > >  ;;
> > >  FreeBSD)
> > >  bsd="yes"
> >
> > No. This is what header files are for.
>
> Yes, but MS did not provide fsync in header files ;-)

I mean you should add a #define in a header file somewhere (eg. vl.h where 
there are already similar things), not do it via a commandline option.

Paul

Re: [Qemu-devel] Recent patches for qemu block-cow.c etc breaks Win32 build becaus e of fsync()

[Johannes Schindelin]

Hi,

On Thu, 8 Jun 2006, Paul Brook wrote:

> I mean you should add a #define in a header file somewhere (eg. vl.h where 
> there are already similar things), not do it via a commandline option.

You mean something like this?

--- vl.h.old	2006-06-08 23:58:55.000000000 +0200
+++ vl.h	2006-06-08 23:59:42.000000000 +0200
@@ -47,6 +47,7 @@
 #endif
 
 #ifdef _WIN32
+#define fsync _commit
 #define lseek _lseeki64
 #define ENOTSUP 4096
 /* XXX: find 64 bit version */

However, I do not know if this breaks cygwin (if cygwin worked at all, 
never checked).

Ciao,
Dscho