Re: [Qemu-devel] [PATCH] qga: fix build for older versions of mingw

[Michael Roth <mdroth@linux.vnet.ibm.com>]

Quoting Denis V. Lunev (2015-07-07 15:04:20)
> On 07/07/15 23:02, Michael Roth wrote:
> > Quoting Peter Maydell (2015-07-07 14:16:29)
> >> On 7 July 2015 at 19:06, Denis V. Lunev <den-lists@parallels.com> wrote:
> >>> On 07/07/15 20:13, Olga Krishtal wrote:
> >>>>
> >>>> Peter, can you try this patch on your system?
> >>>>
> >>>> In mingw older than mingw64-headers-4.0.2-1.el7.noarch
> >>>> header ntddscsi.h can be found in include/ddk, however
> >>>> compiler does not look there. This breaks the compilation.
> >>>>
> >>>> The patch adds this directory in QEMU_INCLUDES.
> >>>>
> >>>> Signed-off-by: Olga Krishtal <okrishtal@parallels.com>
> >>>> ---
> >>>>    configure | 9 ++++++++-
> >>>>    1 file changed, 8 insertions(+), 1 deletion(-)
> >>>>
> >>>> diff --git a/configure b/configure
> >>>> index 52cfe18..ada8895 100755
> >>>> --- a/configure
> >>>> +++ b/configure
> >>>> @@ -3763,7 +3763,6 @@ if test "$usb_redir" != "no" ; then
> >>>>            usb_redir="no"
> >>>>        fi
> >>>>    fi
> >>>> -
> >>>>    ##########################################
> >>>>    # check if we have VSS SDK headers for win
> >>>>
> >>>> @@ -3820,6 +3819,14 @@ if test "$mingw32" = "yes" -a "$guest_agent" !=
> >>>> "no" -a "$guest_agent_with_vss"
> >>>>    fi
> >>>>
> >>>>    ##########################################
> >>>> +# mingw package on Linux is quite different with different versions.
> >>>> ntddscsi.h
> >>>> +# can be found sys-root/mingw/include and in sys-root/mingw/include/ddk
> >>>> +# Add ddk directory to lookup path automatically
> >>>> +if test "$mingw32" = "yes" -a "$guest_agent" != "no"; then
> >>>> +  QEMU_INCLUDES="$QEMU_INCLUDES -I`$cc -print-sysroot`/mingw/include/ddk"
> >>>> +fi
> >>>> +
> >>>> +##########################################
> >>>>    # Guest agent Window MSI  package
> >>>>
> >>>>    if test "$guest_agent" != yes; then
> >>>>
> >>>
> >>> actually this is a dirty hack, but we want to
> >>> know whether additional include directory help or not.
> >>> If this helps, we could think how to avoid extra
> >>> configuration option and ifdef in the code.
> >>
> >> I agree it's a pretty nasty hack :-)
> >>
> >> Unfortunately my build machine is going to be running
> >> flat out processing all the last-minute pull requests.
> >> I won't be able to test this until tomorrow at best.
> >
> > I've gotten an ubuntu 14.04 mingw environment set up that I think matches yours
> > fairly closely, so I can give any proposed fixes a spin.
> >
> > Unfortunately simply pointing to ddk/ntddscsi.h instead of ntddscsi.h
> > doesn't quite do it:
> >
> > In file included from /usr/lib/gcc/i586-mingw32msvc/4.2.1-sjlj/../../../../i586-mingw32msvc/include/ddk/ntddk.h:76,
> >                   from /usr/lib/gcc/i586-mingw32msvc/4.2.1-sjlj/../../../../i586-mingw32msvc/include/ddk/ntddscsi.h:34,
> >                   from /home/mdroth/w/qemu4.git/qga/commands-win32.c:25:
> > /usr/lib/gcc/i586-mingw32msvc/4.2.1-sjlj/../../../../i586-mingw32msvc/include/ddk/winddk.h:1069: error: redefinition of ‘struct _DRIVE_LAYOUT_INFORMATION_MBR’
> > /usr/lib/gcc/i586-mingw32msvc/4.2.1-sjlj/../../../../i586-mingw32msvc/include/ddk/winddk.h:1073: error: redefinition of ‘struct _DRIVE_LAYOUT_INFORMATION_GPT’
> > /usr/lib/gcc/i586-mingw32msvc/4.2.1-sjlj/../../../../i586-mingw32msvc/include/ddk/winddk.h:1080: error: redefinition of ‘struct _PARTITION_INFORMATION_MBR’
> > In file included from /usr/lib/gcc/i586-mingw32msvc/4.2.1-sjlj/../../../../i586-mingw32msvc/include/ddk/ntddk.h:76,
> >                   from /usr/lib/gcc/i586-mingw32msvc/4.2.1-sjlj/../../../../i586-mingw32msvc/include/ddk/ntddscsi.h:34,
> >                   from /home/mdroth/w/qemu4.git/qga/commands-win32.c:25:
> > /usr/lib/gcc/i586-mingw32msvc/4.2.1-sjlj/../../../../i586-mingw32msvc/include/ddk/winddk.h:3575: error: nested redefinition of ‘enum _PARTITION_STYLE’
> > /usr/lib/gcc/i586-mingw32msvc/4.2.1-sjlj/../../../../i586-mingw32msvc/include/ddk/winddk.h:3575: error: redeclaration of ‘enum _PARTITION_STYLE’
> > /usr/lib/gcc/i586-mingw32msvc/4.2.1-sjlj/../../../../i586-mingw32msvc/include/ddk/winddk.h:3576: error: redeclaration of enumerator ‘PARTITION_STYLE_MBR’
> > /usr/lib/gcc/i586-mingw32msvc/4.2.1-sjlj/../../../../i586-mingw32msvc/include/winioctl.h:208: error: previous definition of ‘PARTITION_STYLE_MBR’ was here
> > /usr/lib/gcc/i586-mingw32msvc/4.2.1-sjlj/../../../../i586-mingw32msvc/include/ddk/winddk.h:3578: error: redeclaration of enumerator ‘PARTITION_STYLE_GPT’
> > /usr/lib/gcc/i586-mingw32msvc/4.2.1-sjlj/../../../../i586-mingw32msvc/include/winioctl.h:209: error: previous definition of ‘PARTITION_STYLE_GPT’ was here
> > ...
> >
> > I'm still trying to make sense of how the headers are organized but the
> > impression I'm getting is that stuff in ddk/ is meant for kernel code and
> > doesn't play very nicely with userspace headers. At some point ntddscsi.h
> > got moved out of ddk/ and #ifdef guards were added all over the place to
> > make it useable, so it seems likely we'll have to disable the feature
> > completely for the ddk/ntddscsi.h versions of mingw.
> >
> >>
> >> -- PMM
> >>
> >
> 
> 
> yep :( this seems so thus we need to disable it if the header is not 
> available

Hmm, well it seems it's *almost* possible. I can build with the
following changes (ddk/ntddstor.h replaces winioctl.h):

diff --git a/qga/commands-win32.c b/qga/commands-win32.c
index 609f3c7..6ebb531 100644
--- a/qga/commands-win32.c
+++ b/qga/commands-win32.c
@@ -21,8 +21,11 @@
 #include <ws2ipdef.h>
 #include <iptypes.h>
 #include <iphlpapi.h>
-#include <winioctl.h>
-#include <ntddscsi.h>
+#include <ddk/ntddk.h>
+#include <ddk/ntddscsi.h>
+#include <ddk/ntddstor.h>
 #include <setupapi.h>
 #include <initguid.h>
 #include "qga/guest-agent-core.h"

But I have to modify ddk/winddk.h (pulled in by ddk/ntddk.h) due to an
unfortunately-named 'PCIBus' enum that conflicts with QEMU defs. I
don't think there's a reasonable way to work around this so we'll
have to stick to the original plan.

Since I have a build environment handy I'll go ahead and put together
a patch to disable the feature via a configure check/flag.

>