I'm attempting to make libvirt capable of building with mingw64 toolchain's
pthread.h, instead of directly using Windows thread primitives.
As a quick hack I just changed libvirt logic to force use of pthreads,
since configure is already doing the neccessary pthread.h probes for us
even on win32 eg
diff --git a/src/util/virthread.c b/src/util/virthread.c
index dd1768e..8849d7d 100644
--- a/src/util/virthread.c
+++ b/src/util/virthread.c
@@ -25,7 +25,7 @@
/* On mingw, we prefer native threading over the sometimes-broken
* pthreads-win32 library wrapper. */
-#ifdef WIN32
+#ifdef WINx32
# include "virthreadwin32.c"
#elif defined HAVE_PTHREAD_MUTEXATTR_INIT
# include "virthreadpthread.c"
diff --git a/src/util/virthread.h b/src/util/virthread.h
index 84d3bdc..649285e 100644
--- a/src/util/virthread.h
+++ b/src/util/virthread.h
@@ -111,7 +111,7 @@ int virThreadLocalInit(virThreadLocalPtr l,
void *virThreadLocalGet(virThreadLocalPtr l);
int virThreadLocalSet(virThreadLocalPtr l, void*) ATTRIBUTE_RETURN_CHECK;
-# ifdef WIN32
+# ifdef WINx32
# include "virthreadwin32.h"
# elif defined HAVE_PTHREAD_MUTEXATTR_INIT
# include "virthreadpthread.h"
When attempting to build though I see failures
CC util/libvirt_util_la-virerror.lo
In file included from /usr/i686-w64-mingw32/sys-root/mingw/include/sys/time.h:10:0,
from ../gnulib/lib/sys/time.h:39,
from ../gnulib/lib/sys/select.h:117,
from util/virutil.h:31,
from util/virerror.c:35:
../gnulib/lib/time.h:468:21: error: expected identifier or '(' before '{' token
_GL_FUNCDECL_SYS (localtime_r, struct tm *, (time_t const *restrict __timer,
^
In file included from /usr/i686-w64-mingw32/sys-root/mingw/include/sys/time.h:10:0,
from ../gnulib/lib/sys/time.h:39,
from ../gnulib/lib/sys/select.h:117,
from util/virutil.h:31,
from util/virerror.c:35:
../gnulib/lib/time.h:490:21: error: expected identifier or '(' before '{' token
_GL_FUNCDECL_SYS (gmtime_r, struct tm *, (time_t const *restrict __timer,
^
The problem appears to be that mingw64's pthread.h has the following
craziness:
/* Recursive API emulation. */
#undef localtime_r
#define localtime_r(_Time, _Tm) ({ struct tm *___tmp_tm; \
pthread_testcancel(); \
___tmp_tm = localtime((_Time));\
if (___tmp_tm) { \
*(_Tm) = *___tmp_tm; \
___tmp_tm = (_Tm); \
} \
___tmp_tm; })
#undef gmtime_r
#define gmtime_r(_Time,_Tm) ({ struct tm *___tmp_tm; \
pthread_testcancel(); \
___tmp_tm = gmtime((_Time)); \
if (___tmp_tm) { \
*(_Tm) = *___tmp_tm; \
___tmp_tm = (_Tm); \
} \
___tmp_tm; })
which clashes with the gnulib provided declarations for gmtime_r/localtime_r
gnulib appears to be trying to workaround this problem already - the
generated gnulib/lib/time.h file shows this though, which indicates
its workaround hasn't been activated by 'configure':
/* Some systems don't define struct timespec (e.g., AIX 4.1, Ultrix 4.3).
Or they define it with the wrong member names or define it in <sys/time.h>
(e.g., FreeBSD circa 1997). Stock Mingw does not define it, but the
pthreads-win32 library defines it in <pthread.h>. */
# if ! 1
# if 0
# include <sys/time.h>
# elif 0
# include <pthread.h>
/* The pthreads-win32 <pthread.h> also defines a couple of broken macros. */
# undef asctime_r
# undef ctime_r
# undef gmtime_r
# undef localtime_r
# undef rand_r
# undef strtok_r
# else
ie I would have expected that 'elif' to be 1 instead of 0.
It seems the logic in gnulib/m4/time_h.m4 is incorrect. The configure
check shows
checking for struct timespec in <time.h>... yes
and so time_h.m4 never gets as far as trying to detect the broken
pthreads.h - its hidden in an unreachable else block.
Regards,
Daniel
--
|: http://berrange.com -o- http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org -o- http://virt-manager.org :|
|: http://autobuild.org -o- http://search.cpan.org/~danberr/ :|
|: http://entangle-photo.org -o- http://live.gnome.org/gtk-vnc :|
--
libvir-list mailing list
libvir-list@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/libvir-list