conflicting gettimeofday with MinGW

On Wed, Apr 11, 2007 at 01:45:28PM +0200, Zeugswetter Andreas ADI SD wrote:

mingw-runtime-3.10 introduced a gettimeofday declaration in sys/time.h
that is not compatible with port.h.
(current is mingw-runtime-3.12)

int __cdecl gettimeofday(struct timeval *__restrict__,
void *__restrict__ /* tzp (unused) */);

The problem was already reported by Marc Lepage on June 26, 2006 shortly
after the mingw-runtime release,
(He reported back that it was a MinGW installation issue. But it is a
version issue.).

Now, we could probably make a case that MinGW needs to use "struct
timezone" instead of void,
but that change alone still does not make pg compile, because of the
"#define TIMEZONE_GLOBAL timezone"
in port.h.

Any opinions on how to proceed ?

There doesn't happen to be a #define you can set to make mingw backwards
compatible with itself and not define that header?

We don't use gettimeofday() from them anyway, we use the one in
libpgport...

//Magnus

Zeugswetter Andreas ADI SD wrote:

mingw-runtime-3.10 introduced a gettimeofday declaration in sys/time.h
that is not compatible with port.h.
(current is mingw-runtime-3.12)

int __cdecl gettimeofday(struct timeval *__restrict__,
void *__restrict__ /* tzp (unused) */);

The problem was already reported by Marc Lepage on June 26, 2006 shortly
after the mingw-runtime release,
(He reported back that it was a MinGW installation issue. But it is a
version issue.).

Now, we could probably make a case that MinGW needs to use "struct
timezone" instead of void,
but that change alone still does not make pg compile, because of the
"#define TIMEZONE_GLOBAL timezone"
in port.h.

Any opinions on how to proceed ?

I don't see 3.10 on the download page at http://www.mingw.org/download.shtml

The current release is apparently 3.9.

cheers

andrew

That page is ages out of date. The intended sync is apparently broken.
The current download area is on sourceforge
http://sf.net/project/showfiles.php?group_id=2435

Andreas

Zeugswetter Andreas ADI SD wrote:

That page is ages out of date. The intended sync is apparently broken.
The current download area is on sourceforge
http://sf.net/project/showfiles.php?group_id=2435

*sigh*

And what is in 3.12, which is apparently the current version?

cheers

andrew

That page is ages out of date. The intended sync is

apparently broken.

The current download area is on sourceforge
http://sf.net/project/showfiles.php?group_id=2435

*sigh*

And what is in 3.12, which is apparently the current version?

Sorry that was implied. sys/time.h did not change between 3.10 and 3.12.
There is no #define to remove the declaration.

In cvs the file is marked as dead and gives a reference to cygwin's
w32api.
Maybe the answer is to simply note that the mingw-runtime binary from
sourceforge above 3.9 has a broken sys/time.h (it is not the file from
cygwin but the dead one from cvs).

Andreas

On Wed, Apr 11, 2007 at 06:17:19PM +0200, Zeugswetter Andreas ADI SD wrote:

That page is ages out of date. The intended sync is

apparently broken.

The current download area is on sourceforge
http://sf.net/project/showfiles.php?group_id=2435

*sigh*

And what is in 3.12, which is apparently the current version?

Sorry that was implied. sys/time.h did not change between 3.10 and 3.12.
There is no #define to remove the declaration.

In cvs the file is marked as dead and gives a reference to cygwin's
w32api.
Maybe the answer is to simply note that the mingw-runtime binary from
sourceforge above 3.9 has a broken sys/time.h (it is not the file from
cygwin but the dead one from cvs).

I think that soudns reasonable. Something for the mingw FAQ. Care to
provide bruce with a patch?

//Magnus

Magnus Hagander wrote:

On Wed, Apr 11, 2007 at 06:17:19PM +0200, Zeugswetter Andreas ADI SD wrote:

That page is ages out of date. The intended sync is

apparently broken.

The current download area is on sourceforge
http://sf.net/project/showfiles.php?group_id=2435

*sigh*

And what is in 3.12, which is apparently the current version?

Sorry that was implied. sys/time.h did not change between 3.10 and 3.12.
There is no #define to remove the declaration.

In cvs the file is marked as dead and gives a reference to cygwin's
w32api.
Maybe the answer is to simply note that the mingw-runtime binary from
sourceforge above 3.9 has a broken sys/time.h (it is not the file from
cygwin but the dead one from cvs).

I think that soudns reasonable. Something for the mingw FAQ. Care to
provide bruce with a patch?

No, that means that we can't build with currently released Mingw, AIUI.
I think the solution is probably to rename our replacement gettimeofday
and add a #define to translate the calls. That's pretty much what we've
done in other cases, and has seemed to be a more robust strategy than
supplying replacements for standard library routines with routines of
the same name.

cheers

andrew

Andrew Dunstan wrote:

In cvs the file is marked as dead and gives a reference to cygwin's
w32api.
Maybe the answer is to simply note that the mingw-runtime binary from
sourceforge above 3.9 has a broken sys/time.h (it is not the file from
cygwin but the dead one from cvs).

I think that soudns reasonable. Something for the mingw FAQ. Care to
provide bruce with a patch?

No, that means that we can't build with currently released Mingw, AIUI.
I think the solution is probably to rename our replacement gettimeofday
and add a #define to translate the calls. That's pretty much what we've
done in other cases, and has seemed to be a more robust strategy than
supplying replacements for standard library routines with routines of
the same name.

But if I read you right, they're removing it in the next version? That
would indicate that they included it by mistake?

(But sure, in general that's not a bad idea)

//Magnus