use of tet_fork (hard part)

[Geoff Clare]

I wrote, on 07 Apr 2006:
>
> I don't understand why Jamey's change didn't work either, but I
> have tried an alternative fix which still passes for me.  The basic
> idea is to move the XCheckWindowEvent() on client2 into the child.
> The actual changes are more complicated because of the need to
> rearrange the path checking.  I also changed the opendisplay() to
> a direct XOpenDisplay() to avoid the same problem that Jamey fixed
> in grbsrvr.m and ungrbsrvr.m.

It turns out that this patch causes a problem in later "last close"
tests because it leaves client2 open in the parent.

This is easy to fix for POSIX systems - just close the file descriptor
ConnectionNumber(client2) - but that won't work for non-POSIX systems.
The patch below just calls XCloseDisplay(client2) on non-POSIX
systems, which means the original problem will still exist on
non-POSIX systems but hopefully the chances of the tests being run on
a non-POSIX system which has the problem Jamey originally reported are
small.  (I'm not even sure if the test suite is still usable on
non-POSIX systems, although it was originally designed to be.)

Note that this is an incremental patch - it needs to be applied after
the first patch.

--
Geoff Clare <g.clare at opengroup.org>
The Open Group, Thames Tower, Station Road, Reading, RG1 1LX, England

--- clsdsply.m.patch1	2006-04-07 11:59:00.000000000 +0100
+++ clsdsply.m.new	2006-04-24 10:02:29.000000000 +0100
@@ -494,6 +494,20 @@
 	if (t006p_pass == 1)
 		CHECK;
 
+	/*
+	 * We need to close client2 otherwise it interferes with
+	 * later "last close" tests.  On POSIX systems we can
+	 * directly close the underlying fd.  On other systems
+	 * we use XCloseDisplay() and hope that the fact it was
+	 * used in the child doesn't cause a problem.
+	 */
+	if(config.posix_system == 1) {
+		close(ConnectionNumber(client2));
+	}
+	else {
+		XCloseDisplay(client2);
+	}
+
 	CHECKPASS(2);
 
 >>ASSERTION Good B 1