On Fri, Mar 4, 2011 at 6:12 AM, Harris, Donald <[email protected]> wrote:
> Good morning,
>
> Ok, so I downloaded your latest source code, and when I run do-build, it
> fails with this:
[...]
> wimpiggy/lowlevel/bindings.c:604: warning: parameter names (without types) in
> function declaration
> wimpiggy/lowlevel/bindings.c:606: error: expected â=â, â,â, â;â, âasmâ or
> â__attribute__â before â__Pyx_PyInt_from_py_CARD32â
Ugh, sorry to hear that. I have no idea why Cython would be generating
invalid C code :-(. I suppose upgrading cython might help, but I don't
know.
> Now, as for the xpra log... I attached it to the email. Looking through it I
> didn't see anything that jumped out at me, but what do I know? :)
>
> I did notice a couple of things though after launching the program. Like if
> I right click, it draws a blank window on the screen where the menu would be
> normally. Also if I open yast2 for instance, the buttons actually work and
> if I click around the blank window, it will open up more windows with the
> titles and everything. So it's talking to the server, it's just not drawing
> it completely. Don't know if that means anything, just thought I would throw
> that out there.
Right, it looks like for some reason, the xpra server is failing when
it tries to read graphical contents out of the server-side X server.
I'm attaching a patch that makes the relevant code a little more
paranoid and log a little more. If you apply it and run with '-d all'
again, you should see lines like "failed to name a window pixmap
(expect an X error soon)". If you do, then this is the problem.
(It should apply to whichever version of xpra you're currently using.
Use 'patch -p1 <composite-helper-more-robust.patch' in your source
tree, or even your build tree should probably work -- whatever
directory contains a 'wimpiggy' subdirectory.)
However, I can't figure out why this code wouldn't work -- it looks
like a bug in Xvfb, or in GTK+, or in PyGTK, or something. I wouldn't
be shocked if those pieces have received somewhat less testing on
S/390 than on other architectures. It might be worth trying to use a
more recent version of each of those (in that order). Otherwise, I'm
afraid I'm not sure what to say...
-- Nathaniel
diff -r 1ae63c9cddaa wimpiggy/composite.py
--- a/wimpiggy/composite.py Wed Aug 04 20:54:07 2010 -0700
+++ b/wimpiggy/composite.py Fri Mar 04 08:05:23 2011 -0800
@@ -72,13 +72,16 @@ class CompositeHelper(AutoPropGObjectMix
def invalidate_pixmap(self):
log("invalidating named pixmap", type="pixmap")
if self._listening_to is not None:
- # Don't want to stop listening to self._window!:
- assert self._window not in self._listening_to
- for w in self._listening_to:
- remove_event_receiver(w, self)
+ self._cleanup_listening(self._listening_to)
self._listening_to = None
self._contents_handle = None
+ def _cleanup_listening(self, listening):
+ # Don't want to stop listening to self._window!:
+ assert self._window not in self._listening_to
+ for w in self._listening_to:
+ remove_event_receiver(w, self)
+
def do_get_property_contents_handle(self, name):
if self._contents_handle is None:
log("refreshing named pixmap", type="pixmap")
@@ -99,24 +102,37 @@ class CompositeHelper(AutoPropGObjectMix
# 4) call NameWindowPixmap
# we are safe. (I think.)
listening = []
- win = get_parent(self._window)
- while win is not None and win.get_parent() is not None:
- # We have to use a lowlevel function to manipulate the
- # event selection here, because SubstructureRedirectMask
- # does not roundtrip through the GDK event mask
- # functions. So if we used them, here, we would clobber
- # corral window selection masks, and those don't deserve
- # clobbering. They are our friends! X is driving me
- # slowly mad.
- addXSelectInput(win, const["StructureNotifyMask"])
- add_event_receiver(win, self)
- listening.append(win)
- win = get_parent(win)
- handle = xcomposite_name_window_pixmap(self._window)
- self._contents_handle = handle
- # Don't save the listening set until after NameWindowPixmap
- # has succeeded, to maintain our invariant:
- self._listening_to = listening
+ try:
+ win = get_parent(self._window)
+ while win is not None and win.get_parent() is not None:
+ # We have to use a lowlevel function to manipulate the
+ # event selection here, because
+ # SubstructureRedirectMask does not roundtrip through
+ # the GDK event mask functions. So if we used them,
+ # here, we would clobber corral window selection
+ # masks, and those don't deserve clobbering. They are
+ # our friends! X is driving me slowly mad.
+ addXSelectInput(win, const["StructureNotifyMask"])
+ add_event_receiver(win, self)
+ listening.append(win)
+ win = get_parent(win)
+ handle = xcomposite_name_window_pixmap(self._window)
+ except:
+ try:
+ self._cleanup_listening(listening)
+ except:
+ pass
+ raise
+ if handle is None:
+ log("failed to name a window pixmap (expect an X error soon)",
+ type="pixmap")
+ self._cleanup_listening(listening)
+ else:
+ self._contents_handle = handle
+ # Don't save the listening set until after
+ # NameWindowPixmap has succeeded, to maintain our
+ # invariant:
+ self._listening_to = listening
trap.swallow(set_pixmap)
return self._contents_handle
_______________________________________________
Parti-discuss mailing list
[email protected]
http://lists.partiwm.org/cgi-bin/mailman/listinfo/parti-discuss
