RFE: Atomic query requests for XFIXES

Adam Jackson ajax at redhat.com
Mon Mar 15 09:06:14 PDT 2010

GTK has an internal call path for "get the window under the cursor".
This gets called rather more than you might expect, for things like
tooltip selection and positioning.  It's implemented with a server grab
around a chain of XQueryPointer calls; details, for the curious:


Now the server grab here is a little misguided, since it will not in
fact hold the cursor still.  It _will_ keep the window hierarchy
stable, but that's not the same thing.  Regardless, what the toolkit is
actually looking for here is a single request to give a snapshot of
cursor containment.  If we had that, we wouldn't need the grab, and
we'd be robust in the face of cursor motion.  And the grab is really
quite impolite when the client is remote.

X has this information already, trivially.  The sprite trace already
contains the cursor window and its parentage.  This even works naïvely
for XI2, PickPointer() will do the same thing for FixesQueryPointer as
for the core request.  It might be worth extending this to arbitrary
pointers?  Feedback appreciated.

While we're at it, we could extend this to QueryTree, so you can get
the complete ancestry of any window.  There _was_ a codepath in GTK
that could make use of this feature, but it seems to have been obviated
by the client-side windows change.  It'd be trivial, so I went ahead
and spec'd it, but I'm just as happy to leave it out on grounds of

I'm not aware of any other queries in X where we could eliminate grabs
like this, but I didn't look especially hard.

There are still a few cases in GTK and Qt where server grabs are
required, but they seem to all be around XEMBED and XDND.  It's not
clear to me that the server could provide any additional atomicity to
help with that, but, again, maybe there's something I'm not thinking

- ajax

More information about the xorg-devel mailing list