[4935] in testers
Why panel is bouncing
daemon@ATHENA.MIT.EDU (Mitchell E Berger)
Mon Jul 2 21:10:02 2001
Message-Id: <200107030109.VAA20417@byte-me.mit.edu>
To: testers@MIT.EDU
Date: Mon, 02 Jul 2001 21:09:59 -0400
From: Mitchell E Berger <mitchb@MIT.EDU>
In working with several machines that do everything in slow motion, I've
figured out why panel bounces from the bottom to the top of the screen and
back upon login, sometimes fully drawn and sometimes not.
Panel isn't broken (or at least not in this respect). It is a result of
sawfish being started after panel rather than before. This is a change
that was made in transaction 7755 in source-reviewers, where the description
as well as the comment in the CVS log indicate that the change wasn't needed
for any reason, but seemed like the right thing to do since sawfish might
like certain properties to already be set, and because sawfish might start
its own esd, but that this would not be a problem.
Bob and I have tested reverting the order to starting sawfish before starting
panel, and everything seemed fine; the startup sequence didn't log any errors,
panel didn't bounce, all the menus and launchers seem to work, and esd also
worked correctly and even passed the test script that it comes with.
I believe the reason panel was sometimes drawn and sometimes not when it
bounced is a result of a note Greg made either in his message to
source-reviewers or somewhere else, which says that we can't really enforce
the order in which background processes do things, but can push them in the
right direction by starting them in the right order. Thus, whether panel
had completed drawing by the time sawfish noticed it certainly could be
variable across machines and logins.
Since the panel bounce is a bit disconcerting to those who see it, and it
doesn't seem that the present order is needed for any other reason, could we
switch back to sawfish, then panel? I'd be happy to generate the patch for
it, but I suppose there might be objections, so I thought I'd ask here first.
Mitch
