Pedro Gimeno explains:
"As you most likely already know, every X Window event comes with a
timestamp. You can see it using e.g. xev.
I don't know if this is Debian-specific and I'm no X Window expert, but
the fact is that in my machine this timestamp, according to my
/usr/include/X11/X.h, is a 32-bit unsigned integer and it holds the
event time in milliseconds according to my tests.
This timestamp appears to be in milliseconds from the start of the Unix
epoch, modulo 2^32. The problem is that 32 bits are too little for this
purpose: 2^32 milliseconds are exactly 49 days, 17 hours, 2 minutes and
47.296 seconds.
Now, the WindowMaker code rejects any focus event whose timestamp is
less than the last one seen, using code similar to this:
if (timestamp < LastTimeStamp) return;
LastTimeStamp = timestamp;
This is a disaster when timestamp wraps around because of the 32-bit
limit. Say LastTimeStamp equals 2^32-1 and a focus event comes two
milliseconds after. Its timestamp will equal 1 because of the
wraparound. Obviously, 1 < 2^32-1 so the event will be rejected even if
it comes from the future (relative to LastTimeStamp) and not from the
past. Focus events are no longer accepted because nothing can be greater
than 2^32-1 (unless you click on that exact millisecond, 49 days after).
If you look in http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=102314
you'll see two people confirming the periodicity and predictability of
the bug. Arnaud Giersch was who noted that the period was 2^32
milliseconds (and not 2^31 as I thought it could be because of a signed
int involved) and who noted the coincidence with the Unix epoch.
My fix works by subtracting one timestamp from the other (an operation
which is done modulo 2^32) and rejecting the focus event only if the new
timestamp is within 1 minute (60000 milliseconds) in the past. Given the
periodicity of the timestamp, actually after 49.7 days there will be
another minute in the future during which focus events will not be
accepted; again after 99.4 days, etc., but thanks to the subtraction,
they will now be relative to the *last* focus change and not to the Unix
epoch.
The patch could be simplified to get rid of the compareTimes function.
It's written like that because it comes from the previous patch (which
you can see in the same bug report).
So the date comparison line could end up looking like this:
if (scr->flags.ignore_focus_events || LastFocusChange - timestamp <
60000)
so you can get rid of the compareTimes function. However please respect
the other change of 'int' to 'Time'.
As for why you aren't experiencing this problem, perhaps the event
timestamp does not come from the Unix epoch in your X Window system, but
rather starts counting when you start it or your computer, I don't know.
Or maybe your X Window server's timestamps are 64 bits wide instead of
32 bits, which I don't think because I doubt the X11 protocol specifies
a 64-bit record for the transmission of the timestamp. In either case,
the output of xev will probably help understanding the cause.
Or maybe even it's a click-to-focus only bug and you're not using that
mode. I haven't checked who calls wSetFocusTo.
Note that I can reproduce it just by manipulating the system clock,
which seems to be tied to the event timestamps."
For more information see:
http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=102314
Reported-by: Joey Hess <joeyh@debian.org>
Signed-off-by: Pedro Gimeno <parigalo@formauri.es>
This patch removes wmaker from PowerTop's shame list, where
it appeared with ~3-4 wakeups/second.
It adds the linux kernel's dnotify mechanism (adapted from
the example in Documentation/dnotify.txt in the kernel source),
to detect when a configuration file in ~/GNUStep/Defaults has
changed to load it again on-the-fly. For me it usually means that
modifications to ~/GNUStep/Defaults/WMRootMenu via the 'genmenu'
script are automatically detected and loaded.
The use of dnotify makes the ancient behaviour of polling unecessary
and cuts down the wakeups count.
Other 'apparently' useless timers are also deleted and it's been almost
one year now that I use this patched exclusively without problems, so
I am pretty sure that it doesn't hurt to remove them.
The end result of all this is that wmaker generates 0 (zero) wakeups
when idle in a Linux system.
Signed-off-by: Carlos R. Mafra <crmafra@ift.unesp.br>
small integer and causing unexpected color releases that can crash
Window Maker (Martin Frydl <martin.frydl@systinet.com>)
- fixed a small memory leak in WINGs/wview.c caused by not releasing the
background color of a view (Martin Frydl <martin.frydl@systinet.com>)
on 64bit platforms (Max Loparyev <max@city.veganet.ru>)
- fixed issue with icon colors on big endian platforms
(Max Loparyev <max@city.veganet.ru>)
patch based on a previous patch by Vladimir Nadvornik <nadvornik@suse.cz>
belong to, that works without auto-arranging the icons, while still avoiding
multiple icons in the same spot by moving the miniwindows to a new slot if
their old slot was occupied in the meantime.
- double clicking an appicon will also raise the miniwindows that belong to
that application to the front (along with the normal windows).
- Fixed WPrefs path in the installed WMState (changed after recent WPrefs
installation path changes)
- Fixed wmaker.inst (apparently some very noisy individual was unable to get
right his 10 line patch to rename a path)
the old .AppInfo/... to the new Library/WindowMaker/... on the fly when
Window Maker starts.
This should allow a transparent transition without any need for users to do
anything.
in the first place in his trivial 10 line patch which only attempted to
replace one path with another.
He must be a big time user of his own patch, to not have noticed where the
paths pointed after his change.
- Keep menu and title text at 12px since there is not constrained by width
- Removed some obsoleted options from wconfig.h
- Added a lighter image for the switch panel
