This patch adds the --disable-verbose-compile switch to the
configure script.
When this option is used it reduces the verbosity of compilation
messages to the essential. However compiler warnings are not affected
and thus gain visibility by not standing in the middle of a storm
of other messages.
In summary, the compilation messages are reduced to a stream of
CC array.o
CC bagtree.o
CC configuration.o
CC connection.o
CC data.o
[...]
instead of a mind-boggling flux of
gcc -DHAVE_CONFIG_H -I. -I../src -I../WINGs/WINGs -I../wrlib -I../src
-I/usr/include/freetype2 -I/usr/local/include
-DLOCALEDIR=\"/usr/local/lib/loca\"/usr/local/share/WINGs\"
-DDEBUG -fno-strict-aliasing -g -O2 -c array.c
gcc -DHAVE_CONFIG_H -I. -I../src -I../WINGs/WINGs
-I../wrlib -I../src -I/usr/include/freetype2 -I/usr/local/include
-DLOCALEDIR=\"/usr/local/lib/loca\"/usr/local/share/WINGs\"
-DDEBUG -fno-strict-aliasing -g -O2 -c bagtree.c
[...]
event handler timer.
After upgrading my kernel recently I noticed that dnotify has been
depreciated, so I decided to try and implement the new inotify code in
Window Maker instead.
During testing, I also found that one of the timers which was removed
(the one causing the most wake-ups), calling delayedAction, was
responsible for handling signals. Basically with this timer removed,
signals were only handled after an X event occurs.
After looking at the delayedAction function, I couldn't see the purpose of it.
It certainly wouldn't cause any delay as it was called by the timer every
500ms, so there is no time correlation with when a signal was received.
Also, it appeared to count the signals and call DispatchEvent for each
one, but it appears DispatchEvent would just handle the most recent signal
and take action on that. The signals handled by delayedAction are the
various exit and reset signals, so only one need to be handled. I
therefore have commented out delayedAction (it wasn't called by any other
procedure) and added a call to DispatchEvent imediately after the signal
is registered, in handleExitSig.
I'm not sure what problems this may cause with dead children - these are
only cleaned up after an Xevent now that the timer is removed- but I
haven't observed any problems since a few months ago.
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>
WPrefs tries to use the fonts listed in the WMCreateFont() call (from WPrefs.c),
font = WMCreateFont(scr, "Lucida Sans,URW Gothic L,Times New Roman,serif"
":bold:pixelsize=26:antialias=true");
and 'font' is later used without accounting the possibility of it being NULL in
WMSetLabelFont(WPrefs.nameL, font);
WMReleaseFont(font);
In particular, WMReleaseFont(font) will kill WPrefs ungracefully with the
following message if font=NULL
WPrefs: wfont.c:193: WMReleaseFont: Assertion `font!=((void *)0)' failed.
Aborted
That happens because the return value of WMCreateFont() can be NULL, so this
patch makes WMCreateFont() never return NULL. If the font creation fails for
some reason (see below), we try to use the emergency DEFAULT_FONT and print
debugging information telling what's going on. If the use of DEFAULT_FONT
also fails, then we terminate WPrefs with exit(1) and let the user know
why exactly if failed.
This bug happened because the font "Lucida Sans" (the first possibility
in the initial call to WMCreateFont()) "exists" in my system as a stale
symbolic link to a location which no longer exists after an upgrade from
java from 1.5.0.14 to 1.5.0.15
/etc/alternatives/LucidaSansDemiBold.ttf ->
/usr/lib/jvm/java-1.5.0-sun-1.5.0.14/jre/lib/fonts/LucidaSansDemiBold.ttf
So the call to XftFontOpenName(display, scrPtr->screen, fname) with
"Lucida Sans" being the first possibility for Xft to try out ended up in 'font'
being NULL because, as far as the Xft library was concerned, "Lucida Sans"
produced a positive match (due to it existing as a symbolic link) but in the end
a NULL result was produced due to the missing symbolic link destination. This later
exposed the bug of WMCreateFont() returning font=NULL and WPrefs.c not checking
whether font=NULL before using it. Bang!
If "Lucida Sans" was the _second_ entry to try and the first one had suceeded,
this bug would not have surfaced.
This solves https://qa.mandriva.com/show_bug.cgi?id=39677
Signed-off-by: Carlos R. Mafra <crmafra@ift.unesp.br>
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>
This also fixed an endless loop that could be entered by the previous code
in certain situations, after the patch to fix navigation/selection in a
textfiled with UTF8 chars was applied.
fix the 2 problems mentioend below:
- Fixed buggy handling of UTF8 characters in textfields in WINGs.
- Fixed segfault in WPrefs when some font description is missing from the
configuration file.
WINGs based applications should now work with both KDE and GNOME
applications (Sylvain Reynal <sreynal@nerim.net>)
- better check for the XDND protocol version when interoperating with other
applications. As it seems xdnd version 3 (which WINGs supports) and newer
are not backward compatible with xdnd version 1 and 2. This is why WINGs
applications cannot interoperate with GNUstep applications (which uses
xdnd version 2). Xdnd version 4 and 5 are backwards compatible with
version 3 though. (Sylvain Reynal <sreynal@nerim.net>)
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.
