When applications are embedded in firefox via mozplugger, e.g. ooffice
for viewing .doc files or xpdf for viewing pdfs, and the embedded
application opens a popup window, we end up with a shared appicon for
the embedded app that remains until Window Maker is restarted.
The underlying cause is that mozplugger winds up reparenting the
embedded app's original leader window to a subwindow of itself, so
wmaker's SubstructureNotifyMask on the root window no longer picks up
the DestroyNotify when that original leader is destroyed.
It seems easy to fix once you track down the problem: when fixing
several other properties in fixLeaderProperties(), add
StructureNotifyMask so we get DestroyNotify no matter where the leader
gets reparented to. A more targeted fix would only add
StructureNotifyMask when the fake group's origLeader is set, but I don't
think a few extra structure notifications are going to cause us any real
problems.
Signed-off-by: Brad Jorsch <anomie@users.sourceforge.net>
When an app hides its window but doesn't exit (e.g. apps that "minimize"
to the system tray), the highlighting was not being removed.
When alt-tabbing, the tabbed-to app's icon was not getting highlighted.
Signed-off-by: Brad Jorsch <anomie@users.sourceforge.net>
Autoconf uses multiple levels of variables when defining paths. For
example, ${datadir} by default is ${datarootdir}, which by default is
${prefix}/share, which by default is /usr/local. Substituting from
./configure, as is done by AC_DEFINE or AC_DEFINE_UNQUOTED, does not
expand all these variables. This was causing some of our defines to have
garbage like "${prefix}/share/pixmaps" rather than the intended
"/usr/local/share/pixmaps".
The solution is to generate the files needing these paths from the
Makefile rather than from ./configure, because make does fully expand
all those levels.
Signed-off-by: Brad Jorsch <anomie@users.sourceforge.net>
An appicon is created for each group of windows, as specified by the
WM_HINTS window_group or the WM_CLIENT_LEADER. The "shared_appicon"
feature ignores the group leader specified by the application, replacing
it with a dummy leader matching the window's WM_CLASS. This causes
issues for dockapps, since each instance of a dockapp needs its own
appicon to display the different icon_windows, so shared_appicon is
automatically disabled for dockapp windows.
If the application creates some dockapp windows (no shared_appicon) and
some regular windows (with shared_appicon) with the same leader, it can
unexpectedly end up with two different appicons: one for the real group
leader it set and one for the fake leader created for shared_appicons.
Both of these appicons will try to use the same icon_window, which may
cause the dockapp window to suddenly lose its contents as they are moved
to the fake leader's appicon.
There is a simple fix: if a WApplication already exists (and has an
appicon) for the app-specified group leader window, disable
shared_appicon.
Signed-off-by: Brad Jorsch <anomie@users.sourceforge.net>
On the one hand, libWINGs wasn't linking against -lX11 when it should
have been. And on the other, only libWINGs needs Xft, only wmaker needs
Xrandr, only wmaker and wmsetbg need Xinerama, only libwraster needs
Xmu, and -lpng may not need -lz.
Cleaning this up can help distributions get their dependencies correct,
and might even avoid loading the unused libraries at runtime, so we may
as well do it.
Signed-off-by: Brad Jorsch <anomie@users.sourceforge.net>
On BSD systems, strlcat and strlcpy are included in the C library and
nothing special is needed. On Linux systems they are not, but libbsd may
be available to provide them. Use it if so.
This also adds wstrlcat and wstrlcpy instead of trying to maybe-provide
strlcat and strlcpy themselves, as that way there is no risk of symbol
conflicts. Not bumping the library version at this time, that should be
done (if necessary) before release.
Signed-off-by: Brad Jorsch <anomie@users.sourceforge.net>
Commit acc868d91d tried to change
--with-appspath to --with-gnustepdir, but either it wasn't done
correctly (and no one noticed) or old versions of AC_ARG_WITH worked
completely differently (and in an extremely odd way) because the option
remained --with-appspath and just the help text was changed. Then commit
bde572f673, which claims to have been
fixing some sort of issue with path assignment, broke it completely (i
guess) because the author there trusted the help text.
This completes the change from commit
acc868d91d that makes the option really be
--with-gnustepdir.
Signed-off-by: Brad Jorsch <anomie@users.sourceforge.net>
If you link against a library A that itself links against a library B,
it may or may not work to use symbols from library B in your executable;
this is known as "indirect linking". GNU ld does indirect linking by
default (but it can be disabled using --no-add-needed), but the new
experimental GNU gold linker does not do this. It's an easy fix for us,
as the tests are all already done in ./configure, we just need to tell
the Makefile.ams to use the results.
This should fix Debian bug #556677, if they ever start using this
branch.
[crmafra: Folded Andreas Metzler patch to update debian/changelog]
Signed-off-by: Brad Jorsch <anomie@users.sourceforge.net>
If the appicon is not in the dock/clip, or the dock/clip is not set to
"Keep on Top", there's a good chance you won't actually see the bouncing
because some other window is covering the appicon.
Besides adding the option to raise bouncing windows, this patch adds a
utility method to move a window back into its correct stacking position
after it has been messed with using XRaiseWindow.
Signed-off-by: Brad Jorsch <anomie@users.sourceforge.net>
I have my dock on the right edge of the screen, and my clip right next
to it. So the first icon underneath the clip is 64 pixels from the top
and 64 from the right. It seems silly for it to bounce downwards instead
of sideways like every other icon under the clip.
So make the choice of vertical or horizontal for docked icons be
relative to the dock or clip rather than absolute screen position.
Left/right or up/down is still chosen by screen position, as are
situations where the icon is a chess bishop's move from the clip.
I just spent an hour trying to find out why the latest wmaker-crm was
ignoring a number of the preferences in my configuration file. It turns
out the problem was that commit f41db5 added a value into the middle of
the enum in src/keybind.h, and the setting of
AUTOMAKE_OPTIONS = no-dependencies in all the Makefile.ams meant that
make wasn't smart enough to rebuild the files depending on it.
Would anyone mind if we just remove no-dependencies from
AUTOMAKE_OPTIONS? People can still run ./configure with
--disable-dependency-tracking to disable it at configure time.
On Wed, Sep 15, 2010 at 04:40:52PM +0200, Gilbert wrote:
>
> "." = {NoAppIcon = Yes;};
> "*" = {NoAppIcon = Yes;};
Oh, I see. It lets you override NoAppIcon generically instead of for
each app.
> Exactly, having 'DockApp' as the class name makes them behave more
> consistently internally. This might be fixed in some other way
> internally, but I couldn't figure it out...
Except it doesn't, the only thing it changes inside wmaker is that it
avoids displaying the name as "DockApp" on the settings window.
A "dockapp" is just an application that sets icon_window in WM_HINTS
(which causes a window to be displayed as the app icon) and
initial_state to WithdrawnState (which causes the main window to not
actually appear when the window is mapped). The class has nothing to do
with it.
The problems I mentioned stem from gtk+ not allowing you to set
initial_state to WithdrawnState. Openbox and fluxbox (and maybe others)
don't take the icon_window into their "slit" unless initial_state is
WithdrawnState, so the whole thing completely fails. Wmaker always uses
the icon_window for the app icon, so that part works fine, but the
"main" app window still flashes onscreen for a split second before the
app can call gdk_window_withdraw(), and wmaker applies SharedAppIcon to
it which winds up screwing up the "Kill" menu option.
That said, I've attached a patch to have wmaker treat any window with
class DockApp as if it had initial_state = WithdrawnState, and hopefully
openbox, fluxbox, and the like will pick up the idea too so gtk+
hacks[1] will no longer be needed. If this patch is accpeted, I'll poke
fluxbox and openbox to suggest the idea to them.
[1] E.g. http://wmudmount.svn.sourceforge.net/viewvc/wmudmount/dock.c?r1=8&r2=7
Subject: [PATCH] Improve dockapp recognition
Dockapps are traditionally recognized by having initial_state =
WithdrawnState in WM_HINTS. But some toolkits (e.g. gtk+) will not allow
setting initial_state in this way. So we offer an alternative: any
window with the res_class portion of WM_CLASS set as "DockApp" will be
treated as if it had initial_state = WithdrawnState.
In C, dividing two integers automatically rounds towards zero, so ceil(a
/ b) is useless as the result is truncated before ceil ever sees it. The
correct result for positive integers is obtained by (a + b - 1) / b.
I personally cannot stand stealing Ctrl+Wheel for resizing windows, so I
turned it off. But wmaker is still preventing Ctrl+Wheel from reaching
apps, which is certainly not the behavior I desire.
We grab Mod+Button events for our own purposes, and swallow them using
XAllowEvents(AsyncPointer) to prevent the client window from seeing
them. But if events are coming in fast enough (e.g. via fast wheel
scrolling) so that multiple grabbed events are in the queue, the second
queued event would be allowed to leak through to the client by the
unconditional XAllowEvents(ReplayPointer) a few lines later.
One misuse of CARD32 was missed back when someone fixed the 64-bit
issues (background: X has 8-bit, 16-bit, and "32-bit stored in a long"
data formats; on a 64-bit system, long is 64 bits). This was causing
minimized windows to be restored as shaded, and possibly occasional
crashes, when wmaker was restarted.
Also, throw in a safety check that 10 items really were returned before
trying to access them.
The "make dist" command can be used to generate the release tarball, but
only if the Makefile.am correctly specifies EXTRA_DIST, specifies
headers in foo_SOURCES, and so on.
Initialized variables that are conditionally set. In particular, this
construct is dangerous:
void *p;
if (something) p = couldReturnNull();
if (!p) p = fallbackFunction();
C99 defines new strict aliasing rules to allow compilers to make certain
optimizations. These rules prohibit converting an XEvent to an event
struct (e.g. XShapeEvent) that is not already in the XEvent union using
pointer type punning (e.g. "(XShapeEvent *)&ev"), and vice versa. The
canonical fix seems to be to create a union between XEvent and the
extension event struct to make the aliasing explicit, so do that.
A few internationalized format strings have only one %s in the default
while two strings are being passed. Change those defaults to have two
%s's, which means we also have to update the .po files to match. May as
well throw in the extra %s in the translated version while we're at it.
I noticed a bug today in menu workspace positioning (in the next tree):
if I right-click and hold the button down at the left edge of the
screen, the menu appears (correctly) half off the edge but will *not*
slide back onto the screen. Compare the behavior to right-clicking at
the right edge of the screen: the menu appears half off the edge, but
sliding the mouse to the edge causes the menu to slide until it is fully
visible.
Also, opening a submenu in this state positions the submenu as if the
menu were fully on the screen, leaving a gap between the menu and the
submenu.
If the menu happens to also go off the bottom of the screen, moving the
mouse to the bottom edge causes the issue to be magically fixed as soon
as the sliding upwards begins.
I also note inconsistent behavior when simply right-clicking (without
holding) to bring up the menu: at the right or bottom edge the menu
appears in the correct partially off the edge position, but at the left
edge it immediately jumps to be fully on-screen, almost as if
WrapMenus = YES were set for the left edge only.
I bisected it to d316260395. As far as I
can tell the "fix" there was to position the menu at the correct
negative X position but then lie to wmaker so it thought the menu was at
X=0. Presumably the "WrapMenus" behavior was the intended result.
Since (AFAIK) the window menu is the only one with this problem, why
don't we just check if x is too negative in OpenWindowMenu()?
This patch adds the DockRaiseLowerKey shortcut, which raises/lowers
the dock depending on whether the dock is lowered/raised.
[crmafra: Reformatted Brad's patch against git repo and removed the
DockRaiseKey and DockLowerKey shortcuts ]
Brad Jorsch