fluxbox now properly displays windows that require ARGB visuals when
an external compositor is running. This was done by creating the
container window with the correct visual and colormap when needed.
Closes#2874629
the deleted function was never used, otherwise it would generate an error with other compilers as
well. icc noticed that it was nonsensical even when it wasn't used and complained.
The idea is that connecting to a signal doesn't change it's state or the state of the object
owning the signal (even though it needs to add the functor to the list for later reference).
Emitting, on the other hand, is usually done as a result of a state change and therefore remains
non-const.
Additional benefit of this arrangement is that objects can export const references to signals to
allow connecting, while keeping the ability to emit to themselves.
without this it wasn't possible to construct a Slot returning void from functors returning some
real value because the compiler would complain about "return statement with a value in a function
returning void".
Theoretically, this may produce some unexpected type conversions, because static_cast is slightly
stronger than implicit cast, but I judge the risk to be negligable (the alternative would be to
provide explicit specializations for slots returning void - too much typing)
without them, gcc would compare them by converting them to bool first, which is not exactly what
one would expect. Frankly, I'm surprised it even worked without this.
it is too easy too shoot yourself in the foot with it, other smart pointers also don't allow such
assignments. If you do want to assign to a RefCount pointer, use reset().
ps: assignment between two RefCounts remains possible, of course.
Added some polish around them and, to mark this special occasion, moved them out of the SigImpl
namespace.
PS: This partially reverts commit 0775350fee, since I had to
reintroduce ReturnType template parameter, because it will be used in other places. But Signal classes
remain without the ReturnType, because I still cannot imagine how would it be used.
Commit 690d926 (introduced FbTk::BidiString) broke building without
HAVE_ICONV, because of wrong variable name and use of iconv_t type.
Signed-off-by: Peter Korsgaard <jacmet@sunsite.dk>
otherwise a race condition might appear between copying the default
config files to .fluxbox/ and executing 'fluxbox-update_configs'. under
macosx this lead to 'fluxbox hangs'.
setxkbmap and xmodmap both might change the keycodes. thus fluxbox needs
to remap the keysyms from the currently loaded keytree to new keycodes
after it received a 'MappingNotify' event.
we do not reload() the keys file because:
* the user might work on it right now
* remap only needed symbols is cheaper than parsing the keysfile anyway
The icons coming from _NET_WM_ICON are argb32. fluxbox uses such icons
in entities such as 'clientmenu', 'iconbar', 'titlebar'. These entities
are not related to the depth of the winclient but to fluxbox's default
depth. Using 'winclient.depth()' is a mistake, since fluxbox is unable
to copy pixmaps from 32bit to 24/16/15bit.
It is not necessary either, because fluxbox should extract the argb32 icon
data directly into the pixmap with the correct depth in the first place.
This fixes (reopened) bug #1852693.
Note: The whole icon code in fluxbox is quite messy, lots of copying and
scaling. It might be simpler and fater to just extract the icon when needed
, in just the size that is needed.
Prior to this patch, TextDialog did not handle parentrelative labels;
borrowing code from OSDWindow, now TextDialog properly handles this.
Eventually, I plan to combine the good parts of TextDialog and OSDWindow
into OSDWindow, and then let TextDialog derive from OSDWindow... but for
now, here's the band-aid fix for the issue without the refactorization.
TextDialog windows (such as 'set title', or 'edit clock format') now
properly render their titlebar to match the current style. Previously,
the titlebar font settings were controlled by window.font.*, and this
became apparent because window.font.effect needed to be set to put
'halo' on the titlebar of these TextDialogs. However,
window.font.effect controls the effect on the text within the editbox on
these as well, having these set in the same location is far from
desirable. Specifically because the TextDialog essentially emulates the
titlebar of other windows. So, after this change, window.font.effect
still controls the text in the editbox, but the titlebar area is themed
as the titlebars of other windows.
Text justification of titlebar and border width, is now respected in
TextDialog. Also, previously the edit box was sized in an odd way; one
which left part of the window actually undrawn (so you'd get artifacts
from whatever happened to be on the screen in that position when it
opened); this has been addressed.
TextDialogs, rather than going to the center of the entire desktop, now
center on the _current head_, which prevents the awful positioning where
half of the window is on one screen and half is on another.
Lastly, when hovering your mouse over the iconbar button for a window
with a title that's too wide to fit on the button, a tooltip is
displayed. This tooltip themes itself to match the button, however it
wasn't using the theme properly (similarly to how TextDialog was)... and
as such, also wasn't using the proper font effects. This has been
addressed just as it was for the TextDialog.
previously, the tracker disconnected itself only when the caller passed withTracker = true to the
leave() function. However, the default value was for the parameter was false.
Non disconnecting from signal when stopping tracking creates very dangerous situation because the
signal still holds a pointer to the tracker. This resulted in a segfault when exiting fluxbox,
because the tracker (FluxboxWindow) got destroyed before the signal (BScreen::focusedWindowSig),
and the signal was using an invalid pointer when it tried to disconnect itself from the tracker.
Instead of setting withTracker to true by default or changing all invocations of leave(), I
decided to make the tracker disconnect itself unconditionally because I could not find a use case
for the opposite behaviour.
PS: This message is in fact longer than the actual commit.
previously they declared the parameter, but ignored the return value of the member function. I've
changed it so they pass the return value, if it is not void. MemFunSelectArg didn't have the
ReturnType template parameter, so I added it for consistency. Since I was already editing the, I
made all the operator()s const.
- merged all the common stuff from 0,1,2,3 argument versions into one common base class
- removed ReturnType template parameter as it was instantiated with "void" everywhere and the
current ignores the return value of the callbacks anyway
this was possible (and used) with FbTk::Subject, but the implemetation of FbTk::Signal didn't
support it, which made it impossible to continue with conversion.
Replace CallbackHolder, FunctorHolder and SlotHolder with a (smaller) set of polymorphic classes.
SignalHolder now stores a (smart) pointer to the class.
the previous version of operator*() made no sense. E.g., it violated the invariant
(*ptr).foo <=> ptr->foo. The dereferencing operator now returns a reference to the pointed-to
object, rather than a pointer to it.
I also added a bool conversion operator, which can be used in testing the NULL-ness of the
pointer. Anyone wondering if that could be done in a simpler way is encouraged to read
<http://www.artima.com/cppsource/safebool.html>.
And, finally, I removed the mutable flag from the m_data member, since it does not need it.
basically, i just got rid of Signal[0-3] classes and moved their contents to the appropriate
specialization of FbTk::Signal
also, this fixes the
no matching function for call to 'MemFunIgnoreArgs(FbTk::Signal<void, FbTk::SigImpl::EmptyArg,
FbTk::SigImpl::EmptyArg, FbTk::SigImpl::EmptyArg>&, void (FbTk::SigImpl::Signal0<void>::*)())'
error i had in the following commit.
Introduces a new member function, FbWindow::cardinalProperty()
This change also changes other code that previously used
FbWindow::property() to do the same thing as the new function; this
reduces code duplication.
There are still some bits of code (Ewmh.cc, extractNetWmIcon()) that use
FbWindow::property() to retrieve XA_CARDINAL values, but as the new
method is designed for getting a _single_ property, and that code uses
FbWindow::property() to retrieve the number of values present, and then
grab all of them; it's a different use case. I opted to not try to make
cardinalProperty() into some monolithic all-purpose cardinal method;
FbWindow::property() works just fine for that.
This change also adds an optional (default=NULL) boolean to
FbWindow::textProperty and friends that allows the caller to determine
whether or not a value was actually retrieved. This was necessary for
integrating FbWindow::cardinalProperty with the codebase, and it seemed
to fit with FbWindow::textProperty as well. Prior to this change, if
you got a return value of "", you wouldn't know if you successfully
retrieved the value which happened to be blank, or if you failed to
retrieve the value. Now, you can pass the address of a boolean if you
so choose in order to differentiate these situations; the same applies
to the new FbWindow::cardinalProperty().
and add a configure option for building them (default: off).
This patch fixes an error when configuring the tarball, where configure complained it cannot find
src/tests/Makefile.in.
biggest change: renaming of 'class Layer' to 'class ResourceLayer' in
Layer.hh to handle complaints about 'ambiguous Layer: Layer or FbTk::Layer'.
'ResourceLayer' sounds crappy, but that was the best I could come
up with right now.
changing the layout or the number of the attached monitors
might lead to invisible windows. only those windows should be
positioned to the closest monitor, if they are not visible
already.
m_client.focus() might call ~ClientMenuItem(), thus m_client is not available anymore. the crash was triggered by trying to deiconify a client via the clientmenu. this triggers a signal to remove and destroy the current ClientMenuItem.
warping a transient window to the next workspace creates a cycle between
BScreen::changeWorkspaceID(), BScreen::reassociateWindow(), Workspace::removeWindow(),
FocusControl::unfocusWindow(), FluxboxWindow::setCurrentClient(), FluxboxWindow::focus()
and so on.
for now we just stop allowing transient windows to be warped, it is most
likely a bad idea anyway having the modal dialog on one workspace and the
(dead) main window on another one.
this issue must be analyzed further, since there might be a deeper problem
with the way the focus code works.
FbTk::Menu::updateMenu() recalculates the width() and height() of the
window. this must be done before the menu is moved to fit completely
onto the screen.
the 'menu().raise()' command is not needed, that is done inside
'menu().show()' already.
The issue was caused by a client that got an unmap notify and had a FocusIn event
directly after in the event queue. The focus in event was handled by Fluxbox::revertFocus
before the WinClient.cc removed itself from the search window list.
WindowMenuAccessor returned strange alpha values if compiled
with 'g++ -Os'; unholy black magic happens if template<int> faces
functions returning only 'usigned char'.
With ROT90-SystemTray fluxbox crashed. It is a bit unclear of where
to catch pixmaps / windows with either width or height equal to 0; IMHO
this needs more investigation.
nxserver-3.x creates a XExposeEvent for the unmapped FbTk::Menu(), which is
not configured completely yet (hence unmapped). this causes a call to
FbTk::Menu::clearItem() which then uses a value of 0 for m_rows_per_column
to divide the current index which triggers a SIGFPE.
it is still unclear, why nxserver-3.x creates the XExposeEvent for the
unmapped (menu) window. doing nothing if the menu is unmapped 'fixes'
the problem for now.
many thanks to Lars Engels (bsd-geek.de) to assist me in debugging this
issue.
a newer xlib recently changed the result of 'DisplayString()' a little bit:
instead of returning ':0.0' or ':1.0' it yields ':0' or ':1'. our code to
transform this string into something that includes the currently used
Screen worked only on something like ':0.0'.
we now find the '.' after the ':' and strip that part away.
by using FbCommands::ExecureCmd() we run 'fluxbox-update_configs' in
the bacground while booting a new fluxbox instance already. 'system()'
waits until 'fluxbox-update_configs' is done and then gives the control
back to fluxbox.
before bringing up the first instance of Fluxbox we prepare the directory
and the files it needs. if the config version of exiting files is lower
than what we expect, we upgrade the config files. after that we bring
up Fluxbox.
the old way was problematic because setupConfigFiles() calls
'fluxbox-update_configs' which does its job in the background while
fluxbox continues to boot. 'fluxbox-update_configs' sends a USR2 signal
to the booting fluxbox (it might even be finished, no one knows) which
triggers 'load_rc()' which triggered 'setupConfigFiles()' again which
might trigger 'fluxbox-update_configs' again (on my machine
'fluxbox-update_configs' was called 3 times and left a pretty crippled
'keys' file when it was done).
bootstrapping before bringing up fluxbox resolves the issue. as a bonus:
no need to send USR2 to fluxbox to reload the config file because fluxbox
has not even tried to read it yet.
to quote from 'man putenv':
The string pointed to by string becomes part of the environment,
so altering the string changes the environment.
so, using putenv like
{
std::string foo("FOO=bar");
putenv(foo.c_str());
}
is wrong and leads to a potentially corrupted environment. valgrind
complaint correctly.
FbTk::App seems to be the appropriate place to hold '::seten()'
because it alters the environment of the application.
## FbTk/ImageControl.hh:
* moved 'getSqrt()' over to TextureRender.cc because it is used only there
* removed unused members m_root_window
* use std::vector<> for colors and for the gradient buffers.
## FbTk/ImageControl.cc:
* fixed memory leak in '~ImageControl', 'pixels' where not deleted[]
at all. switched to std::vector, non-issue anymore.
* moved identical code into functions 'allocateUnallocatedColors', 'initColortables'
## FbTk/TextureRenderer:
* removed 'from', 'to', 'interlaced' from class, only used in
'renderGradient()' and are actually part of the given
texture.
* removed 'xtable', 'ytable' from class, only used in
'renderGradient()'
* removed 'colors' from the class, is part of 'control'
* moved render actions (invert, bevel1, bevel2, *gradient), they need more
love but for now this is sufficient
a lot of the parsing-related and menu-building stuff is just plain
stupid boiler plate code. putting the data into a const array
separates the boiler plate data from the actual code.
* ClientPattern.cc: make compiler happy (complaints about XPROP not handled in switch)
* FbTk/FbPixmap.cc FbTk/StringUtil.cc: signed vs unsigned
* FbTk/FbString.cc: missing return
* WinClient.cc: create atoms only once; use helper function
* Window.cc: use a helper function
for each window (windows, menue etc) the initCorners() function of
FbTk/Shape.cc was called (which is ok). On each call the corner pixmaps
(of the Screen the window belongs to) were created and destroyed the
formerly created pixmaps. this is not needed.
the global 's_corners' vector was still used on an application shutdown
(including ctrl-c SIGINT). depending on when 's_corners' was created on
the stack this could lead to an exception ('pure virtual called') while
trying to free the 'FbPixmaps' without having a valid
'FbTk::Application::instance()' left.
our users should migrate seamlessly to our new windows-cycling behaviour
without losing their current way of doing things. so, 'fluxbox-update_configs'
tries its best to add '(workspace=[current])' wherever it looks like a valid
place for it. such changed lines are marked by '!! FBCV13' at the end and can
be easily spotted in case the updater got it wrong.
addtional changes:
* added '-check' to 'fluxbox-update_configs' to check the version number in
'~/.fluxbox/init' vs. the version number of 'fluxbox-update_configs'
* moved the update-code for each version into its own function, easier
housekeeping
commit 882a50fe1d removed the hardcoded limit to
cycle to windows on other workspaces, though it was and IS a good default
behaviour. if users do not want this (they want to cycle to windows on other
workspaces), they should explicitly avoid '(workspace=[current])' in their
keysfile.
these two allow 'tagging' of arbitrary windows with 'tags' (or 'labels'). such 'tagged' windows can then be used in ':NextWindow (@PROP=foo)' commands to quickly cycle through a subset of available windows.
since the 'tags' are applied as real xproperties to a window they survive a restart of fluxbox or even another windowmanager. the user can also set the tags by using xprop(1).
the next step regarding the UI should be to visualize the tags of a window.
reasons:
* fixes#1732115, allows to have chatwindows on another workspaces and reach them via :NextWindow (urgent=yes)
* old behavior can be achieve by using the (workspace=[current]) pattern
a 'BidiString' holds both the logical content and the visual reordered
version of the content of a string. this helps to reduce the number of
calls to reorder the string before drawing it (as introduced in the patch
from Ken Bloom) and to be more consistent in menus and textboxes (drawing
cursors and underlining text).
* parse the string only once for a bunch of chars to check instead of
starting over again and again from the beginning, created a helper
function to do this (FbTk::StringUtil::findCharFromAlphabetAfterTrigger)
* put same code into a function (setClockModeLabel())
* use much simpler code to switch between 12h and 24h mode and replace
the fmt-switches
While 769130f51a did fix one issue, it introduced another by changing the
logic related to the new SignalTracker. The original logic (introduced in
9ad388c5bf) was:
-> in 'leave(Signal)', only call 'disconnect'
-> in 'leaveAll()', call 'disconnect' and 'disconnectTracker'
But 769130f51a inverted this, calling 'disconnectTracker' in both cases but
only 'disconnect' in the 'leaveAll()' case, which would leave unattached signals
around after calling 'leave(Signal)'.
This fix not only repairs the logic, but renames the ambiguous 'disconnect'
boolean to something more explicit: 'withTracker'.
As noted in the previous commit, StrictMouseFocus now works as advertised:
Focus follows mouse on every EnterNotify event (except when the "ClientMenu"
closes or during alt+tab window cycling)
This is not actually implemented yet, but from now on, "MouseFocus" means:
Focus follows mouse only when you are moving the mouse, any EnterNotify events
caused by non-mouse operations (window closing, keycommands, changing
desktops) will *not* shift focus
And once fully-implemented, "StrictMouseFocus" will mean:
Focus follows mouse on every EnterNotify event (except when the "ClientMenu"
closes or during alt+tab window cycling)
the old way of deciding which head to (re)maximize the current window
was to just test if the center of the window is INSIDE which head.
now we calculate the closest head which fixes the problem
using a smaller type ('long' on 64bit is 8 byte, 'int' is 4) results in
strange properties, eg:
_NET_WM_STATE(ATOM) = _NET_WM_STATE_MAXIMIZED_HORZ, undefined atom # 0x2726130
(when xpropping a maximized xterm). this might cause misbehavior in apps.
The MemFunSelectArg* functions can be used to select
a specific argument from a signal. For example this would
select the string argument as argument to the callback:
Signal<void, int, float, string> signal;
signal.connect(MemFunSelectArg2(obj, &Object::takesOneStringArg));
signal.emit(10, 3.14, "hello");
...
void Object::takesOneStringArg(const string& value) {
...
}
For example connecting a function that takes two arguments
to a signal that emits three arguments:
struct Functor {
void show(int a, int b);
};
Functor f;
Signal<void, int, int, int> s3;
s3.connect(MemFunIgnoreArgs(f, &Functor::show));
the idea (as a first patch) for this change was provided by
Thomas Habets (thomas at habets pp se). instead of having a
fixed interval each second, we now calculate the next point in
time based upon the format string used by the clocktool to render
the time. as long as no seconds are shown fluxbox now wakes up
once every minute.
the hardcoded 'OnTitlebar Mouse1 :Raise' (see Window.cc, FluxboxWindow::buttonPressEvent())
is disabled for now, should be added to fluxbox-update_configs
