reparenting the client's window creates a few issues:
• We get bogus UnmapNotify and MapNotify events that are ultimately
generated by amiwm itself.
• We do not get button events from the window, since its frame
(c->parent) that we have called XGrabButtons(3) onto, is not
its parent anymore.
To fix this, do not reparent the client window, but its frame window.
And keep the client window always inside the frame.
So when setting the list of supported _NET atoms, we can just pass a
pointer to ATOM[_NET_SUPPORTED] as the beginning of the array of
supported atoms, and NATOMS - _NET_SUPPORTED as its length.
Previously, adding a new atom into the code required:
1. In <icc.h>, add the line `extern Atom my_new_atom;` declaring the new atom.
2. In <icc.c>, add the line `Atom my_new_atom;` defining the new atom.
3. In <icc.c>, add the line `my_new_atom = XInternAtom(dpy, "MY_NEW_ATOM", False);`
interning this new atom, and assigning the interned value to the variable.
Now, just add a line with the atom's name in the X-MACRO in <icc.h>, and
a XInternAtoms(3) and a set of macros will do the rest.
All new atoms will be referred to as an entry in the ATOMS[] array.
For example:
ATOMS[MY_NEW_ATOM]
The exception are for those REALLY OLD atoms that have already a
dedicated compile-time-known value defined at <X11/Xatom.h>. Those
do not need to be interned; and are referred to with a constant
beginning with the `XA_` prefix. For example:
XA_WM_CLASS
State and control flow for modal mouse actions (dragging/resizing/etc)
was fragmented and scattered through main.c, making code convoluted and
hard to maintain.
I fixed that by applying good structured programming practics:
implementing sub-event-loops for those modal actions, rather than
checking which modal action is currently in action at the main
event-loop.
That commit took me more time than i expected... abstract out all
spaghettish, global-state control flow into a single, encapsulated
scope for each modal action… test if i broke anything… find any dead
code that is not used anymore (implementing that global state we had
before)…
I also put some `static` in global variables and routines, so the
compile can warn whether an unused variable/function is defined (and
help me on removing dead code).
The AmiWM manual was kinda incomplete. It does not describe how the WM
works, does not list all the configuration commands, nor the modules.
This commit rewrites the manual into the mandoc format, fill in the
gaps, and makes its section structure more standardized.
This commit implements the _NET_SUPPORTING_WM_CHECK property, defined
by the EWMH[1] (Extended Window Management Hints, aka wm-spec), which
indicates that a EWMH-compliant window manager is running.
A GTK application, when set to go fullscreen, first checks root window's
_NET_SUPPORTING_WM_CHECK property to determine whether a EWMH-compliant
window manager is running; and then checks the _NET_SUPPORTED property
to determine the supported protocols.
_NET_SUPPORTED was previously implemented, but _NET_SUPPORTING_WM_CHECK
was not.
This commit also fixes a misconception of the code related to screens.
The code deals with two different, unrelated concepts of screens: the
X11 screen, and the AmiWM screen.
The X11 screen, created by the X server independently of whether a
window manager is running, is a kind of “subdisplay”. A X11 display
structure, which describes a connection to the X server, has one or more
screens. Each X11 screen has its own root window and its own set of
clients that are specific to it, and cannot move to another X11 screen.
The user can even run two different window managers at the same time in
the same display, given that each is run in a different X11 screen.
Usually, there is only one X11 screen.
The _NET_SUPPORTING_WM_CHECK and _NET_SUPPORTED properties must be set
on the root window of each screen that AmiWM is managing. Previously,
however, they were set everytime an AmiWM screen (a totally different
concept) was realized. So, if i had 4 AmiWM screens configured in my
~/.amiwmrc file, setsupports() were called four times, even if i only
have a single X11 screen.
[1]: https://specifications.freedesktop.org/wm/latest/index.html
AC_HEADER_TIME is only needed if you need to include time.h and
sys/time.h at the same time.
The truth is this codebase is an unholy mess but at least this fixes
the warnings.
EINTR is happening during suspend/resume, which was causing all of
the modules to die (and requiring a restart of amiwm to re-run all
of its modules.)
This fixes things in FreeBSD suspend/resume so, well, the modules
keep working. :)
This allows a keyboard shortcut to move a client to a different amiwm
screen. That way when you have things like Firefox restart every window
in a single screen, you can quickly move screens to where they should be.
If the clock timer fires when global scr is NULL then we end up calling
redrawmenubar() with a NULL global scr, but a valid local scr.
So just pass in the scr used when doing "stuff" into redrawmenubar(),
making the 'scr' in redrawmenubar() also local scope.
