Here's how it goes:
1. stdin reaches EOF, kmscon_eloop_rm_fd is called on stdin_fd.
2. The fd object is freed, the memory is reused in the meantime.
3. rm_fd is called again in destroy_eloop.
4. con->stdin_fd points to garbage and we get a segmentation fault.
Signed-off-by: Ran Benita <ran234@gmail.com>
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
Move all documentation into EXTRA_DIST, otherwise it gets installed systemwide.
And probably no one is interested in having the README installed in the system.
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
We currently define NDEBUG only if --<set>-debug is not specified at all. That
is, --disable-debug currently has almost no effect.
This fixes this bug and also corrects a wrong parenthesis.
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
This is a first attempt of a freetype2 backend for the font drawing functions.
This avoids the heavy pango/cairo dependencies, although, we might lose
important font handling that pango does for us like correctly drawing combined
characters.
This is still a stub implementation. The drawing operations are not supported,
yet. Use --enable-pango during configure to enable the still working pango
backend.
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
We use a new font factory which is used to create a new font. It will later also
be used to cache fonts and select proper system fonts.
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
Allow converting a ucs4 string into a utf8 string. Also export the default
symbol so other subsystems may use it.
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
Instead of allocating big buffers with kmscon_char's for every character we now
reduce a single character to an integer. This integer is either the Unicode
codepoint or an ID to an internally allocated string.
The so called kmscon_symbol_t integer can now be used as if it were a classic
"char" value without bothering with combining marks etc.
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
Add input subsystem to terminal test app. This currently allows to actually
"write" to the console. There is still much to do to convert the input to proper
data but basic alphanumeric characters work for now.
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
Just a direct mapping to X modifiers for now (= Shift, Control,
Lock, Mod1-5).
This gives us more flexibility to adapt it to our needs in the future.
Also we don't have to include libxkbcommon headers just for the modifier
definitions.
Signed-off-by: Ran Benita <ran234@gmail.com>
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
0 is actually a valid unicode value, so instead we use
KMSCON_INPUT_INVALID which is not legal unicode.
Signed-off-by: Ran Benita <ran234@gmail.com>
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
Just some renames/moving/prefixing to conform to the style of the
other files.
Signed-off-by: Ran Benita <ran234@gmail.com>
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
- Use the number of groups we actually have instead of the maximum
number of groups, when wrapping according to group_wrap control.
- Be careful with negative relative group actions. The group fields of
xkb_state are all unsigned (and have a weird FIXME..), so don't use
them directly.
- Add the grp:ctrl_shift_toggle option to the rmlvo set to also test
locking the previous group (i.e. a negative relative group action).
Signed-off-by: Ran Benita <ran234@gmail.com>
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
We can disable repeats globally or per-key. Also any modifier
keys should not repeat.
Signed-off-by: Ran Benita <ran234@gmail.com>
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
Stuff like grp:ctrl_shift_toggle didn't work because of it.
Signed-off-by: Ran Benita <ran234@gmail.com>
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
This commit introduces a new kmscon_input_event structure which is
delivered to the input consumer. The information provided is
- The evdev scancode (=physical keyboard key).
- The X keysym (a symbol representing what the typist really intended).
- The active modifiers (e.g. Shift, Control..)
- The unicode (UCS-4) value, when it makes sense.
To achieve this we add two dependencies:
- libxkbcommon (this entails xproto and kbproto, but shouldn't bring any of
big X dependencies)
- A function to translate keysyms to unicode values (taken from xlib,
copied in-tree).
libxkbcommon is not an all-around solution, and requires various degrees
of supporting code, depending on how much we wish to support. This
commit implements the common, basic uses:
- Choosing layouts, variants and options
- Switching groups
- Switching shift-levels
- Setting modifiers
Though some stuff is missing. Further information in the comments.
test_input.c is updated to use the new input event interface. You can
change the layouts in input.c:init_input() and see what works and what
doesn't.
Signed-off-by: Ran Benita <ran234@gmail.com>
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
This is currently broken, anyway as we do not unref the object. However, if we
take a reference, then kmscon_input_unref() would never destroy the object as
there are still references. The caller would have to call kmscon_input_sleep()
first to have the desired effect.
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
We either need to adjust our filters to only receive inputX/eventY events or we
simply ignore all other events silently to avoid useless warnings.
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
It should:
- Capture your keypresses
- Announce when a device is added or removed (with --enable-debug)
- stop/start capturing on SIGQUIT (Ctrl-\)
Signed-off-by: Ran Benita <ran234@gmail.com>
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
The diff may look a bit wild but this mostly fixes coding style issues:
- sort includes alphabetically
- add missing includes
- rearrange structures/functions to avoid forward declarations
- adding "data" user-defineable field to input structure (for callbacks)
- correctly handle errors in device_data_arrived
- remove device_added/device_removed
- merge input_init into constructor
- add some log messages for debugging
- remove devices when wake-up fails
- set cb/data fields on eloop connection, not on initialization
- remove all devices when disconnecting eloop so the eloop is guaranteed to be
not used by the input subsystem after disconnection.
- some error handling fixes
- remove some TODOs which are correctly implemented
- Add O_NONBLOCK when opening the device node (I don't know how it actually
worked without it in the data_arrived callback)
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
This does all the boring stuff for an input subsystem:
- Input device discovery using udev
- Input device hotplug (add/remove)
- Reading the evdev data from the devices
- Publishing input events to a callback through the eloop
- Suspending and resuming the input capture
Currently the delivered input events are basically just evdev's
"struct input_event" (type, code, value). More work is required to
have it pass out useful data.
Signed-off-by: Ran Benita <ran234@gmail.com>
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
vte: This adds a dummy VTE subsystem. It will be used to emulate a vt100
terminal based on our console subsystem.
terminal: This ties together several subsystems including the output-, the
console- and the vte-subsystem.
The test_terminal test application can be used to test the terminal. It is
supposed to provide a full vt100 implementation which is displayed on all
available screens.
Most of the functionality here is not implemented yet and only a dummy function.
These will be added in later commits.
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
It doesn't make sense to print the size of new buffer objects as the resize
function does the same. We also currently access uninitialized memory here so it
is useless anyway.
Also move the message of the unref function to be consistent with other objects.
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
We should activate the GL context on initialization to allow other subsystems to
create GL objects.
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
There is no reason to prohibit a compositor from activating its GL context while
being asleep. It is even recommended to activate the GL context when being
asleep as there is no other way to initialize other subsystems without an GL
context.
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
If wakeup fails we should drop master priviledges again, otherwise we might
break other applications.
Also remove the SetMaster call on compositor initialization. It is redundant and
not needed by us.
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
Currently, we set up our VT with SIGUSR1/2 signals and after that the user may
connect the signal handlers to the eloop. However, if we receive a signal in
between, the signal gets lost.
Therefore, this simply merges the kmscon_vt_connect_eloop function into
kmscon_vt_open as there is no obvious reason to split them.
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
If we have only an output but not the related compositor we currently cannot
figure out whether this output is awake. This adds such a helper function.
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
We must not destroy the compositor object if there is still a console using the
GL context. Otherwise we get a SEGFAULT when calling any gl* function.
In future we may need propoper dependencies here. We use reference counts so
this should be easy to implement.
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
