Fixes #142

Mostly useful for testing, where we want the osk to show up reliably
regardless of whether there is a physical keyboard attached. But might
be useful for other use cases too.

Luca Weiss authored 4 years ago and Clayton Craft committed 3 years ago

Fixes #91

Width/height were accidentally swapped when initializing the keyboard
toggle, causing the text position calculation to be wrong and resulting
in the text not being displayed on the toggle surface. Oops.

Previously added destructors fired too late, after the renderer was
destroyed, causing seg faults on musl (though, not on glibc, so it must
be some undefined behavior or ???)

Explicitly calling a cleanup() for these objects to free textures at the
end of the app seems to be working well, no leaks and no
use-after-free/double-free/etc issues. The default destructors for these
objects is used for everything else.

Other objects use TTF_* too, so initializing it in the keyboard object
could introduce some race condition (though I've never seen one first
hand...)

It's still good to do it in a central place. This also calls TTF_Quit()
to clean up.

This is a slight optimization that moves scheduling of a render event so
that it is only scheduled when a repeat character satisfies the repeat
delay, which is the only time a dot will be drawn on the screen for the
character. There's no reason to render things if the key repeats before
the delay since the character is discarded and nothing new is added to
the screen.

The newly added C-u hotkey seems to be generating a text input event in
SDL on some systems (Pinebook Pro), while it did not on other systems
(thinkpad running Arch) in testing the earlier patch.

This returns from the text input handling when the event was fired
because C-u was held down.

This queries input device 'key' capabilities to try and determine if a
physical keyboard is present.

If one is present, it hides the on-screen keyboard.

this is a common binding in applications and helpful when using a
physical keyboard.

When no keyboard is used (-x), the input box is comically large on some
systems (PineBook Pro). This wraps the height of the input box to be
slightly more than the font size. It makes some gross assumptions about
font size and pixels, but seems to work on the two devices I tested on
(librem 5 and PBP)

This option, -x|--no-keyboard, makes osk-sdl behave like a prettier
`cryptsetup open`, with just an input box for devices that have physical
keyboards. The performance of this should be really good when there is
no hw accel support (i.e. using directfb / sw rendering), since it's
only rendering the input box + contents/dots and not the
keyboard/layers/keycaps.

This reduces the render time for a 2 render pass (using sw rendering) by
something close to 50%, as measured on a Librem 5 with SDL2 sw
rendering. RenderClear clears the screen to the wallpaper color, thereby
saving quite a lot of time.

The render while loop was measured as a whole with gettimeofday
(converted to msec):

Before:
        INFO: render loop: 360 ms
        INFO: render loop: 367 ms
        INFO: render loop: 367 ms
        INFO: render loop: 362 ms

After:
        INFO: render loop: 175 ms
        INFO: render loop: 165 ms
        INFO: render loop: 154 ms
        INFO: render loop: 167 ms

The tooltip position is based on the keyboard height, so this draws the
keyboard when animations are disabled *before* tooltips are drawn so
that the tooltip is immediately drawn at the correct location.

When animations are enabled (and, key previews), the keyboard is drawn
*after* the tooltip so that they tooltips don't overlap the key
previews. This is the existing behavior before this change.

Martijn Braam authored 3 years ago and Clayton Craft committed 3 years ago

Added the -o osk.conf option that's loaded after the main config file
and is optional. It can be used to load user overrides from a file of
/boot where it's easier for users to edit than inside the initfs.

There's a cost to calling SDL_GetTicks and it's only used when handling
key repeat events, which is very very rare in the main loop.

DirectFB/framebuffer uses sw rendering and is super slow on most devices
that are trying to run osk-sdl. This disables animations (wavy dots...)
and key preview popups when the renderer is DirectFB to help make
osk-sdl more responsive for devices that don't have hw accel support.

This shows a simple tooltip with a message to the user that the app is
trying to unlock the disk, when the app has animations disabled.

This mechanism was broken, or not really working as intended. The empty
event was ignored by the main loop, and I think what was intended was
for this to trigger a render event (e.g. for displaying an error). This
is necessary when osk-sdl is running without animations, since the
animation loop pushes render events continuously into the queue and this
is will *not* happen when animations are intended to be disabled.

SDL docs recommend using RegisterEvents to get a type rather than just
setting it manually. This also allows us to get a consistent type to
pass to threads, etc for triggering a rendering event from there.

This adds a new tooltip that shows when the app starts, and when no text
is in the input box (including when all text is deleted), informing the
user to input a disk decryption passphrase.

The tooltip disappears as soon as the user starts entering characters,
and the normal password dots replace it in the input box.

I decided to use a tooltip in this way since there was not enough space
on the screen to fit it elsewhere when the screen height is small (e.g.
on the N900), and I think this looks rather nice.

fixes #60

I think this is helpful to have for debug reasons:

	❯ ./_build/osk-sdl -d a -n a -c osk.conf -v
	INFO: osk-sdl v0.62.1
	ERROR: TTF_OpenFont: Couldn't open /usr/share/fonts/ttf-dejavu/DejaVuSans.ttf

fixes #126

The keyboard should manage its own haptic stuff, since keyboard haptic
events are tied to keyboard events.

fixes #125

The GLES stuff doesn't seem to work (with fbdev at least), so this skips
any preference to use GLES when running on DirectFB.

This backend causes SDL to crash when it initializes with haptic
support (SDL uses udev, which isn't available/working in initfs for
directfb for reasons I didn't dig into further).

The key row count was increased in MR !111, as a result it reduced the
calculated keycap size on the keyboard, causing accessibility issues.
The denominator used for calculating the keyboardHeight was chosen by
comparing the calculated key height before !111 with the key height after
!111.

This was tested on a pinephone w/ kmsdrm, and on my laptop with wayland
and various settings for HEIGHT and WIDTH.

Fixes #122

undef authored 4 years ago and Clayton Craft committed 4 years ago

When in keyscript mode, the password should not be considered a log,
as this will both not output anything without verbose mode and output
additional non-password text in verbose mode. Either of these leads to
the user being unable to unlock their disk.

Resolves #120

Johannes Marbach authored 4 years ago and Clayton Craft committed 4 years ago

This change makes osk-sdl use the OpenGL ES driver whenever it is
available. If it is not available, the app falls back to the default
driver.

Additionally, a new command line option -G / --no-gles was added that
allows to opt out of the behavior and fall back to the default driver
from the start.

Closes: #102

This enables haptic feedback when typing on the keyboard. Vibrations are
triggered on touch/mouse down events, which seemed more natural than
triggering it on the up events.

If it's not working, then the module for the haptic/vibration device may
not be enabled or available (e.g. when in initfs)

Closes #95

SDL's manual claims that calling SDL_Quit is safe even if init fails, so
it's set *before* SDL_Init:

> It is safe to call this function even in the case of errors in
> initialization.
https://wiki.libsdl.org/SDL_Quit

fixes #79

This aligns the wrong-password tooltip's position with the input and
applies the same corner rounding (if any).

Johannes Marbach authored 4 years ago and Clayton Craft committed 4 years ago

This avoids repeating the computation in three different spots.

Johannes Marbach authored 4 years ago and Clayton Craft committed 4 years ago

This replaces the three related arguments inputWidth, screenWidth and
y with a single argument for the input rect and rebases all coordinate
computations on the latter.

The benefits are:

- The function signature and call site are simplified
- Computations inside the function are no longer directly based on the
  screen width which makes them more robust against future changes
- The input rect is accessible inside the body and can be used for
  future improvements (like clipping)

Johannes Marbach authored 4 years ago and Clayton Craft committed 4 years ago

This adds key highlighting on click or tap. The highlight colors were
added to the config file. For character keys an additional preview
is shown above the key.