These two fields couldn't have an interesting value at the same time.
As we can no longer rely on a special value to distinguish between
what's the kind, the kind of the key is explicitly encoded in the two
most significative bits of the _flags field.
Extra nice thing: This removes the special values 'EVENT_NONE' and 'CHAR_NONE'.
Two advantages:
- No need to distinguish modifiers in KeyEventHandler. The KeyValue is
enough to decide what action to do.
- Keys are never a Char and Event at the same time, fields can be
merged.
The meaning of the public fields of KeyValue was quite complicated and
not handled consistently accross the app.
Make these fields private and add a more abstract API on top.
The meaning of these fields changed recently and it wasn't an easy
change. I plan on making more changes in the future.
There was no free bits left to add new modifiers. Instead of increasing
the width of the 'flags' field, refactor the way modifiers are
represented and used.
Modifers are now represented as independent values and stored in the
'code' field. A flag is added to distinguish between modifiers and keys
with a key event.
The most notable change is that modifiers can no longer be or-ed into a
single value but have to be represented as an array.
Layouts no longer need to mention every localized keys and dead keys.
They are now placed automatically starting from the second row on the
bottom-right corner.
The "loc " prefix is not removed to still be able to define a more
optimal and consistent placement for some extra keys (eg. 'ß' near 's').
Programming layouts no longer need to place every dead keys.
Each layouts can chose which key are localized instead of specifying it
globally for each key.
Important keys are no longer removed from layouts if the 'extra_keys'
mechanism is not working properly. This can happen if language tags
specified in method.xml don't match the user's language.
Removed some currency symbols from some layouts. They are all in the Fn
layer.
The "closest key" logic must be careful not to reveal keys removed by a
modifier.
Must check [_handler.onPointerSwipe] for every candidate values.
[selected_value] is changed back to [selected_direction].
This adds a new bug: When the direction change, the selected value might
not change but a vibration will be triggered anyway.
getAtDirection was too hard to maintain and might contain bugs.
Change slightly the meaning of directions and implement a the nearest
key calculation as a loop.
Since SDK 21, applications can set the background color of the
navigation bar. This is normally simply an item in a theme but it is
more complicated for keyboards.
When typing fast, a second key might be pressed before the first is
released.
Clearing modifiers earlier would prevent this but would break modifiers
placed in corners (especially the accent keys). Instead, don't take
latched modifiers into account when registering the second press.
A new flag is needed to not interfere with holding modifers, which is
merged with the norepeat flag.
The required version of fontforge (from 2020!) is not available in many
distros. This is an annoying for contributors and greatly complicated
the CI and F-Droid scripts.
The generated font file is now included in the sources. Fontforge is
still needed when adding new glyphs but this is not a common operation.
Requires two new diacritics: ogonek and dot_above.
The new accents are also added to the Latvian layout as the two language
can be close but not to the other localized layouts. A new mechanism is
needed to reproducibly add extra keys to layouts without manual
placement.
Allow specifying a layout for programming and add a key for switching to
it easily.
The switching key is placed on the top edge of the space bar.
The option has no effect by default because the ergonomic isn't ideal,
it needs to be enabled explicitly.
Users of Latin-script languages certainly prefer to use one layout (for
programming or not). This feature might be removed in favor of a better
language-switching mechanisms in the future.
Revert to previous version, the original google material icon,
which is not rounded, and it looks better with the rest of the keyboard,
also more correct to represent the open box unicode space char,
instead of a smile
The previous computation was very different depending on the device's
screen and accessibility options.
Given that the keyboard is supposed to fill all the space in one
dimension, the dpi unit makes little sense.
The formula doesn't mean anything in particular, it takes into account
both dimensions (x and y) and should be close to the dpi in the
diagonals (which is the direction of swipe).
This changes the actual value, on a 1920x1080 480dpi screen, the value
is increased by 20%.
The accents are taken from the Roboto Regular font (Apache 2.0), added
the dotted circle and exported with FontForge.
The argument to FontForge's Import function are changed, the fifth
argument controls the accuracy and needed to be lowered otherwise the
grave accent wouldn't be rendered at all.
When changing any numeric setting and cancelling this change, the change would not
be persisted, but the summary in the setting screen would show the last position
of the slider instead of the correct unchanged value.
This commit fixes this on the most simple way by resetting the slider position to
the persisted value.
Modifiers can temporarily remove a key from the layout by returning
'null'.
Make sure pointer handling code handle these modified keys gracefully
and doesn't trigger a key event and a vibration for the removed key.
