This does a bunch of crazy stuff:
- The ability to set a unicode mode (right now only Linux+ibus or
MacOS-RALT) in the keymap. This will be changeable at runtime soon, to
allow a single keyboard to be able to send table flips and whatever
other crazy stuff on any OS the board is plugged into (something that's
not currently doable on QMK, so yay us?)
- As part of the above, there is now just one user-facing macro for
unicode codepoint submission,
`kmk.common.macros.unicode.unicode_sequence`. Users should never use the
platform-specific macros, partly because they just outright won't work.
There's all sorts of fun stuff in these methods now, thank goodness
MicroPython supports the `yield from` construct.
- Keycode (these should really be renamed Keysym or something) objects
that are intended to not be pressed, or not be released. Right now these
properties are completely ignored if not part of a macro, and it's
probably sane to keep it that way. This was necessary to support MacOS's
"hold RALT while typing the codepoint characters" flow.
- Other refactor-y bits, like moving macro support to `kmk/common`
rather than sitting at the top level of the tree. One day `kmk/common`
may make sense to surface at top level `kmk/`, but that's a discussion
for another day.
Wow, what a trip this was. Layer support is now fully implemented. Other
changes here mostly revolve around the event dispatching model: more
floating state (hidden in clases wherever) has been purged, with the
reducer (now mutable, comments inline) serving, as it should, as the
sole source of truth. Thunk support has been added to our fake Redux
clone, allowing Action Creators to handle sequences of events (which is
arguably a cleaner way of handling matrix changes when not all matrix
changes should result in a new HID report - in the case of internal
keys). A whole class has been deprecated (Keymap) which only served as
another arbitor of state: instead, the MatrixScanner has been made
smarter and handles diffing internally, dispatching an Action when
needed (and allowing the reducer to parse the keymap and figure out what
key is pressed - this is the infinitely cleaner solution when layers
come into play).
