TODO

Here's a evergreen list of TODO items in no particular order:

* Implement support for multiple screens.
* More klasses!
  - Clock
  - Spinner / Throbber
  - Label (done)
  - Text Box (done)
  - Input Field
  - File Picker
  - Button
  - Slider
  - Progress Bar
  - Check Box
  - Frames (done)
* Convert standard event signals from strings to integers for perforance.
* Rewrite KMIO code using protothreads for better code maintenace.
* Shuffle APIs into more logical files.  Decks, Screens, Windows, States
  aren't really arranged in a clear cut manner.
* Support freezing screen.
* valgrind: "Invalid read of size 4" in ncurses' color tree
  (tsearch/tfind/tdelete) reached via vdk_color_init /
  _vdk_color_init_extended at startup, and via delscreen at screen
  teardown.  Investigate whether the extended-color init can avoid the
  freed-block reads, else add a suppression.  (Seen 2026-06-17 from vwm.)
* vkmio: unify the two mouse decoders.  vk_kmio_gpm (GPM) and
  _vk_kmio_parse_sgr (xterm/SGR) each build the ncurses bstate
  independently, so the same mouse semantics live in two places and can
  drift apart (the btn1 drag-desync class of bug).  Collapse onto the
  SGR decoder: have the GPM handler compute the SGR Cb (button = low 2
  bits; motion = 0x20; wheel = 0x40|bit from wdy; shift/ctrl/alt =
  0x04/0x10/0x08; M/m from GPM_DOWN/UP) and call _vk_kmio_parse_sgr
  directly.  It is already a pure body->MEVENT decoder, so do NOT
  re-inject synthesized escape bytes through getch -- pass it the
  encoding, or split it so GPM hands over the integer Cb and skips the
  sscanf.  Wins: one canonical bstate builder (kills the divergence
  bugs); delete vk_gpm.def with its X_GPM tables and the ~50-line
  mapping tail; and drop the GPM cooked-click code (GPM_CLICK_STRICT /
  X_GPM_COOKED) since vk_dblclick already does double-click by timing
  downstream -- nothing uses GPM's native click count.  The mapping is
  clean because both event streams are self-describing (no held-state).
  Caveats: recently stabilized, historically bug-prone code -- do it as
  a spike; only testable on a real console with GPM; perf-neutral or
  better (removes the static tables).  (Noted 2026-06-17.)