Conversation
…ro-editor` With the fork we can still use raw escape sequences until something similar is present upstream. Adapt to the style and clipboard interface changes: - replace Decompose() with dedicated getters - replace attributes with dedicated getters - replace underline attribute with new interface We have to update our required go toolchain version to 1.25 due to the fact that tcell updated its toolchain dependency to 1.25.
Several small API adaptations in the screen wrapper:
- Screen.SetPaste(bool) is replaced by Screen.EnablePaste() and
Screen.DisablePaste().
- Screen.CanDisplay is no longer exposed; drop the pass-through.
- Screen.RegisterRawSeq / UnregisterRawSeq are gone. The v3 input
processor handles the CSI-u keyboard protocol natively, covering
the cases raw-seq registration was used for.
- Add InitMockScreen helper for tests that migrated from the
removed SimulationScreen to vt.MockTerm.
Assisted-by: Claude:claude-opus-4-7
- EventKey.Rune() is replaced by Str(). Call sites take the first
rune where a scalar is still required. Multi-rune grapheme
clusters get truncated to their first rune on the keystroke path;
the paste path handles wide input via accumulation (below).
- NewEventKey takes a string instead of a rune.
- EventPaste no longer carries payload text. Between Start and End
the v3 input processor streams EventKey events, so bufpane
accumulates them into a buffer and replays the whole block
through paste() on End. This preserves single-undo and
smartpaste behaviour for bracketed pastes and prevents
plugin-driven autoclose etc. from firing per-character.
- command.go: SetPaste(bool) call is split into EnablePaste /
DisablePaste.
- events.go: drop the *EventRaw branch from ConstructEvent and
eventEscSeq. CSI-u covers the original use case natively.
- rawpane: the EscSeq accessor is no longer a method on the event
type; use the internal eventEscSeq helper.
- bindings.go: tcell v3 dropped the KeyCtrlSpace / KeyCtrlLeftSq /
KeyCtrlBackslash / KeyCtrlRightSq / KeyCtrlCarat /
KeyCtrlUnderscore constants, and its CSI-u disambiguate path
delivers Ctrl+Shift+letter, Ctrl+Alt+letter, and Ctrl+non-letter
as KeyRune+str+mod rather than the KeyCtrlX form. Add
ctrlNameAliases for the long-form names, gate the KeyCtrlX
return on mod == ModCtrl exactly to mirror tcell's NewEventKey
fold (key.go:282-291), and lowercase the rune in the
single-character emission so the unshifted base kitty sends
matches bindings written as CtrlShiftP etc. Plain Ctrl+letter
keeps the legacy KeyCtrlX path. CtrlLeftSq parses but does not
fire because tcell consumes 0x1B as the start of an escape
sequence, matching v2 behaviour.
- tcell v3 drops KeyBackspace2 and only populates KeyBackspace
Assisted-by: Claude:claude-opus-4-7
action tcell
- tcell v3 exposes an EventQ() channel directly. The PollEvent
goroutine micro used to plumb events into its own channel is no
longer needed; select on Screen.EventQ() alongside the existing
timer / redraw channels.
- EventError.Err() is gone. Use errors.Is(e, io.EOF) to distinguish
clean EOF from real errors.
Assisted-by: Claude:claude-opus-4-7
- Mirror the bracketed-paste accumulator on the termpane side.
Between EventPaste{Start} and EventPaste{End} the intervening
EventKeys are buffered together with \x1b[200~ / \x1b[201~
framing and the whole frame is written to the child pty in one
go. This preserves bracketed-paste semantics for programs
running inside termpane (so e.g. bash skips alias expansion on
the pasted text).
- When forwarding keystrokes to the pty, translate CSI-u
Ctrl+letter events (KeyRune + ModCtrl + single-letter Str) back
to the legacy single-byte C0 control. Child shells and readline
speak the legacy byte protocol regardless of the outer
terminal's keyboard encoding.
- Use the eventEscSeq() helper for raw key forwarding now that
EscSeq is not a method on the event type.
Assisted-by: Claude:claude-opus-4-7
tcell v3 drops SimulationScreen. The package's vt.MockTerm is the closest replacement: it satisfies tcell.Screen with an in-memory cell grid and event queue, so the existing test harness adapts with minimal changes. PostEvent(NewEventResize(...)) replaces the old InjectResize helper. Assisted-by: Claude:claude-opus-4-7
tcell v3 removes the legacy KeyCtrlSpace / KeyCtrlLeftSq /
KeyCtrlBackslash / KeyCtrlRightSq / KeyCtrlCarat / KeyCtrlUnderscore
constants. Micro's bindings parser previously accepted the matching
string names ("CtrlSpace", "CtrlLeftSq", ...) as aliases; those are
now invalid and the parser emits a TermMessage warning when it sees
one.
Update the shipped comment plugin and the keybindings help page to
use the supported forms.
Assisted-by: Claude:claude-opus-4-7
|
Currently I'm unsure if this should be added within @Andriamanitra: |
|
I just tested all the bindings in my bindings.json (and some random ones) in micro versions before and after this PR. Most notable change is that bunch of key combinations that didn't get parsed correctly before just started working (for example Ctrli, Ctrlm, modifiers with non-ascii letters). There were however a few key combinations that were emitting valid events before that now show up as a different event in the raw event viewer:
I assume the results will vary a bit depending on OS/terminal emulator/keyboard layout, but on my setup (Linux/Ghostty/Nordic kbd layout) the breakage doesn't seem to be too bad. Footnotes |
According to
In general it is now more consistent to receive all modifiers instead of special key combinations. micro/runtime/plugins/comment/comment.lua Lines 219 to 220 in 1c8b82b to micro/runtime/plugins/comment/comment.lua Lines 219 to 220 in d6daf2a via f5e19bb A 100% downward compatibility is quite hard here. 🤔
This case seems to be terminal dependent, because I still receive |
3 participants
This work was initially done by @Tubbles and Claude. 😉
Many thanks for this preparation!
With this we are finally able to have proper pastes in the Windows Terminal App.
TODOs:
Related to micro-editor/tcell#5.
Fixes #3571
Fixes #4173