turbo-editors/turbo-go · 2026-08-31-terminal-windows.md

⑂ main

2026-08-31-terminal-windows.md · 40 lines · 4.4 KBmarkdown    Blame HistoryRaw
   1# Handoff — 2026-08-31 — terminal windows
5**Terminal windows are done and green.** `F8`, or **Window ▸ New terminal**, opens a window running a real shell in a real pseudo-terminal. Ticket 0007.
7- New package `internal/terminal`: `Session` (pty + process), `Parser` + `Screen` (VT/ANSI emulation, scrollback, alternate screen), `Encode` (keys → terminal bytes), `View` (the `ui` widget). Coverage 95.7 %.
8- Wired into `app` via `terminals.go`; `app` coverage went from ~71 % to 82.8 %.
9- Two theme keys added, `terminal.text` and `terminal.cursor`, set in all three shipped themes. The theme suite fails if a shipped theme misses one, so that is enforced rather than remembered.
10- Whole suite green under `-race` across eight consecutive full runs.
11- Quality gate **PASS**: 0 errors, 0 warnings, 0 smells, complexity 1023.
12- Docs complete in EN and FR — three new pages each, six existing pages updated each. `docs/diagrams/packages.drawio` was regenerated and then **checked against `go list` programmatically**: it matches the real import graph edge for edge.
13- Ticket `0015` created for the Windows/ConPTY port.
15**Nothing has been committed.** The working tree carries the whole feature, the docs, the memory update and the new ticket.
17## In flight
19Nothing. The feature is finished through Phase 8 of `methodical-dev`.
21## Next steps
231. **Commit.** Suggested message and file list are in the session's final summary; `git status` shows `.tickets/`, `.vscode/` and `kits/` as untracked from before this work — decide what belongs in the repository before staging everything.
242. **Try a terminal window on a real terminal.** Everything here was verified against a real pty, but nobody has yet opened one inside a running `turbo-go` on a physical terminal emulator and run `vim` or `htop`. That is the check most likely to find something.
253. **Decide about ticket 0007.** It is the terminal-window ticket and is still `state: open`; closing it was left to the user.
27## Open questions / blockers
29- **Should `F1`…`F12` be reachable inside a terminal?** They are currently reserved by the editor, so `htop`'s function-key menu cannot be used. The reasoning is written up in `docs/*/explanation/terminal-windows.md`; if the user disagrees, the change is one predicate — `editorOwnedKey` in `internal/app/app.go`.
30- **`internal/terminal/pty_darwin.go` has never been run.** It compiles and passes `go vet`, and uses the documented `TIOCPTYGRANT` / `TIOCPTYUNLK` / `TIOCPTYGNAME` ioctls, but this sandbox is Linux. The first Mac run is the test.
31- **The tickets' `tasks:` schema is unknown.** Every existing ticket has `tasks: []`, so `0015`'s checklist went into the `body` rather than into an invented task shape. If IssueSpec defines one, move it.
33## Watch out for
35- **A pseudo-terminal echoes the command line.** A test that types `echo red` and waits for `red` on screen passes *before* the shell has run anything — it matched the echo. Wait for something only the output can produce: a colour, or a string the typed line spells differently (`echo turbo''-go-works` produces `turbo-go-works`). This is the same class of mistake as the gopls-answers-from-disk trap from 2026-08-30.
36- **Never assert on a screen cell while a live shell is writing to it.** A drawing test doing that passed by luck and was hiding a real fault: the cell it sampled, (0,0), is where the *cursor* is drawn on a fresh screen, not the text. `newOfflineView` in `view_test.go` builds a `View` with no session behind it — `Draw` needs none — and is the deterministic way to test drawing.
37- **`tcell.KeyCtrlC` is 67, not 3.** tcell reports a control byte as `KeyCtrlSpace + b`, and `KeyCtrlSpace` is 64. Code testing `key < 0x20` for a control key matches nothing, silently, and every Ctrl-key would send no bytes at all.
38- **The routing order in `App.keyLayers()` is load-bearing.** A focused terminal deliberately sits *above* the global shortcuts. Move it below and `Ctrl-W` closes the window instead of deleting a word — there is a test for exactly that, `TestAFocusedTerminalKeepsTheKeysAShellNeeds`.
39- **`terminal.*` theme keys do not fall back to `editor.*`.** Dotted fallback runs along the dots and stops at `default`. A new theme must set both groups.
40- **`internal/app/README.md`'s file table was stale** — it listed an `actions.go` that has not existed for some time. It is corrected now; the lesson is that a table of files rots silently.

📦 Turbo Go 3d7798b k33g 11h ago	1	# Handoff — 2026-08-31 — terminal windows
	2
	3	## State
	4
	5	Terminal windows are done and green. `F8`, or Window ▸ New terminal, opens a window running a real shell in a real pseudo-terminal. Ticket 0007.
	6
	7	- New package `internal/terminal`: `Session` (pty + process), `Parser` + `Screen` (VT/ANSI emulation, scrollback, alternate screen), `Encode` (keys → terminal bytes), `View` (the `ui` widget). Coverage 95.7 %.
	8	- Wired into `app` via `terminals.go`; `app` coverage went from ~71 % to 82.8 %.
	9	- Two theme keys added, `terminal.text` and `terminal.cursor`, set in all three shipped themes. The theme suite fails if a shipped theme misses one, so that is enforced rather than remembered.
	10	- Whole suite green under `-race` across eight consecutive full runs.
	11	- Quality gate PASS: 0 errors, 0 warnings, 0 smells, complexity 1023.
	12	- Docs complete in EN and FR — three new pages each, six existing pages updated each. `docs/diagrams/packages.drawio` was regenerated and then checked against `go list` programmatically: it matches the real import graph edge for edge.
	13	- Ticket `0015` created for the Windows/ConPTY port.
	14
	15	Nothing has been committed. The working tree carries the whole feature, the docs, the memory update and the new ticket.
	16
	17	## In flight
	18
	19	Nothing. The feature is finished through Phase 8 of `methodical-dev`.
	20
	21	## Next steps
	22
	23	1. Commit. Suggested message and file list are in the session's final summary; `git status` shows `.tickets/`, `.vscode/` and `kits/` as untracked from before this work — decide what belongs in the repository before staging everything.
	24	2. Try a terminal window on a real terminal. Everything here was verified against a real pty, but nobody has yet opened one inside a running `turbo-go` on a physical terminal emulator and run `vim` or `htop`. That is the check most likely to find something.
	25	3. Decide about ticket 0007. It is the terminal-window ticket and is still `state: open`; closing it was left to the user.
	26
	27	## Open questions / blockers
	28
	29	- Should `F1`…`F12` be reachable inside a terminal? They are currently reserved by the editor, so `htop`'s function-key menu cannot be used. The reasoning is written up in `docs/*/explanation/terminal-windows.md`; if the user disagrees, the change is one predicate — `editorOwnedKey` in `internal/app/app.go`.
	30	- `internal/terminal/pty_darwin.go` has never been run. It compiles and passes `go vet`, and uses the documented `TIOCPTYGRANT` / `TIOCPTYUNLK` / `TIOCPTYGNAME` ioctls, but this sandbox is Linux. The first Mac run is the test.
	31	- The tickets' `tasks:` schema is unknown. Every existing ticket has `tasks: []`, so `0015`'s checklist went into the `body` rather than into an invented task shape. If IssueSpec defines one, move it.
	32
	33	## Watch out for
	34
	35	- A pseudo-terminal echoes the command line. A test that types `echo red` and waits for `red` on screen passes before the shell has run anything — it matched the echo. Wait for something only the output can produce: a colour, or a string the typed line spells differently (`echo turbo''-go-works` produces `turbo-go-works`). This is the same class of mistake as the gopls-answers-from-disk trap from 2026-08-30.
	36	- Never assert on a screen cell while a live shell is writing to it. A drawing test doing that passed by luck and was hiding a real fault: the cell it sampled, (0,0), is where the cursor is drawn on a fresh screen, not the text. `newOfflineView` in `view_test.go` builds a `View` with no session behind it — `Draw` needs none — and is the deterministic way to test drawing.
	37	- `tcell.KeyCtrlC` is 67, not 3. tcell reports a control byte as `KeyCtrlSpace + b`, and `KeyCtrlSpace` is 64. Code testing `key < 0x20` for a control key matches nothing, silently, and every Ctrl-key would send no bytes at all.
	38	- The routing order in `App.keyLayers()` is load-bearing. A focused terminal deliberately sits above the global shortcuts. Move it below and `Ctrl-W` closes the window instead of deleting a word — there is a test for exactly that, `TestAFocusedTerminalKeepsTheKeysAShellNeeds`.
	39	- *`terminal.` theme keys do not fall back to `editor.`.* Dotted fallback runs along the dots and stops at `default`. A new theme must set both groups.
	40	- `internal/app/README.md`'s file table was stale — it listed an `actions.go` that has not existed for some time. It is corrected now; the lesson is that a table of files rots silently.