render: editable CheckBox, ChoicePicker, Slider, and DateTimeInput (finish #15)

[joestump]

What

Finishes the editing work #15 started: the four remaining input components are now focusable and editable, and their edited values round-trip to the agent in a Button's ActionEvent.Context.

• CheckBox — joins the focus ring; Space/Enter toggles. Toggling a bound value replaces the {binding} placeholder with real local state.
• ChoicePicker — Up/Down move a highlight (clamped, no wrap), Space toggles the highlighted option. multipleSelection toggles membership; single-select replaces the selection (radio semantics). The edited value is a []string normalized to option-declaration order, honoring the native DynamicStringList shape.
• Slider — Left/Right step within [min, max]. The v0.9 schema has no step field, so spans >= 1 step by 1 and smaller spans step by span/16 (one bar cell), keeping fractional sliders adjustable. Bound values seed from min.
• DateTimeInput — reuses the TextField rune-edit path against its string value (appendText/deleteRune/editSeed), including cleared-to-empty "(unset)" semantics. No separate editor was needed.

How

Each component mirrors the existing hand-rolled TextField pattern in render/render.go rather than importing bubbles models: edits live in lazily initialized maps on Surface (checkValues, choiceValues, sliderValues, plus choiceCursor for the picker highlight; DateTimeInput shares fieldValues) and shadow the static literal for both rendering and value readout. The new edit paths live in render/inputs.go; collectFocusables now admits all input kinds via an isInteractive helper; gatherFieldValues reads edited state first so dispatch-time context gathering carries bool/[]string/float64/string values (Slider and DateTimeInput literals are now also reported when unedited, matching the other kinds). deleteSurface clears the new state; edits survive Apply merges like TextField edits do.

Chrome stays monochrome: focused inputs mark their value line with the same "▎" cue as the TextField cursor, and a focused picker draws a "▏" gutter with "▎" on the highlighted option. Idle rendering is byte-identical to before. The stale comment at render/fields.go:4 and the package/event docs are updated.

The activateButtonCtx test helper previously pressed Enter on every focusable while hunting for the button; Enter now toggles a focused CheckBox, so it tabs until focus is on an actual Button before activating.

Tests

go build ./... && go vet ./... && go test ./... green. New render/inputs_test.go (17 tests) follows the textfield_test.go patterns: focus-ring membership and order, toggle/highlight/step behavior with live re-render assertions, clamping at bounds and degenerate ranges, binding-placeholder shadowing for CheckBox and Slider, fractional-span stepping, DateTimeInput rune editing and clear-to-"(unset)", Space's dual role (toggle vs. text input), an end-to-end check that Button activation emits a ClientMessage whose Context carries all four edited value types, and edit survival across Apply merges.

Notes

• The website/docs/wire-format.md "Not yet" item ("Editing beyond TextField") from the Done-when is deliberately not touched here — website/ docs are handled in the dedicated post-merge docs pass.

Fixes #42

🤖 Posted on behalf of @joestump by Claude.

[joestump]

Adversarial review against the #42 Done-when. Verified locally at a914437: go build ./... && go vet ./... && go test ./... green, worktree clean.

Verdict: no blockers

What I actively hunted and confirmed correct:

• Key routing — traced Key.String() through bubbletea v2 → ultraviolet: space genuinely reports "space" (never " "), so the explicit case "space" in render/render.go:259 is required and correct, and its dual role holds: toggle on CheckBox/ChoicePicker, literal " " insertion on TextField/DateTimeInput (TestSpaceStillInsertsIntoTextField guards this). Enter still activates buttons first (render/render.go:253) and only then falls to the CheckBox toggle; arrows only act on the matching focused component, so nothing steals keys from text editing.
• State across Apply merges — applyComponents (render/composite.go:71) never touches the edit maps and re-anchors focus by ID; DeleteSurface (render/composite.go:49-52) clears all four new maps. Ran extra probes beyond the PR's tests: options shrinking under a live cursor (cursor clamps at read time in pickerCursor, no panic, next Space toggles a valid option), zero-option picker (guarded, no panic), slider literal outside [min,max] (first step clamps into range).
• ChoicePicker multi-value — edited value is a real []string normalized to option-declaration order (togglePickerOption, render/inputs.go:64-88); single-select is radio semantics; TestButtonContextCarriesEditedInputValues asserts the ClientMessage.Context types end-to-end (bool / []string / float64 / string).
• Nil/missing guards — every edit path re-resolves s.byID[id] and checks the component kind; focusedComponent returns a zero Component on an empty ring; degenerate slider range (max <= min) refuses to step.
• Idle-render stability — unedited/unfocused output is byte-identical: the new slider readout (strconv.FormatFloat(_, 'f', -1, 64), render/fields.go:191) matches dynNumString's literal branch exactly, and all focus cues are gated on isFocused.
• Test quality — the new tests assert rendered content, FieldValues, and dispatched Context, not mere execution; the activateButtonCtx rewrite (render/context_test.go:30-56) correctly stops blind-Enter from corrupting CheckBox state.
• Ticket compliance — all Done-when items met except website/docs/wire-format.md, which is correctly deferred to the post-merge docs pass per repo policy and noted in the PR body. render/fields.go:4 comment updated.

Nits (non-blocking)

1. Stale picker selection can round-trip after a merge — verified by probe: select "a", then an updateComponents replaces the picker's options with only "x"; FieldValues/Context still report ["a"], a value naming no current option. This mirrors the established TextField edit-shadowing semantics and self-heals on the next toggle (normalization drops unknown values), but a sentence in the choiceValues doc comment (render/render.go:1129) acknowledging it would help the next reader.
2. render/render.go:480 — the appendText doc comment wraps awkwardly after the edit: "…on first edit. On the first edit," reads as an accidental duplication; reflow the paragraph.
3. render/context.go:76 — the Slider case reads c.Slider.Value.Literal inline while every sibling kind goes through a resolveDynamic* helper; a two-line resolveDynamicNumber would keep the pattern uniform.

🤖 Posted on behalf of @joestump by Claude.