@devries-cameron20 / t3code Mirror · read-only / commits / 1001bc71fc
Fork 0

The best way to control Claude Code, Codex and more from any computer or phone

Mirrored from https://github.com/Camo67/t3code.git

Synced Nov 21, 19:45 UTC

fix(voice): dictation survives draft changes and Escape in the citation box (#32682)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

Julius Marminge committed Nov 12, 2026 at 11:35 UTC 1001bc71fcc80ba8dc32c60a0ae4bbe85efea44b
19 files changed +350 -159
apps/mobile/src/features/settings/SettingsScheduledTasksRouteScreen.tsx
-4
@@ -87,9 +87,7 @@ import { useScheduledTaskDestinationLabel, useScheduledTaskEditor } from "./sche
87 import {
88 discardScheduledTaskEditor,
89 markScheduledTaskEditorSaved,
90 - scheduledTaskEditorSessionAtom,
90 } from "./scheduled-task-editor-state";
92 -import { appAtomRegistry } from "../../state/atom-registry";
91 import {
92 formatNextScheduledTaskRun,
93 formatScheduledTaskInterval,
@@ -519,8 +517,6 @@ function SettingsScheduledTaskEditorScreen({ title }: { readonly title: string }
517 const voiceInput = useVoiceInputController({
518 ownerKey,
519 label: editor?.draft.title.trim() || title,
522 - subscribeToDraftChanges: (onChange) =>
523 - appAtomRegistry.subscribe(scheduledTaskEditorSessionAtom, onChange),
520 selection,
521 readDraftMessage: () => {
522 const current = readEditor();
apps/mobile/src/features/threads/ThreadComposer.tsx
-3
@@ -57,13 +57,11 @@ import { armAgentAwarenessLiveActivityForLocalWork } from "../agent-awareness/re
57 import { scopedThreadKey } from "../../lib/scopedEntities";
58 import {
59 getComposerDraftSnapshot,
60 - composerDraftsAtom,
60 setComposerDraftText,
61 composerContextImportsAtom,
62 countComposerDraftAttachmentsAfterSelection,
63 useComposerDraft,
64 } from "../../state/use-composer-drafts";
66 -import { appAtomRegistry } from "../../state/atom-registry";
65 import type { ComposerDocumentAttachment } from "../../lib/composerContext";
66 import { useProject, useVisibleThreadShells } from "../../state/entities";
67 import { useMentionableEnvironments } from "../../state/environments";
@@ -579,7 +577,6 @@ export const ThreadComposer = function ThreadComposer(props: ThreadComposerProps
577 ownerKey: composerDraftKey,
578 label: props.selectedThread.title || "Untitled thread",
579 readDraftMessage: () => getComposerDraftSnapshot(composerDraftKey).text,
582 - subscribeToDraftChanges: (onChange) => appAtomRegistry.subscribe(composerDraftsAtom, onChange),
580 selection: composerMenu.selection,
581 onChangeDraftMessage: (text) => setComposerDraftText(composerDraftKey, text),
582 onChangeSelection: composerMenu.onSelectionChange,
apps/mobile/src/features/voice-input/lazyVoiceRecorder.test.ts
-1
@@ -143,7 +143,6 @@ describe("createLazyVoiceRecorder", () => {
143 ownerKey: "thread",
144 text: "",
145 selection: { start: 0, end: 0 },
146 - revision: 0,
146 }),
147 commitDraft: () => undefined,
148 onStateChange: () => undefined,
apps/mobile/src/features/voice-input/useVoiceInputController.ts
-2
@@ -15,7 +15,6 @@ export function useVoiceInputController(input: {
15 readonly ownerKey: string | null;
16 readonly label: string;
17 readonly readDraftMessage: () => string | null;
18 - readonly subscribeToDraftChanges: (onChange: () => void) => () => void;
18 readonly selection: ComposerEditorSelection;
19 readonly disabled?: boolean;
20 readonly onChangeDraftMessage: (value: string) => void;
@@ -58,7 +57,6 @@ export function useVoiceInputController(input: {
57 }
58 },
59 captured.selection,
61 - captured.subscribeToDraftChanges,
60 ),
61 label: captured.label,
62 });
apps/mobile/src/features/voice-input/voiceInputSession.test.ts
+40 -54
@@ -1,6 +1,5 @@
1 import { beforeEach, describe, expect, it, vi } from "vite-plus/test";
2 import type { PreparedVoiceTranscription } from "@t3tools/client-runtime/voice-input";
3 -import { Atom, AtomRegistry } from "effect/reactivity";
3 import { resetVoiceInputGlobalsForTests } from "../../../../../packages/client-runtime/src/voice-input/controller";
4 import {
5 acquireAudioSession,
@@ -23,19 +22,9 @@ import {
22 createVoiceInputTarget,
23 VoiceInputSession,
24 type VoiceInputSessionDependencies,
26 - type VoiceInputTarget,
25 type VoiceRecordingSample,
26 } from "./voiceInputSession";
27
30 -function createTarget(
31 - ownerKey: string,
32 - readText: () => string | null,
33 - commit: VoiceInputTarget["commitDraft"],
34 - selection: { start: number; end: number },
35 -) {
36 - return createVoiceInputTarget(ownerKey, readText, commit, selection, () => () => {});
37 -}
38 -
28 function createSession(
29 overrides?: Partial<
30 Pick<
@@ -116,7 +105,7 @@ describe("global voice input", () => {
105 if (outcome === "activation failure") {
106 audio.setActive.mockRejectedValueOnce(new Error("Audio activation failed"));
107 }
119 - await session.start(createTarget("first", () => "", vi.fn(), { start: 0, end: 0 }));
108 + await session.start(createVoiceInputTarget("first", () => "", vi.fn(), { start: 0, end: 0 }));
109 expect(pause).toHaveBeenCalledOnce();
110 expect(isAudioSessionOwner(playback)).toBe(false);
111 if (outcome === "complete") {
@@ -154,9 +143,9 @@ describe("global voice input", () => {
143 async ({ selection, expected, cursor }) => {
144 const { session } = createSession();
145 const commit = vi.fn();
157 - await session.start(createTarget("first", () => "hello world", commit, selection));
146 + await session.start(createVoiceInputTarget("first", () => "hello world", commit, selection));
147 await session.start(
159 - createTarget("second", () => "other prompt", vi.fn(), { start: 12, end: 12 }),
148 + createVoiceInputTarget("second", () => "other prompt", vi.fn(), { start: 12, end: 12 }),
149 );
150 await session.controller.stop();
151 expect(commit).toHaveBeenCalledWith(expected, {
@@ -175,7 +164,7 @@ describe("global voice input", () => {
164 let visibleDraft = "first";
165 const targetKey = visibleDraft;
166 await session.start(
178 - createTarget(
167 + createVoiceInputTarget(
168 targetKey,
169 () => drafts.get(targetKey) ?? null,
170 (text) => drafts.set(targetKey, text),
@@ -207,7 +196,7 @@ describe("global voice input", () => {
196 const firstCommit = vi.fn();
197 const secondCommit = vi.fn();
198 const starting = session.start(
210 - createTarget("first", () => "first", firstCommit, { start: 5, end: 5 }),
199 + createVoiceInputTarget("first", () => "first", firstCommit, { start: 5, end: 5 }),
200 );
201 await preparationEntered.promise;
202 let stopping: Promise<void> | null = null;
@@ -226,7 +215,7 @@ describe("global voice input", () => {
215 await transcriptionEntered.promise;
216 }
217 await session.start(
229 - createTarget("second", () => "second", secondCommit, { start: 6, end: 6 }),
218 + createVoiceInputTarget("second", () => "second", secondCommit, { start: 6, end: 6 }),
219 );
220 expect(session.ownerKey).toBe("first");
221 expect(session.controller.currentState.phase).toBe(phase);
@@ -247,7 +236,9 @@ describe("global voice input", () => {
236 const { session } = createSession();
237 let text: string | null = "first";
238 const commit = vi.fn();
250 - await session.start(createTarget("first", () => text, commit, { start: 5, end: 5 }));
239 + await session.start(
240 + createVoiceInputTarget("first", () => text, commit, { start: 5, end: 5 }),
241 + );
242 text = change === "removed" ? null : "edited prompt";
243 await session.controller.stop();
244 expect(commit).not.toHaveBeenCalled();
@@ -255,10 +246,26 @@ describe("global voice input", () => {
246 },
247 );
248
249 + it("retries a failed start against the draft as edited since", async () => {
250 + const { session, prepare } = createSession();
251 + prepare.mockRejectedValueOnce(new Error("model unavailable"));
252 + let text = "first";
253 + const commit = vi.fn();
254 + await session.start(createVoiceInputTarget("first", () => text, commit, { start: 5, end: 5 }));
255 + expect(session.controller.currentState.phase).toBe("error");
256 + text = "first draft";
257 + await session.retry();
258 + await session.controller.stop();
259 + expect(prepare).toHaveBeenCalledTimes(2);
260 + expect(commit).toHaveBeenCalledWith("first spoken text draft", { start: 17, end: 17 });
261 + });
262 +
263 it("finishes the original draft at the recording limit while it is off screen", async () => {
264 const { session, recorder } = createSession();
265 const commit = vi.fn();
261 - await session.start(createTarget("first", () => "first", commit, { start: 5, end: 5 }));
266 + await session.start(
267 + createVoiceInputTarget("first", () => "first", commit, { start: 5, end: 5 }),
268 + );
269 await session.controller.handleRecorderStatus({
270 isFinished: true,
271 hasError: false,
@@ -269,41 +276,14 @@ describe("global voice input", () => {
276 expect(session.controller.currentState.phase).toBe("idle");
277 });
278
272 - it("rejects a transcript when its off-screen draft changes and returns to the original text", async () => {
273 - const registry = AtomRegistry.make();
274 - const draft = Atom.make("hello world");
275 - const unsubscribe = vi.fn();
276 - const { session } = createSession();
277 - const commit = vi.fn();
278 - const target = createVoiceInputTarget(
279 - "first",
280 - () => registry.get(draft),
281 - commit,
282 - { start: 6, end: 6 },
283 - (onChange) => {
284 - const stop = registry.subscribe(draft, onChange);
285 - return () => {
286 - stop();
287 - unsubscribe();
288 - };
289 - },
290 - );
291 - await session.start(target);
292 - registry.set(draft, "changed");
293 - registry.set(draft, "hello world");
294 - await session.controller.stop();
295 - expect(commit).not.toHaveBeenCalled();
296 - expect(session.controller.currentState.error).toContain("draft changed");
297 - expect(unsubscribe).toHaveBeenCalledTimes(1);
298 - registry.dispose();
299 - });
300 -
279 it("stops recording when its queued edit is discarded", async () => {
280 const { session, recorder } = createSession();
281 const commit = vi.fn();
282 let draft: string | null = "queued prompt";
283 const ownerKey = "thread~queued-edit~run";
306 - await session.start(createTarget(ownerKey, () => draft, commit, { start: 13, end: 13 }));
284 + await session.start(
285 + createVoiceInputTarget(ownerKey, () => draft, commit, { start: 13, end: 13 }),
286 + );
287 session.cancel(ownerKey);
288 draft = null;
289 await session.controller.stop();
@@ -319,7 +299,9 @@ describe("global voice input", () => {
299 it("keeps another prompt's recording when a queued edit is discarded", async () => {
300 const { session, recorder } = createSession();
301 const commit = vi.fn();
322 - await session.start(createTarget("other prompt", () => "hello", commit, { start: 5, end: 5 }));
302 + await session.start(
303 + createVoiceInputTarget("other prompt", () => "hello", commit, { start: 5, end: 5 }),
304 + );
305 session.cancel("thread~queued-edit~run");
306 expect(recorder.stop).not.toHaveBeenCalled();
307 expect(session.controller.currentState.phase).toBe("recording");
@@ -338,12 +320,12 @@ describe("global voice input", () => {
320 const oldCommit = vi.fn();
321 const nextCommit = vi.fn();
322 const firstStart = session.start(
341 - createTarget("first", () => "first", oldCommit, { start: 5, end: 5 }),
323 + createVoiceInputTarget("first", () => "first", oldCommit, { start: 5, end: 5 }),
324 );
325 await preparationEntered.promise;
326 session.cancel("first");
327 const nextStart = session.start(
346 - createTarget("second", () => "second", nextCommit, { start: 6, end: 6 }),
328 + createVoiceInputTarget("second", () => "second", nextCommit, { start: 6, end: 6 }),
329 );
330 expect(prepare).toHaveBeenCalledTimes(1);
331 expect(recorder.record).not.toHaveBeenCalled();
@@ -366,7 +348,9 @@ describe("global voice input", () => {
348 const snapshots: number[] = [];
349 session.subscribe(() => snapshots.push(session.getSnapshot().elapsedSeconds));
350
369 - await session.start(createTarget("first", () => "first", vi.fn(), { start: 5, end: 5 }));
351 + await session.start(
352 + createVoiceInputTarget("first", () => "first", vi.fn(), { start: 5, end: 5 }),
353 + );
354 expect(session.getSnapshot().state.phase).toBe("recording");
355 expect(keepAwake).toHaveBeenCalledTimes(1);
356 // Silence on a silent history publishes nothing.
@@ -388,7 +372,9 @@ describe("global voice input", () => {
372 expect(session.getSnapshot().elapsedSeconds).toBe(2);
373
374 sample = { durationMillis: 0, metering: -100 };
391 - await session.start(createTarget("first", () => "first", vi.fn(), { start: 5, end: 5 }));
375 + await session.start(
376 + createVoiceInputTarget("first", () => "first", vi.fn(), { start: 5, end: 5 }),
377 + );
378 expect(session.getSnapshot().elapsedSeconds).toBe(0);
379 expect(publishAudioLevels).toHaveBeenLastCalledWith(expect.not.arrayContaining([1]));
380 expect(keepAwake).toHaveBeenCalledTimes(2);
apps/mobile/src/features/voice-input/voiceInputSession.ts
+8 -24
@@ -19,7 +19,6 @@ export type VoiceInputTarget = {
19 readonly label?: string;
20 readonly readDraft: () => VoiceDraftSnapshot | null;
21 readonly commitDraft: VoiceInputControllerDependencies["commitDraft"];
22 - readonly subscribe: () => () => void;
22 };
23
24 export type VoiceInputSnapshot = {
@@ -52,27 +51,14 @@ export function createVoiceInputTarget(
51 readText: () => string | null,
52 commitDraft: VoiceInputTarget["commitDraft"],
53 selection: VoiceDraftSnapshot["selection"],
55 - subscribeToChanges: (onChange: () => void) => () => void,
54 ): VoiceInputTarget {
57 - let revision = 0;
55 return {
56 ownerKey,
57 readDraft: () => {
58 const text = readText();
62 - if (text === null) return null;
63 - return { ownerKey, text, selection, revision };
59 + return text === null ? null : { ownerKey, text, selection };
60 },
61 commitDraft,
66 - subscribe: () => {
67 - let previousText = readText();
68 - return subscribeToChanges(() => {
69 - const text = readText();
70 - if (text !== previousText) {
71 - previousText = text;
72 - revision += 1;
73 - }
74 - });
75 - },
62 };
63 }
64
@@ -84,7 +70,7 @@ export class VoiceInputSession {
70 readonly controller: VoiceInputController;
71 private readonly dependencies: VoiceInputSessionDependencies;
72 private target: VoiceInputTarget | null = null;
87 - private unsubscribe: (() => void) | null = null;
73 + private startText: string | null = null;
74 private snapshot: VoiceInputSnapshot = {
75 state: IDLE_STATE,
76 ownerKey: null,
@@ -102,7 +88,11 @@ export class VoiceInputSession {
88 this.controller = new VoiceInputController({
89 ...dependencies,
90 onStateChange: (state) => this.handleStateChange(state),
105 - readDraft: () => this.target?.readDraft() ?? null,
91 + // The target's selection is fixed at start, so a draft changed since refuses the transcript.
92 + readDraft: () => {
93 + const draft = this.target?.readDraft() ?? null;
94 + return draft?.text === this.startText ? draft : null;
95 + },
96 commitDraft: (text, selection) => this.target?.commitDraft(text, selection),
97 });
98 }
@@ -141,23 +131,17 @@ export class VoiceInputSession {
131 start(target: VoiceInputTarget): Promise<void> {
132 if (voiceInputBlocksSubmission(this.controller.currentState)) return Promise.resolve();
133 this.target = target;
144 - this.unsubscribe = target.subscribe();
134 + this.startText = target.readDraft()?.text ?? null;
135 return this.controller.start();
136 }
137
138 dispose(): void {
139 this.controller.dispose();
140 this.stopRecordingActivity();
151 - this.unsubscribe?.();
152 - this.unsubscribe = null;
141 this.listeners.clear();
142 }
143
144 private handleStateChange(state: VoiceInputState): void {
157 - if (!voiceInputBlocksSubmission(state)) {
158 - this.unsubscribe?.();
159 - this.unsubscribe = null;
160 - }
145 const previousPhase = this.snapshot.state.phase;
146 let elapsedSeconds = this.snapshot.elapsedSeconds;
147 if (state.phase !== previousPhase) {
apps/web/src/components/ComposerPromptEditorTiptap.citationDraft.browser.test.tsx
+26 -1
@@ -31,10 +31,12 @@ const promptBefore = `See ${serializeAssistantCitation(citation)} `;
31 function Composer({
32 question,
33 editorRef,
34 + disabled = false,
35 send = "clears",
36 }: {
37 question: boolean;
38 editorRef: React.RefObject<ComposerPromptEditorHandle | null>;
39 + disabled?: boolean;
40 send?: "clears" | "blocked" | "refusedLater";
41 }) {
42 const [prompt, setPrompt] = useState(promptBefore);
@@ -57,7 +59,7 @@ function Composer({
59 cursor={0}
60 contextRecords={new Map()}
61 skills={[]}
60 - disabled={false}
62 + disabled={disabled}
63 placeholder=""
64 onChange={(next) => (question ? setAnswer(next) : setPrompt(next))}
65 onCitationDraftSave={(replacedValue, savedValue) => {
@@ -159,3 +161,26 @@ it.each(["blocked", "refusedLater"] as const)(
161 .toHaveAttribute("data-prompt", `See ${serializeAssistantCitation(citation)}x `);
162 },
163 );
164 +
165 +it("cancels a new citation while dictation makes the composer read-only", async () => {
166 + const editorRef = createRef<ComposerPromptEditorHandle>();
167 + await using screen = await render(<Composer question={false} editorRef={editorRef} disabled />);
168 +
169 + const sourceAnchor = findAssistantCitationSourceAnchor(document, citation);
170 + if (!sourceAnchor) throw new Error("Citation source did not resolve");
171 + await act(async () =>
172 + editorRef.current?.requestCitationComment({
173 + previousValue: "See ",
174 + value: promptBefore,
175 + citationStart: "See ".length,
176 + sourceAnchor,
177 + insertedSpaces: { before: false, after: true },
178 + }),
179 + );
180 + await expect.element(page.getByLabelText("Comment on selected text")).toBeVisible();
181 +
182 + await page.getByRole("button", { name: "Cancel" }).click();
183 +
184 + await expect.element(page.getByLabelText("Comment on selected text")).not.toBeInTheDocument();
185 + expect(screen.container.querySelector("output")?.dataset.prompt).toBe("See ");
186 +});
apps/web/src/components/ComposerPromptEditorTiptap.tsx
+4 -2
@@ -468,7 +468,7 @@ function ComposerCitationNodeView({ node, editor, getPos }: NodeViewProps) {
468
469 const removeOnCancel = commentTarget?.removeOnCancel;
470 const onRemove = () => {
471 - if (!editor.isEditable || !removeOnCancel) return;
471 + if (!removeOnCancel) return;
472 const pos = nodePos();
473 if (pos === null) return;
474 const { doc } = editor.state;
@@ -477,7 +477,9 @@ function ComposerCitationNodeView({ node, editor, getPos }: NodeViewProps) {
477 const end = pos + current.nodeSize;
478 const from = removeOnCancel.before && doc.textBetween(pos - 1, pos) === " " ? pos - 1 : pos;
479 const to = removeOnCancel.after && doc.textBetween(end, end + 1) === " " ? end + 1 : end;
480 - editor.chain().focus().deleteRange({ from, to }).run();
480 + // A read-only composer (dictation) can still receive a new citation, so Cancel must undo it.
481 + const chain = editor.chain();
482 + (editor.isEditable ? chain.focus() : chain).deleteRange({ from, to }).run();
483 };
484
485 const sentRef = useRef(false);
apps/web/src/components/chat/ChatComposer.tsx
-1
@@ -1810,7 +1810,6 @@ export const ChatComposer = memo(function ChatComposer(props: ChatComposerProps)
1810 projectId: activeProjectId,
1811 projectName,
1812 ownerKey: questionAttachmentTarget ?? composerDraftTargetKey,
1813 - draftText: prompt,
1813 disabled: dictationDisabled,
1814 readDraft: () => {
1815 const snapshot = readComposerSnapshot();
apps/web/src/components/preview/PreviewAnnotationSpeech.tsx
+1 -5
@@ -5,7 +5,7 @@ import type {
5 ScopedThreadRef,
6 } from "@t3tools/contracts";
7 import { useNavigate } from "@tanstack/react-router";
8 -import { useEffect, useRef, useState } from "react";
8 +import { useEffect, useRef } from "react";
9
10 import { resolveSpeechPresentation } from "~/components/chat/ComposerSpeechButton";
11 import { VoiceInputSetup } from "~/components/chat/VoiceInputSetup";
@@ -40,7 +40,6 @@ export function PreviewAnnotationSpeech({
40 projectId ? scopeProjectRef(threadRef.environmentId, projectId) : null,
41 );
42 const draft = useRef({ text: "", cursor: 0, selectionEnd: 0 });
43 - const [text, setText] = useState("");
43 const keys = useRef<DictationKeyHandlers | null>(null);
44 const guestFocused = useRef(false);
45 // oxlint-disable-next-line t3code/no-unreasoned-state-sync -- notices focus returning from the preview guest to this window
@@ -60,7 +59,6 @@ export function PreviewAnnotationSpeech({
59 projectId,
60 projectName: project?.title,
61 ownerKey: JSON.stringify(["preview-annotation", threadRef, tabId, config.sessionId]),
63 - draftText: text,
62 offerSetup: false,
63 readDraft: () => ({
64 text: draft.current.text,
@@ -68,7 +66,6 @@ export function PreviewAnnotationSpeech({
66 }),
67 commitDraft: (next, selection) => {
68 draft.current = { text: next, cursor: selection.start, selectionEnd: selection.end };
71 - setText(next);
69 void publish(current.current.voiceState, { text: next, cursor: selection.start });
70 },
71 });
@@ -136,7 +133,6 @@ export function PreviewAnnotationSpeech({
133 cursor: Math.min(event.cursor, event.text.length),
134 selectionEnd: Math.min(event.selectionEnd ?? event.cursor, event.text.length),
135 };
139 - setText(event.text);
136 if (event.action === "focus" || event.action === "key" || event.action === "start")
137 guestFocused.current = true;
138 if (event.action === "blur") guestFocused.current = false;
apps/web/src/components/settings/VoiceSettingsPanel.tsx
-1
@@ -1083,7 +1083,6 @@ function TranscriptionTest(props: {
1083 projectId: props.projectId,
1084 projectName: props.projectName,
1085 ownerKey: JSON.stringify([props.environmentId, props.projectId]),
1086 - draftText: draft,
1086 cleanup: false,
1087 readDraft: () => ({
1088 text: draftRef.current,
apps/web/src/speech/useDictationShortcut.browser.test.tsx
+132 -2
@@ -1,17 +1,30 @@
1 import { DEFAULT_RESOLVED_KEYBINDINGS } from "@t3tools/shared/keybindings";
2 +import { createRef, type RefObject } from "react";
3 import { expect, it, vi } from "vite-plus/test";
4 import { render } from "../../test/render";
4 -import { useDictationShortcut } from "./useDictationShortcut";
5 +import { useDictationShortcut, type DictationKeyHandlers } from "./useDictationShortcut";
6
7 vi.mock("../hooks/useSettings", () => ({ useClientSettings: () => "hold" }));
8
8 -function Probe({ speech }: { speech: Parameters<typeof useDictationShortcut>[0]["speech"] }) {
9 +type ShortcutInput = Parameters<typeof useDictationShortcut>[0];
10 +
11 +function Probe({
12 + speech,
13 + targetRef,
14 + subscribeKeys,
15 +}: {
16 + speech: ShortcutInput["speech"];
17 + targetRef?: RefObject<HTMLElement | null>;
18 + subscribeKeys?: ShortcutInput["subscribeKeys"];
19 +}) {
20 useDictationShortcut({
21 keybindings: DEFAULT_RESOLVED_KEYBINDINGS,
22 speech,
23 disabled: false,
24 terminalOpen: false,
25 modelPickerOpen: false,
26 + ...(targetRef ? { targetRef } : {}),
27 + ...(subscribeKeys ? { subscribeKeys, ownsFocus: () => true } : {}),
28 });
29 return null;
30 }
@@ -80,3 +93,120 @@ it("finishes a released hold after startup even if the chord was pressed again w
93 expect(speech.start).toHaveBeenCalledOnce();
94 expect(speech.stop).toHaveBeenCalledOnce();
95 });
96 +
97 +const escape = () =>
98 + new KeyboardEvent("keydown", { key: "Escape", code: "Escape", bubbles: true, cancelable: true });
99 +
100 +const recording = () => ({
101 + ...actions(true),
102 + state: { phase: "recording" as const, error: null, errorAction: null },
103 +});
104 +
105 +it("lets Escape in another dictation editor reach that editor instead of canceling the recording", async () => {
106 + const speech = recording();
107 + const commentRef = createRef<HTMLTextAreaElement>();
108 + const editorEscape = vi.fn();
109 + await using screen = await render(
110 + <>
111 + <Probe speech={speech} />
112 + <Probe speech={actions(true)} targetRef={commentRef} />
113 + <div
114 + role="dialog"
115 + aria-label="Edit citation comment"
116 + onKeyDown={(event) => {
117 + if (event.key === "Escape") editorEscape();
118 + }}
119 + >
120 + <textarea aria-label="Comment" ref={commentRef} />
121 + <button type="button">Cancel</button>
122 + </div>
123 + </>,
124 + );
125 + for (const control of [
126 + screen.getByRole("textbox", { name: "Comment" }).element(),
127 + screen.getByRole("button", { name: "Cancel" }).element(),
128 + ]) {
129 + if (!(control instanceof HTMLElement)) throw new Error("expected a comment editor control");
130 + control.focus();
131 + control.dispatchEvent(escape());
132 + }
133 + expect(speech.cancel).not.toHaveBeenCalled();
134 + expect(editorEscape).toHaveBeenCalledTimes(2);
135 +});
136 +
137 +it("lets Escape in another editor of the recording editor's dialog reach that editor", async () => {
138 + const speech = recording();
139 + const replyRef = createRef<HTMLTextAreaElement>();
140 + const otherReplyRef = createRef<HTMLTextAreaElement>();
141 + await using screen = await render(
142 + <>
143 + <Probe speech={speech} targetRef={replyRef} />
144 + <Probe speech={actions(true)} targetRef={otherReplyRef} />
145 + <div role="dialog" aria-label="Pull request">
146 + <textarea aria-label="Reply A" ref={replyRef} />
147 + <textarea aria-label="Reply B" ref={otherReplyRef} />
148 + </div>
149 + </>,
150 + );
151 + const other = screen.getByRole("textbox", { name: "Reply B" }).element();
152 + if (!(other instanceof HTMLElement)) throw new Error("expected a reply editor");
153 + other.focus();
154 + other.dispatchEvent(escape());
155 + expect(speech.cancel).not.toHaveBeenCalled();
156 +});
157 +
158 +it("cancels the recording on Escape outside other shown dictation editors", async () => {
159 + const speech = recording();
160 + const reviewRef = createRef<HTMLTextAreaElement>();
161 + const commentRef = createRef<HTMLTextAreaElement>();
162 + await using screen = await render(
163 + <>
164 + <Probe speech={speech} targetRef={reviewRef} />
165 + <Probe speech={actions(true)} targetRef={commentRef} />
166 + <input aria-label="Search" />
167 + <div role="dialog" aria-label="Pull request">
168 + <textarea aria-label="Review" ref={reviewRef} />
169 + <div hidden>
170 + <textarea aria-label="Comment" ref={commentRef} />
171 + </div>
172 + <button type="button">Submit</button>
173 + </div>
174 + </>,
175 + );
176 + for (const control of [
177 + screen.getByRole("textbox", { name: "Search" }).element(),
178 + screen.getByRole("button", { name: "Submit" }).element(),
179 + ]) {
180 + if (!(control instanceof HTMLElement)) throw new Error("expected a focusable control");
181 + control.focus();
182 + control.dispatchEvent(escape());
183 + }
184 + expect(speech.cancel).toHaveBeenCalledTimes(2);
185 +});
186 +
187 +it.each([
188 + ["another editor is recording", false],
189 + ["the preview itself is recording", true],
190 +])(
191 + "routes Escape relayed from the preview guest to the preview when %s",
192 + async (_name, previewRecords) => {
193 + const composer = previewRecords ? actions(true) : recording();
194 + const preview = previewRecords ? recording() : actions(true);
195 + let relayed: DictationKeyHandlers | undefined;
196 + await using _screen = await render(
197 + <>
198 + <Probe speech={composer} />
199 + <Probe
200 + speech={preview}
201 + subscribeKeys={(handlers) => {
202 + relayed = handlers;
203 + return () => {};
204 + }}
205 + />
206 + </>,
207 + );
208 + relayed?.keydown(escape());
209 + expect(composer.cancel).not.toHaveBeenCalled();
210 + expect(preview.cancel).toHaveBeenCalledTimes(previewRecords ? 1 : 0);
211 + },
212 +);
apps/web/src/speech/useDictationShortcut.ts
+33 -3
@@ -60,6 +60,31 @@ function ownsShortcut(id: symbol) {
60 return entries.find(([, read]) => !read().targetRef && !read().ownsFocus)?.[0] === id;
61 }
62
63 +const targetScope = (field: HTMLElement | null | undefined) =>
64 + field?.closest('[role="dialog"]') ?? field;
65 +
66 +// Keys an outside editor (the preview guest) relays arrive on window, so remember their editor.
67 +const relayedFrom = new WeakMap<Event, symbol>();
68 +
69 +// Escape inside another shown dictation editor, or the dialog hosting it, belongs to that editor.
70 +// The editor itself is checked before dialogs, since one dialog can host several editors.
71 +function inOtherShortcutTarget(id: symbol, event: Event) {
72 + const source = relayedFrom.get(event);
73 + if (source) return source !== id;
74 + const target = event.target;
75 + if (!(target instanceof Node)) return false;
76 + const own = shortcutTargets.get(id)?.().targetRef?.current;
77 + const others = [...shortcutTargets]
78 + .filter(([other]) => other !== id)
79 + .map(([, read]) => read().targetRef?.current)
80 + .filter((field): field is HTMLElement => !!field?.getClientRects().length);
81 + for (const scope of [(field?: HTMLElement | null) => field, targetScope]) {
82 + if (scope(own)?.contains(target)) return false;
83 + if (others.some((field) => scope(field)?.contains(target))) return true;
84 + }
85 + return false;
86 +}
87 +
88 export function useDictationShortcut(input: ShortcutInput) {
89 const id = useRef(Symbol("dictation-target"));
90 const subscribeKeys = input.subscribeKeys;
@@ -103,7 +128,8 @@ export function useDictationShortcut(input: ShortcutInput) {
128 if (
129 event.key === "Escape" &&
130 (pressed.current !== null ||
106 - (speech.state.phase !== "idle" && speech.state.phase !== "error"))
131 + (speech.state.phase !== "idle" && speech.state.phase !== "error")) &&
132 + !inOtherShortcutTarget(id.current, event)
133 ) {
134 event.preventDefault();
135 event.stopImmediatePropagation();
@@ -146,9 +172,13 @@ export function useDictationShortcut(input: ShortcutInput) {
172 };
173 };
174 const onBlur = () => finish(null);
175 + const relay = (event: KeyboardEvent) => {
176 + relayedFrom.set(event, id.current);
177 + window.dispatchEvent(event);
178 + };
179 const unsubscribe = subscribeKeys?.({
150 - keydown: (event) => window.dispatchEvent(event),
151 - keyup: (event) => window.dispatchEvent(event),
180 + keydown: relay,
181 + keyup: relay,
182 blur: onBlur,
183 });
184 window.addEventListener("keydown", onKeyDown, true);
apps/web/src/speech/useEnvironmentSpeechInput.browser.test.tsx
-1
@@ -256,7 +256,6 @@ function Probe(props: {
256 projectName: props.projectName,
257 cleanup: props.cleanup,
258 ownerKey: "draft",
259 - draftText: mocks.draft,
259 readDraft: () => ({
260 text: mocks.draft,
261 selection: { start: mocks.draft.length, end: mocks.draft.length },
apps/web/src/speech/useEnvironmentSpeechInput.ts
+1 -28
@@ -92,7 +92,6 @@ type HookInput = {
92 readonly projectId?: ProjectId | null | undefined;
93 readonly projectName?: string | null | undefined;
94 readonly ownerKey: string;
95 - readonly draftText: string;
95 readonly disabled?: boolean | undefined;
96 /**
97 * Until a destination is chosen, voice input is offered and its first gesture opens Voice
@@ -212,19 +211,6 @@ export function useEnvironmentSpeechInput(input: HookInput) {
211 useEffect(() => {
212 latestRef.current = { input, clientSettings, cleanupPrepared, projectWords };
213 });
215 - const draftRevisionRef = useRef({ ownerKey: input.ownerKey, text: input.draftText, revision: 0 });
216 - // oxlint-disable-next-line t3code/no-unreasoned-state-sync -- tracks committed draft revisions so a late transcript cannot overwrite an edited draft
217 - useEffect(() => {
218 - const revision = draftRevisionRef.current;
219 - if (revision.ownerKey !== input.ownerKey || revision.text !== input.draftText) {
220 - draftRevisionRef.current = {
221 - ownerKey: input.ownerKey,
222 - text: input.draftText,
223 - revision: revision.revision + 1,
224 - };
225 - }
226 - }, [input.ownerKey, input.draftText]);
227 -
214 // oxlint-disable-next-line t3code/no-unreasoned-state-sync -- owns the recording controller bound to the consented destination; it is disposed with the consent
215 useEffect(() => {
216 if (!prepared || !grantId || typeof navigator === "undefined") return;
@@ -242,20 +228,7 @@ export function useEnvironmentSpeechInput(input: HookInput) {
228 const readDraft = (): VoiceDraftSnapshot => {
229 const current = latestRef.current.input;
230 const draft = current.readDraft();
245 - const revision = draftRevisionRef.current;
246 - if (revision.ownerKey !== current.ownerKey || revision.text !== draft.text) {
247 - draftRevisionRef.current = {
248 - ownerKey: current.ownerKey,
249 - text: draft.text,
250 - revision: revision.revision + 1,
251 - };
252 - }
253 - return {
254 - ownerKey: current.ownerKey,
255 - text: draft.text,
256 - selection: draft.selection,
257 - revision: draftRevisionRef.current.revision,
258 - };
231 + return { ownerKey: current.ownerKey, text: draft.text, selection: draft.selection };
232 };
233 const platform = createBrowserVoiceInputPlatform({
234 prepared,
apps/web/src/speech/useTextFieldSpeech.tsx
-1
@@ -31,7 +31,6 @@ export function useTextFieldSpeech(input: {
31 environmentId: input.environmentId,
32 projectId: input.projectId,
33 ownerKey: input.ownerKey,
34 - draftText: input.text,
34 disabled: input.disabled,
35 offerSetup: false,
36 readDraft: () => {
docs/user/voice-input.md
+2
@@ -83,6 +83,8 @@ macOS and Ctrl elsewhere). Press again, or the check mark, to finish; `Esc`
83 discards the recording and leaves your draft unchanged. Recordings last up to
84 five minutes. Text is inserted at the cursor or replaces the selection you had
85 when recording started, and the editor is read-only while dictation runs.
86 +Changes that reach the draft meanwhile, such as a cited message, are kept;
87 +the text then goes after your selection instead of replacing it.
88
89 Pressing Send while dictating finishes the recording and sends the draft once
90 the transcript has been inserted. Cancelling also cancels that queued send. In
packages/client-runtime/src/voice-input/controller.test.ts
+63 -9
@@ -49,7 +49,6 @@ function draft(overrides: Partial<VoiceDraftSnapshot> = {}): VoiceDraftSnapshot
49 ownerKey: "environment:thread",
50 text: "hello world",
51 selection: { start: 6, end: 11 },
52 - revision: 1,
52 ...overrides,
53 };
54 }
@@ -102,7 +101,7 @@ describe("environment voice input", () => {
101 expect(harness.controller.currentState.phase).toBe("error");
102 expect(harness.deleted).toEqual([]);
103
105 - harness.setDraft(draft({ text: "updated ", revision: 2, selection: { start: 8, end: 8 } }));
104 + harness.setDraft(draft({ text: "updated ", selection: { start: 8, end: 8 } }));
105 await harness.controller.start();
106 expect(transcribe.mock.calls.map(([uri]) => uri)).toEqual([
107 "file:///voice.m4a",
@@ -280,19 +279,74 @@ describe("resolveTranscriptCommit", () => {
279 });
280 });
281
283 - it("does not replace text after the owner, text, or revision changes", () => {
282 + it.each([
283 + [
284 + "at the start of the selection",
285 + { start: 6, end: 11 },
286 + "hello [cite]world",
287 + "hello [cite]world new text",
288 + ],
289 + [
290 + "inside the selection",
291 + { start: 6, end: 11 },
292 + "hello wo[cite]rld",
293 + "hello wo[cite]rld new text",
294 + ],
295 + ["at the caret", { start: 6, end: 6 }, "hello [cite] world", "hello [cite] new text world"],
296 + [
297 + "that repeats the selection",
298 + { start: 0, end: 5 },
299 + "hello there hello world",
300 + "hello there hello new text world",
301 + ],
302 + [
303 + "that repeats the selection next to it",
304 + { start: 0, end: 5 },
305 + "hello hello world",
306 + "hello hello new text world",
307 + ],
308 + ])("keeps text inserted %s while recording", (_, selection, current, expected) => {
309 + const captured = draft({ text: "hello world", selection });
310 + const result = resolveTranscriptCommit(
311 + captured,
312 + draft({ text: current, selection: { start: 0, end: 0 } }),
313 + "new text",
314 + "en-US",
315 + );
316 + expect(result).toMatchObject({ kind: "commit", text: expected });
317 + });
318 +
319 + it("keeps every citation when several were inserted while recording", () => {
320 + const captured = draft({ text: "[todo", selection: { start: 0, end: 1 } });
321 + const current = "[A](t3-citation://v1/a)[todo[B](t3-citation://v1/b)";
322 + const result = resolveTranscriptCommit(
323 + captured,
324 + draft({ text: current, selection: { start: 0, end: 0 } }),
325 + "spoken",
326 + "en-US",
327 + );
328 + expect(result).toMatchObject({ kind: "commit", text: `${current} spoken` });
329 + });
330 +
331 + it("keeps surrogate pairs whole when an emoji changed while recording", () => {
332 + const captured = draft({ text: "🔔", selection: { start: 0, end: 0 } });
333 + const result = resolveTranscriptCommit(
334 + captured,
335 + draft({ text: "🤔", selection: { start: 0, end: 0 } }),
336 + "hello",
337 + "en-US",
338 + );
339 + expect(result).toMatchObject({ kind: "commit", text: "🤔hello" });
340 + });
341 +
342 + it("does not insert into a missing draft or another owner's draft", () => {
343 const captured = draft();
344 + expect(resolveTranscriptCommit(captured, null, "text", "en-US")).toEqual({ kind: "stale" });
345 expect(
346 resolveTranscriptCommit(captured, draft({ ownerKey: "other" }), "text", "en-US"),
347 ).toEqual({
348 kind: "stale",
349 });
290 - expect(resolveTranscriptCommit(captured, draft({ text: "newer" }), "text", "en-US")).toEqual({
291 - kind: "stale",
292 - });
293 - expect(resolveTranscriptCommit(captured, draft({ revision: 2 }), "text", "en-US")).toEqual({
294 - kind: "stale",
295 - });
350 });
351 });
352
packages/client-runtime/src/voice-input/controller.ts
+40 -17
@@ -50,7 +50,6 @@ export type VoiceDraftSnapshot = {
50 readonly ownerKey: string;
51 readonly text: string;
52 readonly selection: { readonly start: number; readonly end: number };
53 - readonly revision: number;
53 };
54
55 export type VoiceRecorderStatus = {
@@ -109,18 +108,37 @@ type TranscriptCommitResult =
108
109 const WORD_SPACELESS_LANGUAGES = new Set(["ja", "zh", "yue", "th", "lo", "km", "my"]);
110
111 +const isHighSurrogate = (code: number) => code >= 0xd800 && code <= 0xdbff;
112 +const isLowSurrogate = (code: number) => code >= 0xdc00 && code <= 0xdfff;
113 +
114 +// Where `position` in `before` lands in `text`, placed after anything inserted at it.
115 +function mapPosition(before: string, text: string, position: number, suffixFirst: boolean) {
116 + const shared = Math.min(before.length, text.length);
117 + const matchesFromEnd = (offset: number) =>
118 + before[before.length - 1 - offset] === text[text.length - 1 - offset];
119 + let prefix = 0;
120 + let suffix = 0;
121 + // Matching is per UTF-16 unit; never let a boundary split a surrogate pair.
122 + const scanSuffix = (limit: number) => {
123 + while (suffix < limit && matchesFromEnd(suffix)) suffix += 1;
124 + if (suffix > 0 && isLowSurrogate(before.charCodeAt(before.length - suffix))) suffix -= 1;
125 + };
126 + if (suffixFirst) scanSuffix(shared);
127 + while (prefix < shared - suffix && before[prefix] === text[prefix]) prefix += 1;
128 + if (prefix > 0 && isHighSurrogate(before.charCodeAt(prefix - 1))) prefix -= 1;
129 + if (!suffixFirst) scanSuffix(shared - prefix);
130 + if (position < prefix) return position;
131 + if (position >= before.length - suffix) return position + text.length - before.length;
132 + return text.length - suffix;
133 +}
134 +
135 export function resolveTranscriptCommit(
136 captured: VoiceDraftSnapshot,
137 current: VoiceDraftSnapshot | null,
138 transcript: string,
139 locale: string,
140 ): TranscriptCommitResult {
118 - if (
119 - !current ||
120 - current.ownerKey !== captured.ownerKey ||
121 - current.text !== captured.text ||
122 - current.revision !== captured.revision
123 - ) {
141 + if (!current || current.ownerKey !== captured.ownerKey) {
142 return { kind: "stale" };
143 }
144
@@ -129,14 +147,24 @@ export function resolveTranscriptCommit(
147 return { kind: "empty" };
148 }
149
132 - const isEmptySelection = captured.selection.start === captured.selection.end;
150 + const { text } = current;
151 + // An edit made while recording (a citation, a paste) is never replaced. The edit's position is
152 + // ambiguous over repeated text, so insert after the selection under both extreme alignments.
153 + const unchanged = text === captured.text;
154 + const end = unchanged
155 + ? captured.selection.end
156 + : Math.max(
157 + mapPosition(captured.text, text, captured.selection.end, false),
158 + mapPosition(captured.text, text, captured.selection.end, true),
159 + );
160 + const start = unchanged ? captured.selection.start : end;
161 + const isEmptySelection = start === end;
162 const language = locale.split(/[-_]/)[0]!.toLowerCase();
163 let insertion = replacement;
164 if (isEmptySelection && !WORD_SPACELESS_LANGUAGES.has(language)) {
165 const auto = language === "auto";
137 - const caret = captured.selection.start;
138 - const left = captured.text.slice(Math.max(0, caret - 2), caret);
139 - const right = captured.text.slice(caret, caret + 2);
166 + const left = text.slice(Math.max(0, start - 2), start);
167 + const right = text.slice(start, start + 2);
168 const leftEdge = auto ? /[A-Za-z0-9.!?,:;)\]}'"]$/ : /(?:\p{L}|\p{M}|\p{N}|[.!?,:;)\]}'"])$/u;
169 const rightEdge = auto ? /^[A-Za-z0-9([{'"]/ : /^[\p{L}\p{M}\p{N}([{'"]/u;
170 const leftNeedsBoundary =
@@ -150,12 +178,7 @@ export function resolveTranscriptCommit(
178 insertion = `${leftNeedsBoundary ? " " : ""}${replacement}${rightNeedsBoundary ? " " : ""}`;
179 }
180
153 - const result = replaceTextRange(
154 - captured.text,
155 - captured.selection.start,
156 - captured.selection.end,
157 - insertion,
158 - );
181 + const result = replaceTextRange(text, start, end, insertion);
182 return {
183 kind: "commit",
184 text: result.text,