@mengine/medeo-client 1.2.1-alpha.5 → 1.2.1-alpha.6

package/dist/{document-ClBRZtLn.js → document-49OdvviW.js}

@@ -1,6 +1,7 @@
 import { Mirror, schema } from "loro-mirror";
 import { z } from "zod";
 import { LoroDoc } from "loro-crdt";
+import { produce } from "immer";
 //#region src/client/base64.ts
 function bytesToBase64(bytes) {
 	let binary = "";
@@ -87,8 +88,7 @@ function jsonTransform() {
 const trackItem = schema.LoroMap({
 	part_id: schema.String(),
 	time_position: schema.String().transform(jsonTransform()),
-	fallback_abs_ms: schema.Number({ required: false }),
-	duration_override_ms: schema.Number({ required: false })
+	fallback_abs_ms: schema.Number({ required: false })
 });
 const track = schema.LoroMap({
 	id: schema.String(),
@@ -111,7 +111,7 @@ const speechPart = schema.LoroMap({
 	volume: schema.Number(),
 	audio_storage_key: schema.String(),
 	origin_speech_id: schema.String(),
-	voice: schema.String({ required: false }).transform(jsonTransform()),
+	voice: schema.String().transform(jsonTransform()),
 	caption_ids: schema.String().transform(jsonTransform())
 }, { required: false });
 const captionStyle = schema.LoroMap({
@@ -322,71 +322,6 @@ function syncAggregatedClipsTimePosition(doc) {
 	}
 }
 /**
-* Resolve authoritative `time_position` edges without inventing parents from
-* geometric overlap. Roots stay absolute/sequential; every anchored descendant
-* follows its declared parent, regardless of whether that parent is video, BGM,
-* or speech.
-*/
-function resolveAnchoredClipsTimePosition(doc) {
-	const items = [
-		...doc.main_track,
-		...doc.bgm_track,
-		...doc.speech_track,
-		...doc.caption_track
-	];
-	const byPart = indexByPart(items);
-	const resolved = /* @__PURE__ */ new Set();
-	const visiting = /* @__PURE__ */ new Set();
-	const resolve = (item) => {
-		if (resolved.has(item.part_id)) return item.abs_time_position;
-		if (visiting.has(item.part_id)) throw new Error(`anchored position cycle at ${item.part_id}`);
-		visiting.add(item.part_id);
-		const position = item.time_position;
-		if (position.mode === "absolute") item.abs_time_position = Math.round(position.offsetMs);
-		else if (position.mode === "anchored") {
-			const parent = byPart.get(position.anchorPartId);
-			if (parent != null) item.abs_time_position = resolve(parent) + Math.round(position.offsetMs);
-		}
-		visiting.delete(item.part_id);
-		resolved.add(item.part_id);
-		return item.abs_time_position;
-	};
-	for (const item of items) resolve(item);
-}
-/** Preserve the historical orphan fallback without re-parenting valid roots. */
-function recoverOrphanedSpeechAnchorsByTime(doc) {
-	const placedPartIds = new Set([
-		...doc.main_track,
-		...doc.bgm_track,
-		...doc.speech_track,
-		...doc.caption_track
-	].map((item) => item.part_id));
-	const ranges = doc.main_track.map((item) => ({
-		part_id: item.part_id,
-		start: item.abs_time_position,
-		end: item.abs_time_position + partDurationMs(doc, item.part_id)
-	}));
-	if (ranges.length === 0) return;
-	for (const speech of doc.speech_track) {
-		const position = speech.time_position;
-		if (position.mode !== "anchored" || placedPartIds.has(position.anchorPartId)) continue;
-		const range = ranges.find((candidate) => candidate.start <= speech.abs_time_position && speech.abs_time_position < candidate.end) ?? (speech.abs_time_position < ranges[0].start ? ranges[0] : ranges[ranges.length - 1]);
-		const relative = Math.max(0, Math.round(speech.abs_time_position - range.start));
-		let aggregation = doc.aggregations.find((candidate) => candidate.body_part_id === range.part_id);
-		if (aggregation == null) {
-			aggregation = {
-				body_part_id: range.part_id,
-				attachments: []
-			};
-			doc.aggregations.push(aggregation);
-		}
-		aggregation.attachments.push({
-			part_id: speech.part_id,
-			relative_time_position: relative
-		});
-	}
-}
-/**
 * Reassign each speech to the video clip whose time range contains its start,
 * rebuilding `aggregations`. A speech before the first clip or after the last
 * falls back to the first / last clip respectively (FE: see §4 note — FE falls
@@ -589,14 +524,14 @@ function hostVideoOf(doc, speechPartId) {
 //#endregion
 //#region src/timeline-core/entrypoints.ts
 /**
-* The full solve pipeline: arrange → resolve declared parent edges →
-* resolve-overlap → fill-gaps → recalc. The single cascade the read-side projection runs; ops
+* The full solve pipeline: arrange → sync → reassign → resolve-overlap →
+* fill-gaps → recalc. The single cascade the read-side projection runs; ops
 * never call it (they write only facts — RFC 02 §7/§10).
 */
 function cascadeAfterVideoClipChanges(doc, makeEmptyPart) {
 	arrangeMainTrackSeamlessly(doc);
-	resolveAnchoredClipsTimePosition(doc);
-	recoverOrphanedSpeechAnchorsByTime(doc);
+	syncAggregatedClipsTimePosition(doc);
+	reassignSpeechesToVideoClipsByTime(doc);
 	resolveAllSpeechOverlaps(doc);
 	fillMainTrackTimeGaps(doc, makeEmptyPart);
 	recalculateTimelineDuration(doc);
@@ -670,7 +605,7 @@ function videoDocumentToTimelineDoc(document) {
 		caption_track: captionItems,
 		bgm_track: bgmItems,
 		part_library: partLibrary,
-		aggregations: seedAggregations([...mainItems, ...bgmItems], speechItems),
+		aggregations: seedAggregations(mainItems, speechItems),
 		timeline: {
 			duration_ms: 0,
 			unit_time_ms: document.timeline?.unit_time_ms ?? 0
@@ -694,18 +629,19 @@ function seedAbs(position, fallbackAbsMs) {
 	}
 }
 /**
-* Build the read-view `aggregations` parent-map from anchored speech items.
-* Video and BGM roots are both valid hosts; absolute speech roots have no entry.
-* Caption→speech ownership remains in `speech_part_id` / `caption_ids`.
+* Build the cascade `aggregations` parent-map from anchored items: an anchored
+* item's `anchorPartId` is the host's `part_id` (= `body_part_id`), `offsetMs`
+* is `relative_time_position`. Only main-track (video) hosts form aggregations;
+* caption→speech relations are recomputed by the cascade from `caption.start_ms`.
 */
-function seedAggregations(hostItems, speechItems) {
-	const hostPartIds = new Set(hostItems.map((item) => item.part_id));
+function seedAggregations(mainItems, speechItems) {
+	const mainPartIds = new Set(mainItems.map((item) => item.part_id));
 	const byHost = /* @__PURE__ */ new Map();
 	const order = [];
 	for (const speech of speechItems) {
 		if (speech.time_position.mode !== "anchored") continue;
 		const host = speech.time_position.anchorPartId;
-		if (!hostPartIds.has(host)) continue;
+		if (!mainPartIds.has(host)) continue;
 		let aggregation = byHost.get(host);
 		if (aggregation == null) {
 			aggregation = {
@@ -838,8 +774,7 @@ const trackItemTimePositionSchema = z.discriminatedUnion("mode", [
 const trackItemSchema = z.object({
 	part_id: z.string(),
 	time_position: trackItemTimePositionSchema,
-	fallback_abs_ms: finiteNumber.optional(),
-	duration_override_ms: z.number().finite().positive().optional()
+	fallback_abs_ms: finiteNumber.optional()
 }).passthrough();
 const trackSchema = z.object({
 	id: z.string().optional(),
@@ -871,7 +806,7 @@ const speechPartSchema = z.object({
 	volume: finiteNumber,
 	audio_storage_key: z.string(),
 	origin_speech_id: z.string(),
-	voice: z.unknown().optional(),
+	voice: z.unknown(),
 	caption_ids: z.array(z.string())
 }).passthrough();
 const captionPartSchema = z.object({
@@ -1392,7 +1327,7 @@ function fromVideoDocument(document) {
 		above_main_tracks: aboveTracks.map((t) => draftTrack(t, view.absByPartId)),
 		below_main_tracks: belowTracks.map((t) => draftTrack(t, view.absByPartId)),
 		part_aggregations: view.aggregations,
-		part_library: toReadViewPartLibrary(view.partLibrary, view.durationMs, view.derivedFillerPartIds, captionDurationOverrides(document)),
+		part_library: toReadViewPartLibrary(view.partLibrary, view.durationMs, view.derivedFillerPartIds),
 		version: document.meta.version
 	};
 }
@@ -1413,7 +1348,7 @@ function fromVideoDocument(document) {
 * and they are not part of the document (see `fromVideoDocument`). An empty clip a
 * writer placed deliberately is NOT in that set and is projected normally.
 */
-function toReadViewPartLibrary(partLibrary, durationMs, derivedFillerPartIds, captionDurationByPartId) {
+function toReadViewPartLibrary(partLibrary, durationMs, derivedFillerPartIds) {
 	const out = {};
 	for (const [partId, part] of Object.entries(partLibrary)) {
 		if (derivedFillerPartIds.has(partId)) continue;
@@ -1433,7 +1368,7 @@ function toReadViewPartLibrary(partLibrary, durationMs, derivedFillerPartIds, ca
 			const caption = clone(part.caption);
 			out[partId] = { caption: {
 				...caption,
-				duration_ms: captionDurationByPartId.get(partId) ?? caption.initial_duration_ms
+				duration_ms: caption.initial_duration_ms
 			} };
 		} else if (part.bgm != null) out[partId] = { bgm: {
 			...clone(part.bgm),
@@ -1442,14 +1377,6 @@ function toReadViewPartLibrary(partLibrary, durationMs, derivedFillerPartIds, ca
 	}
 	return out;
 }
-function captionDurationOverrides(document) {
-	const result = /* @__PURE__ */ new Map();
-	for (const track of document.tracks ?? []) {
-		if (track.parts_kind !== "caption") continue;
-		for (const item of track.items ?? []) if (item.duration_override_ms != null) result.set(item.part_id, item.duration_override_ms);
-	}
-	return result;
-}
 /** Deep-clone a part payload, dropping the derived read-view `duration_ms`. */
 function omitDurationMs(part) {
 	const { duration_ms: _drop, ...rest } = clone(part);
@@ -1612,7 +1539,6 @@ function rowToTrack(row) {
 				time_position: item.time_position
 			};
 			if (item.fallback_abs_ms != null) result.fallback_abs_ms = item.fallback_abs_ms;
-			if (item.duration_override_ms != null) result.duration_override_ms = item.duration_override_ms;
 			return result;
 		})
 	};
@@ -1687,7 +1613,7 @@ function createMirrorVideoDocument(document, options = {}) {
 		doc,
 		schema: videoDocumentMirrorSchema
 	}).setState((draft) => {
-		seedDraft(draft, document);
+		writeVideoDocumentToDraft(draft, document);
 	}, { origin: options.origin ?? "mengine.bootstrap" });
 	return doc;
 }
@@ -1695,8 +1621,8 @@ function createMirrorVideoDocument(document, options = {}) {
 function createMirrorVideoDocumentAdapter(document, options) {
 	return new MirrorVideoDocumentAdapter(createMirrorVideoDocument(document, options));
 }
-/** Write a whole `VideoDocument` into the mirror draft (used to seed a fresh doc). */
-function seedDraft(draft, document) {
+/** Write a whole `VideoDocument` into a draft (seed a fresh doc / plain-memory state). */
+function writeVideoDocumentToDraft(draft, document) {
 	draft.meta = {
 		schema_version: document.meta.schema_version,
 		draft_id: document.meta.draft_id ?? void 0,
@@ -1721,10 +1647,124 @@ function toTrackRow(track) {
 		items: (track.items ?? []).map((item) => ({
 			part_id: item.part_id,
 			time_position: item.time_position,
-			fallback_abs_ms: item.fallback_abs_ms,
-			duration_override_ms: item.duration_override_ms
+			fallback_abs_ms: item.fallback_abs_ms
 		}))
 	};
 }
 //#endregion
-export { TIMELINE_SKELETON_DURATION_MS as A, base64ToBytes as B, reassignSpeechesToVideoClipsByTime as C, resolveAnchoredClipsTimePosition as D, resolveAllSpeechOverlaps as E, VIDEO_DOCUMENT_SCHEMA_VERSION as F, effectiveVideoClipDurationMs as I, partUnionToDraft as L, partDurationMs as M, safeDurationMs as N, resolveSpeechOverlapByShiftingVideos as O, DEFAULT_UNIT_TIME_MS as P, recordEntries as R, fillMainTrackTimeGaps as S, recoverOrphanedSpeechAnchorsByTime as T, bytesToBase64 as V, findLaneTrack as _, buildInitialVideoDocument as a, cascadeAfterVideoClipChanges as b, fromVideoDocument as c, assertValidVideoDocument as d, validateVideoDocument as f, ensureLaneTrack as g, LANE_KINDS_IN_STACK_ORDER as h, readVideoDocumentFromDraft as i, isEmptyVideoClip as j, syncAggregatedClipsTimePosition as k, toVideoDocument as l, videoDocumentSchema as m, createMirrorVideoDocument as n, buildSpeechHostMap as o, partUnionSchema as p, createMirrorVideoDocumentAdapter as r, derivePositionFromAbs as s, MirrorVideoDocumentAdapter as t, VideoDocumentValidationError as u, laneTrackId as v, recalculateTimelineDuration as w, arrangeMainTrackSeamlessly as x, solveVideoDocument as y, videoDocumentMirrorSchema as z };
+//#region src/editor/id-gen.ts
+/**
+* Part-id generation, aligned with the online ecosystem.
+*
+* The authoritative online producers — agent-harness (`@harness/shared`
+* `genObjId`) and director.v2 (`common/obj_id.py` `gen_obj_id`) — both mint part
+* ids as `` `${prefix}_${ulid()}` ``, and real captured drafts use exactly that
+* shape (`clip_…` / `spe_…` / `cap_…` / `bgm_…`, each a 26-char ULID). The engine
+* previously emitted `vc_<base36 timestamp><6 random>`, a different prefix AND a
+* different encoding — the sole cross-repo id divergence. This module removes it
+* by emitting the same `<prefix>_<ULID>` bytes.
+*
+* The ULID is generated inline (Crockford Base32, 48-bit time + 80-bit random)
+* rather than pulling the `ulid` npm package: the randomness class matches the
+* old generator (both `Math.random`-based) and it keeps `@mengine/medeo-client`
+* dependency-free for a purely mechanical id string. Part ids only need to be
+* unique and lexicographically time-sortable, which this satisfies.
+*/
+/** Crockford Base32 alphabet (no I, L, O, U), per the ULID spec. */
+const CROCKFORD = "0123456789ABCDEFGHJKMNPQRSTVWXYZ";
+const TIME_LEN = 10;
+const RANDOM_LEN = 16;
+function encodeTime(now) {
+	let out = "";
+	let ms = now;
+	for (let i = TIME_LEN - 1; i >= 0; i--) {
+		const mod = ms % 32;
+		out = CROCKFORD[mod] + out;
+		ms = (ms - mod) / 32;
+	}
+	return out;
+}
+function encodeRandom() {
+	let out = "";
+	for (let i = 0; i < RANDOM_LEN; i++) out += CROCKFORD[Math.floor(Math.random() * 32)];
+	return out;
+}
+/** A 26-char Crockford Base32 ULID (10-char time + 16-char random). */
+function ulid() {
+	return encodeTime(Date.now()) + encodeRandom();
+}
+function generatePartId(prefix) {
+	return `${prefix}_${ulid()}`;
+}
+//#endregion
+//#region src/document/plain-memory-adapter.ts
+/**
+* Pure in-memory `SemanticDocumentAdapter` — no Loro/WASM. Holds a
+* `VideoDocumentDraft` object and applies each `transact` via immer `produce`,
+* journaling every audit that actually mutated state (plus any ids minted
+* during that transact).
+*
+* Known benign difference vs `MirrorVideoDocumentAdapter`: the editor's
+* `setPart` assigns a fresh part object on every call, so a same-value rewrite
+* produces a new immer state and IS journaled here, while the mirror's deep
+* diff emits no commit. Terminal `snapshot()` stays equal; replay is idempotent.
+*/
+var PlainMemoryAdapter = class {
+	state;
+	_journal = [];
+	baseIdFactory;
+	/** Non-null only while a `transact` edit callback is running. */
+	pendingIds = null;
+	/**
+	* Recording wrapper around the underlying factory. Callers (sandbox editor)
+	* use this so every minted id is appended to the current transact's list.
+	*/
+	idFactory;
+	constructor(document, options) {
+		assertValidVideoDocument(document);
+		const draft = {};
+		writeVideoDocumentToDraft(draft, document);
+		this.state = draft;
+		this.baseIdFactory = options?.idFactory ?? generatePartId;
+		this.idFactory = (prefix) => {
+			const id = this.baseIdFactory(prefix);
+			if (this.pendingIds != null) this.pendingIds.push(id);
+			return id;
+		};
+	}
+	get journal() {
+		return this._journal;
+	}
+	hasContent() {
+		return (this.state.meta?.schema_version ?? "") !== "";
+	}
+	snapshot() {
+		return readVideoDocumentFromDraft(this.state);
+	}
+	/**
+	* Apply one op via immer. A throw in `edit` discards the draft (state and
+	* journal unchanged). When `produce` returns the same reference, there was
+	* no structural change — skip journal, matching mirror "no change, no commit".
+	*/
+	transact(edit, audit) {
+		this.pendingIds = [];
+		try {
+			const next = produce(this.state, edit);
+			if (next !== this.state) {
+				this.state = next;
+				this._journal.push({
+					...audit,
+					generated_ids: this.pendingIds
+				});
+			}
+		} finally {
+			this.pendingIds = null;
+		}
+	}
+};
+/** Build a `PlainMemoryAdapter` seeded with `document`. */
+function createPlainMemoryAdapter(document, options) {
+	return new PlainMemoryAdapter(document, options);
+}
+//#endregion
+export { resolveSpeechOverlapByShiftingVideos as A, partUnionToDraft as B, solveVideoDocument as C, reassignSpeechesToVideoClipsByTime as D, fillMainTrackTimeGaps as E, safeDurationMs as F, videoDocumentMirrorSchema as H, DEFAULT_UNIT_TIME_MS as I, VIDEO_DOCUMENT_SCHEMA_VERSION as L, TIMELINE_SKELETON_DURATION_MS as M, isEmptyVideoClip as N, recalculateTimelineDuration as O, partDurationMs as P, effectiveVideoClipDurationMs as R, laneTrackId as S, arrangeMainTrackSeamlessly as T, base64ToBytes as U, recordEntries as V, bytesToBase64 as W, partUnionSchema as _, createMirrorVideoDocument as a, ensureLaneTrack as b, readVideoDocumentFromDraft as c, derivePositionFromAbs as d, fromVideoDocument as f, validateVideoDocument as g, assertValidVideoDocument as h, MirrorVideoDocumentAdapter as i, syncAggregatedClipsTimePosition as j, resolveAllSpeechOverlaps as k, buildInitialVideoDocument as l, VideoDocumentValidationError as m, createPlainMemoryAdapter as n, createMirrorVideoDocumentAdapter as o, toVideoDocument as p, generatePartId as r, writeVideoDocumentToDraft as s, PlainMemoryAdapter as t, buildSpeechHostMap as u, videoDocumentSchema as v, cascadeAfterVideoClipChanges as w, findLaneTrack as x, LANE_KINDS_IN_STACK_ORDER as y, speedOf as z };