Make Matroska block decoding lazy

dev/convert.html

@@ -54,7 +54,8 @@
 				source
 			}),
 			output,
-			audio: {
+			audio: (_, n) => ({
+				discard: n > 1,
 				//codec: 'pcm-s16',
 				//sampleRate: 16000,
 				//numberOfChannels: 1,
@@ -65,7 +66,7 @@
 				//sampleRate: 4000
 				//discard: true
 				//forceTranscode: true,
-			},
+			}),
 			/*
 			video: {
 				discard: true,
@@ -91,6 +92,7 @@
 			},
 			*/
 			video: () => ({
+				discard: true,
 				//discard: true,
 				//crop: {
 				//	left: 0,
@@ -124,7 +126,7 @@
 				//width: 200,
 				//height: 100,
 			}),
-			tags: {
+			tags: {} ?? {
 				title: 'Bigggy',
 				artist: 'Buck Bunny',
 				images: [{
@@ -142,8 +144,8 @@
 				}
 			},
 			trim: {
-				start: 10,
+				//start: 10,
-				end: 20
+				//end: 20
 			},
 		});
 		console.log(conversion);

package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "mediabunny",
-  "version": "1.16.0",
+  "version": "1.16.1",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "mediabunny",
-      "version": "1.16.0",
+      "version": "1.16.1",
       "license": "MPL-2.0",
       "workspaces": [
         "packages/*"
@@ -7749,9 +7749,9 @@
       }
     },
     "node_modules/mediabunny": {
-      "version": "1.15.2",
+      "version": "1.16.0",
-      "resolved": "https://registry.npmjs.org/mediabunny/-/mediabunny-1.15.2.tgz",
+      "resolved": "https://registry.npmjs.org/mediabunny/-/mediabunny-1.16.0.tgz",
-      "integrity": "sha512-OvIuSK10wNdz+7Dd4bmeq/IWkYG0bkm33FQb29vmyDztpr47ru8XSxYansa3EHLy6lLEm4rUADUENOcmrnN0jg==",
+      "integrity": "sha512-JXNR8MVo4bMHN4WMLSBjCUMBAxPjEJKtfAqBJI7gJayMOMeLp00K3WE5CH5uLSyhk0yCr26bGNjIEBF0/pQjZA==",
       "license": "MPL-2.0",
       "peer": true,
       "workspaces": [
@@ -12242,7 +12242,7 @@
     },
     "packages/mp3-encoder": {
       "name": "@mediabunny/mp3-encoder",
-      "version": "1.16.0",
+      "version": "1.16.1",
       "license": "MPL-2.0",
       "devDependencies": {
         "@types/emscripten": "^1.40.1"

package.json

@@ -1,7 +1,7 @@
 {
   "name": "mediabunny",
   "author": "Vanilagy",
-  "version": "1.16.0",
+  "version": "1.16.1",
   "description": "Pure TypeScript media toolkit for reading, writing, and converting media files, directly in the browser.",
   "type": "module",
   "workspaces": [

packages/mp3-encoder/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@mediabunny/mp3-encoder",
   "author": "Vanilagy",
-  "version": "1.16.0",
+  "version": "1.16.1",
   "description": "MP3 encoder extension for Mediabunny, based on LAME.",
   "main": "./dist/bundles/mediabunny-mp3-encoder.mjs",
   "module": "./dist/bundles/mediabunny-mp3-encoder.mjs",

src/codec-data.ts

@@ -12,6 +12,7 @@ import {
 	assert,
 	assertNever,
 	Bitstream,
+	getUint24,
 	last,
 	readExpGolomb,
 	readSignedExpGolomb,
@@ -116,7 +117,7 @@ const findNalUnitsInLengthPrefixed = (packetData: Uint8Array, lengthSize: 1 | 2
 		} else if (lengthSize === 2) {
 			nalUnitLength = dataView.getUint16(offset, false);
 		} else if (lengthSize === 3) {
-			nalUnitLength = (dataView.getUint16(offset, false) << 8) + dataView.getUint8(offset + 2);
+			nalUnitLength = getUint24(dataView, offset, false);
 		} else if (lengthSize === 4) {
 			nalUnitLength = dataView.getUint32(offset, false);
 		} else {

src/matroska/matroska-demuxer.ts

@@ -116,7 +116,7 @@ type Cluster = {
 };
 
 type ClusterTrackData = {
-	track: InternalTrack | null;
+	track: InternalTrack;
 	startTimestamp: number;
 	endTimestamp: number;
 	firstKeyFrameTimestamp: number | null;
@@ -141,6 +141,7 @@ type ClusterBlock = {
 	referencedTimestamps: number[];
 	data: Uint8Array;
 	lacing: BlockLacing;
+	decoded: boolean;
 };
 
 type CuePoint = {
@@ -667,7 +668,7 @@ export class MatroskaDemuxer extends Demuxer {
 					const nextEntry = trackData.presentationTimestamps[i + 1]!;
 					currentBlock.duration = nextEntry.timestamp - currentBlock.timestamp;
 				} else if (currentBlock.duration === 0) {
-					if (track?.defaultDuration != null) {
+					if (track.defaultDuration != null) {
 						if (currentBlock.lacing === BlockLacing.None) {
 							currentBlock.duration = track.defaultDuration;
 						} else {
@@ -695,13 +696,11 @@ export class MatroskaDemuxer extends Demuxer {
 			trackData.startTimestamp = firstBlock.timestamp;
 			trackData.endTimestamp = lastBlock.timestamp + lastBlock.duration;
 
-			if (track) {
+			insertSorted(track.clusters, cluster, x => x.elementStartPos);
-				insertSorted(track.clusters, cluster, x => x.elementStartPos);
 
-				const hasKeyFrame = trackData.firstKeyFrameTimestamp !== null;
+			const hasKeyFrame = trackData.firstKeyFrameTimestamp !== null;
-				if (hasKeyFrame) {
+			if (hasKeyFrame) {
-					insertSorted(track.clustersWithKeyFrame, cluster, x => x.elementStartPos);
+				insertSorted(track.clustersWithKeyFrame, cluster, x => x.elementStartPos);
-				}
 			}
 		}
 
@@ -714,8 +713,13 @@ export class MatroskaDemuxer extends Demuxer {
 	getTrackDataInCluster(cluster: Cluster, trackNumber: number) {
 		let trackData = cluster.trackData.get(trackNumber);
 		if (!trackData) {
+			const track = cluster.segment.tracks.find(x => x.id === trackNumber);
+			if (!track) {
+				return null;
+			}
+
 			trackData = {
-				track: cluster.segment.tracks.find(x => x.id === trackNumber) ?? null,
+				track,
 				startTimestamp: 0,
 				endTimestamp: 0,
 				firstKeyFrameTimestamp: null,
@@ -728,7 +732,7 @@ export class MatroskaDemuxer extends Demuxer {
 		return trackData;
 	}
 
-	expandLacedBlocks(blocks: ClusterBlock[], track: InternalTrack | null) {
+	expandLacedBlocks(blocks: ClusterBlock[], track: InternalTrack) {
 		// https://www.matroska.org/technical/notes.html#block-lacing
 
 		for (let blockIndex = 0; blockIndex < blocks.length; blockIndex++) {
@@ -737,6 +741,12 @@ export class MatroskaDemuxer extends Demuxer {
 				continue;
 			}
 
+			// Decode the block data if it hasn't been decoded yet (needed for lacing expansion)
+			if (!originalBlock.decoded) {
+				originalBlock.data = this.decodeBlockData(track, originalBlock.data);
+				originalBlock.decoded = true;
+			}
+
 			const slice = FileSlice.tempFromBytes(originalBlock.data);
 
 			const frameSizes: number[] = [];
@@ -819,7 +829,7 @@ export class MatroskaDemuxer extends Demuxer {
 				const frameSize = frameSizes[i]!;
 				const frameData = readBytes(slice, frameSize);
 
-				const blockDuration = originalBlock.duration || (frameCount * (track?.defaultDuration ?? 0));
+				const blockDuration = originalBlock.duration || (frameCount * (track.defaultDuration ?? 0));
 
 				// Distribute timestamps evenly across the block duration
 				const frameTimestamp = originalBlock.timestamp + (blockDuration * i / frameCount);
@@ -832,6 +842,7 @@ export class MatroskaDemuxer extends Demuxer {
 					referencedTimestamps: originalBlock.referencedTimestamps,
 					data: frameData,
 					lacing: BlockLacing.None,
+					decoded: true,
 				});
 			}
 
@@ -1331,18 +1342,17 @@ export class MatroskaDemuxer extends Demuxer {
 				const trackNumber = readVarInt(slice);
 				if (trackNumber === null) break;
 
+				const trackData = this.getTrackDataInCluster(this.currentCluster, trackNumber);
+				if (!trackData) break; // Not a track we care about
+
 				const relativeTimestamp = readI16Be(slice);
 
 				const flags = readU8(slice);
 				const isKeyFrame = !!(flags & 0x80);
 				const lacing = (flags >> 1) & 0x3 as BlockLacing; // If the block is laced, we'll expand it later
 
-				const trackData = this.getTrackDataInCluster(this.currentCluster, trackNumber);
+				const blockData = readBytes(slice, size - (slice.filePos - dataStartPos));
-				let blockData = readBytes(slice, size - (slice.filePos - dataStartPos));
+				const hasDecodingInstructions = trackData.track.decodingInstructions.length > 0;
-
-				if (trackData.track) {
-					blockData = this.decodeBlockData(trackData.track, blockData);
-				}
 
 				trackData.blocks.push({
 					timestamp: relativeTimestamp, // We'll add the cluster's timestamp to this later
@@ -1351,6 +1361,7 @@ export class MatroskaDemuxer extends Demuxer {
 					referencedTimestamps: [],
 					data: blockData,
 					lacing,
+					decoded: !hasDecodingInstructions,
 				});
 			}; break;
 
@@ -1374,17 +1385,16 @@ export class MatroskaDemuxer extends Demuxer {
 				const trackNumber = readVarInt(slice);
 				if (trackNumber === null) break;
 
+				const trackData = this.getTrackDataInCluster(this.currentCluster, trackNumber);
+				if (!trackData) break;
+
 				const relativeTimestamp = readI16Be(slice);
 
 				const flags = readU8(slice);
 				const lacing = (flags >> 1) & 0x3 as BlockLacing; // If the block is laced, we'll expand it later
 
-				const trackData = this.getTrackDataInCluster(this.currentCluster, trackNumber);
+				const blockData = readBytes(slice, size - (slice.filePos - dataStartPos));
-				let blockData = readBytes(slice, size - (slice.filePos - dataStartPos));
+				const hasDecodingInstructions = trackData.track.decodingInstructions.length > 0;
-
-				if (trackData.track) {
-					blockData = this.decodeBlockData(trackData.track, blockData);
-				}
 
 				this.currentBlock = {
 					timestamp: relativeTimestamp, // We'll add the cluster's timestamp to this later
@@ -1393,6 +1403,7 @@ export class MatroskaDemuxer extends Demuxer {
 					referencedTimestamps: [],
 					data: blockData,
 					lacing,
+					decoded: !hasDecodingInstructions,
 				};
 				trackData.blocks.push(this.currentBlock);
 			}; break;
@@ -1600,24 +1611,26 @@ export class MatroskaDemuxer extends Demuxer {
 	}
 
 	decodeBlockData(track: InternalTrack, rawData: Uint8Array) {
+		assert(track.decodingInstructions.length > 0); // This method shouldn't be called otherwise
+
 		let currentData = rawData;
 
-		// In the vast number of cases there are exactly zero decoding instructions
+		for (const instruction of track.decodingInstructions) {
-		for (let i = 0; i < track.decodingInstructions.length; i++) {
-			const instruction = track.decodingInstructions[i]!;
 			assert(instruction.data);
 
 			switch (instruction.data.type) {
 				case 'decompress': {
 					switch (instruction.data.algorithm) {
 						case ContentCompAlgo.HeaderStripping: {
-							const prefix = instruction.data.settings ?? new Uint8Array(0);
+							if (instruction.data.settings && instruction.data.settings.length > 0) {
-							const newData = new Uint8Array(prefix.length + currentData.length);
+								const prefix = instruction.data.settings;
+								const newData = new Uint8Array(prefix.length + currentData.length);
 
-							newData.set(prefix, 0);
+								newData.set(prefix, 0);
-							newData.set(currentData, prefix.length);
+								newData.set(currentData, prefix.length);
 
-							currentData = newData;
+								currentData = newData;
+							}
 						}; break;
 
 						default: {
@@ -1959,6 +1972,12 @@ abstract class MatroskaTrackBacking implements InputTrackBacking {
 		const block = trackData.blocks[blockIndex];
 		assert(block);
 
+		// Perform lazy decoding if needed
+		if (!block.decoded) {
+			block.data = this.internalTrack.demuxer.decodeBlockData(this.internalTrack, block.data);
+			block.decoded = true;
+		}
+
 		const data = options.metadataOnly ? PLACEHOLDER_DATA : block.data;
 		const timestamp = block.timestamp / this.internalTrack.segment.timestampFactor;
 		const duration = block.duration / this.internalTrack.segment.timestampFactor;

src/reader.ts

@@ -6,7 +6,7 @@
  * file, You can obtain one at https://mozilla.org/MPL/2.0/.
  */
 
-import { assert, clamp, MaybePromise, toDataView } from './misc';
+import { assert, clamp, getUint24, MaybePromise, toDataView } from './misc';
 import { Source } from './source';
 
 export class Reader {
@@ -157,9 +157,10 @@ export const readU16Be = (slice: FileSlice) => {
 };
 
 export const readU24Be = (slice: FileSlice) => {
-	const high = readU16Be(slice);
+	const value = getUint24(slice.view, slice.bufferPos, false);
-	const low = readU8(slice);
+	slice.bufferPos += 3;
-	return high * 0x100 + low;
+
+	return value;
 };
 
 export const readI16Be = (slice: FileSlice) => {