@ahmedrowaihi/mediabunny-server 1.65.0-beta.261 → 1.65.0-beta.263

package/dist/bundles/mediabunny-server.cjs

@@ -735,29 +735,6 @@ var MediabunnyServer = (() => {
   var assertNever = (x) => {
     throw new Error(`Unexpected value: ${x}`);
   };
-  var getUint24 = (view, byteOffset, littleEndian) => {
-    const byte1 = view.getUint8(byteOffset);
-    const byte2 = view.getUint8(byteOffset + 1);
-    const byte3 = view.getUint8(byteOffset + 2);
-    if (littleEndian) {
-      return byte1 | byte2 << 8 | byte3 << 16;
-    } else {
-      return byte1 << 16 | byte2 << 8 | byte3;
-    }
-  };
-  var setUint24 = (view, byteOffset, value, littleEndian) => {
-    value = value >>> 0;
-    value = value & 16777215;
-    if (littleEndian) {
-      view.setUint8(byteOffset, value & 255);
-      view.setUint8(byteOffset + 1, value >>> 8 & 255);
-      view.setUint8(byteOffset + 2, value >>> 16 & 255);
-    } else {
-      view.setUint8(byteOffset, value >>> 16 & 255);
-      view.setUint8(byteOffset + 1, value >>> 8 & 255);
-      view.setUint8(byteOffset + 2, value & 255);
-    }
-  };
   var SECOND_TO_MICROSECOND_FACTOR = 1e6 * (1 + Number.EPSILON);
   var isWebKitCache = null;
   var isWebKit = () => {
@@ -1093,170 +1070,6 @@ var MediabunnyServer = (() => {
     }
   };
 
-  // packages/server/src/video-decoder.ts
-  var NodeAvVideoDecoder = class extends import_mediabunny2.CustomVideoDecoder {
-    constructor() {
-      super(...arguments);
-      this.codecContext = null;
-      // Bookkeeping to restore the original timing information
-      this.preciseTimings = [];
-    }
-    // eslint-disable-next-line @typescript-eslint/no-unused-vars
-    static supports(codec, config) {
-      return codec === "avc" || codec === "hevc" || codec === "vp8" || codec === "vp9" || codec === "av1";
-    }
-    async init() {
-      this.frame = new NodeAv3.Frame();
-      this.frame.alloc();
-      this.packet = new NodeAv3.Packet();
-      this.packet.alloc();
-    }
-    async initCodecContext(packet) {
-      assert(this.codecContext === null);
-      const codecId = CODEC_TO_CODEC_ID[this.codec];
-      assert(codecId !== void 0);
-      let codec;
-      if (this.codec === "vp9" && packet.sideData.alpha) {
-        codec = NodeAv3.Codec.findDecoderByName(NodeAv3.FF_DECODER_LIBVPX_VP9) ?? NodeAv3.Codec.findDecoder(codecId);
-      } else if (
-        // This check used to be "is not prefer-hardware", meaning it would default to using software decode. I
-        // didn't leave a comment for that back then so I actually don't know what it was for. I'm sure it was to
-        // work around an issue but, I don't know. Not using hardware decode at all by defaults feels wrong to me,
-        // so I changed it to what it is right now. If an issue is encountered, I can always change it.
-        this.config.hardwareAcceleration === "prefer-software" || this.codec === "av1"
-      ) {
-        codec = NodeAv3.Codec.findDecoder(codecId);
-      } else {
-        codec = await getHardwareDecoderCodec(codecId) ?? NodeAv3.Codec.findDecoder(codecId);
-      }
-      if (!codec) {
-        throw new Error(`Unable to obtain libav codec for '${this.codec}'.`);
-      }
-      const codecContext = new NodeAv3.CodecContext();
-      codecContext.allocContext3(codec);
-      this.pixelAspectRatio = simplifyRational({
-        num: (this.config.displayAspectWidth ?? this.config.codedWidth ?? 0) * (this.config.codedHeight ?? 0),
-        den: (this.config.displayAspectHeight ?? this.config.codedHeight ?? 0) * (this.config.codedWidth ?? 0) || 1
-      });
-      codecContext.width = this.config.codedWidth ?? 0;
-      codecContext.height = this.config.codedHeight ?? 0;
-      codecContext.codecType = NodeAv3.AVMEDIA_TYPE_VIDEO;
-      codecContext.codecId = codecId;
-      codecContext.extraData = this.config.description ? Buffer.from(toUint8Array(this.config.description)) : null;
-      codecContext.sampleAspectRatio = new NodeAv3.Rational(this.pixelAspectRatio.num, this.pixelAspectRatio.den);
-      codecContext.threadCount = 0;
-      const options = NodeAv3.Dictionary.fromObject({ strict: NodeAv3.FF_COMPLIANCE_STRICT });
-      const ret = await codecContext.open2(codec, options);
-      NodeAv3.FFmpegError.throwIfError(ret, "Open codec context");
-      this.codecContext = codecContext;
-    }
-    async decode(packet) {
-      if (this.codecContext === null) {
-        await this.initCodecContext(packet);
-      }
-      assert(this.codecContext);
-      this.packet.isKeyframe = packet.type === "key";
-      this.packet.data = Buffer.from(packet.data);
-      this.packet.timeBase = { num: 1, den: 1e6 };
-      this.packet.pts = BigInt(packet.microsecondTimestamp);
-      this.packet.dts = NodeAv3.AV_NOPTS_VALUE;
-      this.packet.duration = BigInt(packet.microsecondDuration);
-      if (packet.sideData.alpha) {
-        const matroskaBlockAdditional = Buffer.alloc(8 + packet.sideData.alpha.byteLength);
-        matroskaBlockAdditional[7] = 1;
-        matroskaBlockAdditional.set(packet.sideData.alpha, 8);
-        this.packet.addSideData(NodeAv3.AV_PKT_DATA_MATROSKA_BLOCKADDITIONAL, matroskaBlockAdditional);
-      }
-      const preciseTimingIndex = binarySearchLessOrEqual(
-        this.preciseTimings,
-        packet.microsecondTimestamp,
-        (x) => x.microsecondTimestamp
-      );
-      const existingEntry = preciseTimingIndex !== -1 ? this.preciseTimings[preciseTimingIndex] : null;
-      if (existingEntry && existingEntry.microsecondTimestamp === packet.microsecondTimestamp) {
-        if (existingEntry.timestamp !== packet.timestamp) {
-          existingEntry.timestampIsValid = false;
-        }
-        if (existingEntry.duration !== packet.duration) {
-          existingEntry.durationIsValid = false;
-        }
-      } else {
-        this.preciseTimings.splice(preciseTimingIndex + 1, 0, {
-          microsecondTimestamp: packet.microsecondTimestamp,
-          timestamp: packet.timestamp,
-          duration: packet.duration,
-          timestampIsValid: true,
-          durationIsValid: true
-        });
-        if (this.preciseTimings.length > 128) {
-          this.preciseTimings.shift();
-        }
-      }
-      const ret = await this.codecContext.sendPacket(this.packet);
-      NodeAv3.FFmpegError.throwIfError(ret, "Send packet");
-      this.packet.unref();
-      while (true) {
-        const receiveRet = await this.codecContext.receiveFrame(this.frame);
-        if (receiveRet === NodeAv3.AVERROR_EAGAIN || receiveRet === NodeAv3.AVERROR_EOF) {
-          break;
-        }
-        this.receiveFrame(receiveRet);
-      }
-    }
-    receiveFrame(ret) {
-      NodeAv3.FFmpegError.throwIfError(ret, "Receive frame");
-      this.frame.sampleAspectRatio = new NodeAv3.Rational(this.pixelAspectRatio.num, this.pixelAspectRatio.den);
-      let timestamp = Number(this.frame.pts) / 1e6;
-      let duration = Number(this.frame.duration) / 1e6;
-      const preciseTimingIndex = binarySearchLessOrEqual(
-        this.preciseTimings,
-        Number(this.frame.pts),
-        (x) => x.microsecondTimestamp
-      );
-      const entry = preciseTimingIndex !== -1 ? this.preciseTimings[preciseTimingIndex] : null;
-      if (entry && entry.microsecondTimestamp === Number(this.frame.pts)) {
-        if (entry.timestampIsValid) {
-          timestamp = entry.timestamp;
-        }
-        if (entry.durationIsValid) {
-          duration = entry.duration;
-        }
-      }
-      const clone = this.frame.clone();
-      if (!clone) {
-        throw new Error("Frame clone allocation failed.");
-      }
-      this.onSample(new import_mediabunny2.VideoSample(new AvFrameVideoSampleResource(clone), {
-        timestamp,
-        duration
-      }));
-    }
-    async flush() {
-      if (!this.codecContext) {
-        return;
-      }
-      const ret = await this.codecContext.sendPacket(null);
-      NodeAv3.FFmpegError.throwIfError(ret, "Flush decoder");
-      while (true) {
-        const receiveRet = await this.codecContext.receiveFrame(this.frame);
-        if (receiveRet === NodeAv3.AVERROR_EAGAIN || receiveRet === NodeAv3.AVERROR_EOF) {
-          break;
-        }
-        this.receiveFrame(receiveRet);
-      }
-      this.codecContext.flushBuffers();
-    }
-    close() {
-      this.codecContext?.freeContext();
-      this.frame.free();
-      this.packet.free();
-    }
-  };
-
-  // packages/server/src/video-encoder.ts
-  var import_mediabunny3 = __require("mediabunny");
-  var NodeAv4 = __toESM(__require("node-av"));
-
   // shared/bitstream.ts
   var Bitstream = class _Bitstream {
     constructor(bytes) {
@@ -1445,100 +1258,6 @@ var MediabunnyServer = (() => {
     return { header, bitstream };
   };
 
-  // src/sei.ts
-  var SEI_FRAMINGS = {
-    avc: {
-      // forbidden_zero_bit = 0, nal_ref_idc = 0 (an SEI is never a reference picture), nal_unit_type = 6.
-      get nalHeader() {
-        return [6 /* SEI */];
-      },
-      isSeiNalUnit: (firstByte) => extractNalUnitTypeForAvc(firstByte) === 6 /* SEI */,
-      // AVC VCL NAL unit types are 1..5; the parameter sets, AUD and SEI that may precede them are all above.
-      isVcl: (firstByte) => {
-        const type = extractNalUnitTypeForAvc(firstByte);
-        return type >= 1 /* NON_IDR_SLICE */ && type <= 5 /* IDR */;
-      },
-      iterateNalUnits: (packetData, decoderConfig) => iterateAvcNalUnits(packetData, decoderConfig),
-      concatNalUnits: (nalUnits, decoderConfig) => concatAvcNalUnits(nalUnits, decoderConfig)
-    },
-    hevc: {
-      // nuh_layer_id = 0, nuh_temporal_id_plus1 = 1
-      get nalHeader() {
-        return [39 /* PREFIX_SEI_NUT */ << 1, 1];
-      },
-      isSeiNalUnit: (firstByte) => {
-        const type = extractNalUnitTypeForHevc(firstByte);
-        return type === 39 /* PREFIX_SEI_NUT */ || type === 40 /* SUFFIX_SEI_NUT */;
-      },
-      // HEVC VCL NAL unit types are 0..31; everything at or above VPS_NUT precedes them in an access unit.
-      isVcl: (firstByte) => extractNalUnitTypeForHevc(firstByte) < 32 /* VPS_NUT */,
-      iterateNalUnits: (packetData, decoderConfig) => iterateHevcNalUnits(packetData, decoderConfig),
-      concatNalUnits: (nalUnits, decoderConfig) => concatHevcNalUnits(nalUnits, decoderConfig)
-    }
-  };
-  var appendSeiMessage = (out, message) => {
-    for (const value of [message.payloadType, message.payload.length]) {
-      let remaining = value;
-      while (remaining >= 255) {
-        out.push(255);
-        remaining -= 255;
-      }
-      out.push(remaining);
-    }
-    out.push(...message.payload);
-  };
-  var buildSeiNalUnit = (messages, codec) => {
-    if (messages.length === 0) {
-      return null;
-    }
-    const rbsp = [];
-    for (const message of messages) {
-      appendSeiMessage(rbsp, message);
-    }
-    rbsp.push(128);
-    const framing = SEI_FRAMINGS[codec];
-    const escaped = addEmulationPreventionBytes(new Uint8Array(rbsp));
-    const nalUnit = new Uint8Array(framing.nalHeader.length + escaped.length);
-    nalUnit.set(framing.nalHeader, 0);
-    nalUnit.set(escaped, framing.nalHeader.length);
-    return nalUnit;
-  };
-  var spliceSeiNalUnit = (packetData, decoderConfig, codec, seiNalUnit) => {
-    const framing = SEI_FRAMINGS[codec];
-    const nalUnits = [];
-    let firstVclIndex = -1;
-    for (const location of framing.iterateNalUnits(packetData, decoderConfig)) {
-      const nalUnit = packetData.subarray(location.offset, location.offset + location.length);
-      if (firstVclIndex === -1 && framing.isVcl(nalUnit[0])) {
-        firstVclIndex = nalUnits.length;
-      }
-      nalUnits.push(nalUnit);
-    }
-    if (firstVclIndex === -1) {
-      throw new Error(`Expected a VCL NAL unit in the ${codec.toUpperCase()} access unit, found none.`);
-    }
-    nalUnits.splice(firstVclIndex, 0, seiNalUnit);
-    return framing.concatNalUnits(nalUnits, decoderConfig);
-  };
-
-  // src/closed-captions.ts
-  var USER_DATA_REGISTERED_ITU_T_T35 = 4;
-  var ATSC_CC_DATA_HEADER = [181, 0, 49, 71, 65, 57, 52, 3];
-  var buildClosedCaptionSeiMessage = (bytePairs) => {
-    const payload = new Uint8Array(ATSC_CC_DATA_HEADER.length + 2 + bytePairs.length * 3 + 1);
-    payload.set(ATSC_CC_DATA_HEADER, 0);
-    let pos = ATSC_CC_DATA_HEADER.length;
-    payload[pos++] = 192 | bytePairs.length;
-    payload[pos++] = 255;
-    for (const bytePair of bytePairs) {
-      payload[pos++] = 252 | bytePair.type;
-      payload[pos++] = bytePair.byte0;
-      payload[pos++] = bytePair.byte1;
-    }
-    payload[pos] = 255;
-    return { payloadType: USER_DATA_REGISTERED_ITU_T_T35, payload };
-  };
-
   // src/codec.ts
   var PCM_AUDIO_CODECS = [
     "pcm-s16",
@@ -1864,39 +1583,6 @@ var MediabunnyServer = (() => {
       };
     }
   };
-  var iterateNalUnitsInLengthPrefixed = function* (packetData, lengthSize) {
-    let offset = 0;
-    const dataView = new DataView(packetData.buffer, packetData.byteOffset, packetData.byteLength);
-    while (offset + lengthSize <= packetData.length) {
-      let nalUnitLength;
-      if (lengthSize === 1) {
-        nalUnitLength = dataView.getUint8(offset);
-      } else if (lengthSize === 2) {
-        nalUnitLength = dataView.getUint16(offset, false);
-      } else if (lengthSize === 3) {
-        nalUnitLength = getUint24(dataView, offset, false);
-      } else {
-        assert(lengthSize === 4);
-        nalUnitLength = dataView.getUint32(offset, false);
-      }
-      offset += lengthSize;
-      yield {
-        offset,
-        length: Math.min(nalUnitLength, packetData.length - offset)
-      };
-      offset += nalUnitLength;
-    }
-  };
-  var iterateAvcNalUnits = (packetData, decoderConfig) => {
-    if (decoderConfig.description) {
-      const bytes = toUint8Array(decoderConfig.description);
-      const lengthSizeMinusOne = bytes[4] & 3;
-      const lengthSize = lengthSizeMinusOne + 1;
-      return iterateNalUnitsInLengthPrefixed(packetData, lengthSize);
-    } else {
-      return iterateNalUnitsInAnnexB(packetData);
-    }
-  };
   var extractNalUnitTypeForAvc = (byte) => {
     return byte & 31;
   };
@@ -1915,68 +1601,7 @@ var MediabunnyServer = (() => {
     }
     return result.subarray(0, resultLength);
   };
-  var addEmulationPreventionBytes = (data) => {
-    const result = [];
-    let zeroCount = 0;
-    for (const byte of data) {
-      if (zeroCount === 2 && byte <= 3) {
-        result.push(3);
-        zeroCount = 0;
-      }
-      result.push(byte);
-      zeroCount = byte === 0 ? zeroCount + 1 : 0;
-    }
-    return new Uint8Array(result);
-  };
   var ANNEX_B_START_CODE = new Uint8Array([0, 0, 0, 1]);
-  var concatNalUnitsInAnnexB = (nalUnits) => {
-    const totalLength = nalUnits.reduce((a, b) => a + ANNEX_B_START_CODE.byteLength + b.byteLength, 0);
-    const result = new Uint8Array(totalLength);
-    let offset = 0;
-    for (const nalUnit of nalUnits) {
-      result.set(ANNEX_B_START_CODE, offset);
-      offset += ANNEX_B_START_CODE.byteLength;
-      result.set(nalUnit, offset);
-      offset += nalUnit.byteLength;
-    }
-    return result;
-  };
-  var concatNalUnitsInLengthPrefixed = (nalUnits, lengthSize) => {
-    const totalLength = nalUnits.reduce((a, b) => a + lengthSize + b.byteLength, 0);
-    const result = new Uint8Array(totalLength);
-    let offset = 0;
-    for (const nalUnit of nalUnits) {
-      const dataView = new DataView(result.buffer, result.byteOffset, result.byteLength);
-      switch (lengthSize) {
-        case 1:
-          dataView.setUint8(offset, nalUnit.byteLength);
-          break;
-        case 2:
-          dataView.setUint16(offset, nalUnit.byteLength, false);
-          break;
-        case 3:
-          setUint24(dataView, offset, nalUnit.byteLength, false);
-          break;
-        case 4:
-          dataView.setUint32(offset, nalUnit.byteLength, false);
-          break;
-      }
-      offset += lengthSize;
-      result.set(nalUnit, offset);
-      offset += nalUnit.byteLength;
-    }
-    return result;
-  };
-  var concatAvcNalUnits = (nalUnits, decoderConfig) => {
-    if (decoderConfig.description) {
-      const bytes = toUint8Array(decoderConfig.description);
-      const lengthSizeMinusOne = bytes[4] & 3;
-      const lengthSize = lengthSizeMinusOne + 1;
-      return concatNalUnitsInLengthPrefixed(nalUnits, lengthSize);
-    } else {
-      return concatNalUnitsInAnnexB(nalUnits);
-    }
-  };
   var extractAvcDecoderConfigurationRecord = (packetData) => {
     try {
       const spsUnits = [];
@@ -2323,24 +1948,6 @@ var MediabunnyServer = (() => {
     bitstream.skipBits(5);
     bitstream.skipBits(5);
   };
-  var concatHevcNalUnits = (nalUnits, decoderConfig) => {
-    if (decoderConfig.description) {
-      return concatNalUnitsInLengthPrefixed(nalUnits, hevcNalUnitLengthSize(decoderConfig.description));
-    } else {
-      return concatNalUnitsInAnnexB(nalUnits);
-    }
-  };
-  var hevcNalUnitLengthSize = (description) => {
-    const bytes = toUint8Array(description);
-    return (bytes[21] & 3) + 1;
-  };
-  var iterateHevcNalUnits = (packetData, decoderConfig) => {
-    if (decoderConfig.description) {
-      return iterateNalUnitsInLengthPrefixed(packetData, hevcNalUnitLengthSize(decoderConfig.description));
-    } else {
-      return iterateNalUnitsInAnnexB(packetData);
-    }
-  };
   var extractNalUnitTypeForHevc = (byte) => {
     return byte >> 1 & 63;
   };
@@ -3316,9 +2923,193 @@ var MediabunnyServer = (() => {
   var AC3_REGISTRATION_DESCRIPTOR = new Uint8Array([5, 4, 65, 67, 45, 51]);
   var EAC3_REGISTRATION_DESCRIPTOR = new Uint8Array([5, 4, 69, 65, 67, 51]);
 
+  // src/closed-captions.ts
+  var MAX_CLOSED_CAPTION_BYTE_PAIRS = 31;
+  var parseClosedCaptionBytePairs = (ccData) => {
+    const bytePairs = [];
+    for (let i = 0; i + 2 < ccData.length; i += 3) {
+      const header = ccData[i];
+      if ((header & 4) === 0) {
+        continue;
+      }
+      bytePairs.push({
+        type: header & 3,
+        byte0: ccData[i + 1],
+        byte1: ccData[i + 2]
+      });
+    }
+    if (bytePairs.length === 0 || bytePairs.length > MAX_CLOSED_CAPTION_BYTE_PAIRS) {
+      return null;
+    }
+    return bytePairs;
+  };
+
+  // packages/server/src/video-decoder.ts
+  var NodeAvVideoDecoder = class extends import_mediabunny2.CustomVideoDecoder {
+    constructor() {
+      super(...arguments);
+      this.codecContext = null;
+      // Bookkeeping to restore the original timing information
+      this.preciseTimings = [];
+    }
+    // eslint-disable-next-line @typescript-eslint/no-unused-vars
+    static supports(codec, config) {
+      return codec === "avc" || codec === "hevc" || codec === "vp8" || codec === "vp9" || codec === "av1";
+    }
+    async init() {
+      this.frame = new NodeAv3.Frame();
+      this.frame.alloc();
+      this.packet = new NodeAv3.Packet();
+      this.packet.alloc();
+    }
+    async initCodecContext(packet) {
+      assert(this.codecContext === null);
+      const codecId = CODEC_TO_CODEC_ID[this.codec];
+      assert(codecId !== void 0);
+      let codec;
+      if (this.codec === "vp9" && packet.sideData.alpha) {
+        codec = NodeAv3.Codec.findDecoderByName(NodeAv3.FF_DECODER_LIBVPX_VP9) ?? NodeAv3.Codec.findDecoder(codecId);
+      } else if (
+        // This check used to be "is not prefer-hardware", meaning it would default to using software decode. I
+        // didn't leave a comment for that back then so I actually don't know what it was for. I'm sure it was to
+        // work around an issue but, I don't know. Not using hardware decode at all by defaults feels wrong to me,
+        // so I changed it to what it is right now. If an issue is encountered, I can always change it.
+        this.config.hardwareAcceleration === "prefer-software" || this.codec === "av1"
+      ) {
+        codec = NodeAv3.Codec.findDecoder(codecId);
+      } else {
+        codec = await getHardwareDecoderCodec(codecId) ?? NodeAv3.Codec.findDecoder(codecId);
+      }
+      if (!codec) {
+        throw new Error(`Unable to obtain libav codec for '${this.codec}'.`);
+      }
+      const codecContext = new NodeAv3.CodecContext();
+      codecContext.allocContext3(codec);
+      this.pixelAspectRatio = simplifyRational({
+        num: (this.config.displayAspectWidth ?? this.config.codedWidth ?? 0) * (this.config.codedHeight ?? 0),
+        den: (this.config.displayAspectHeight ?? this.config.codedHeight ?? 0) * (this.config.codedWidth ?? 0) || 1
+      });
+      codecContext.width = this.config.codedWidth ?? 0;
+      codecContext.height = this.config.codedHeight ?? 0;
+      codecContext.codecType = NodeAv3.AVMEDIA_TYPE_VIDEO;
+      codecContext.codecId = codecId;
+      codecContext.extraData = this.config.description ? Buffer.from(toUint8Array(this.config.description)) : null;
+      codecContext.sampleAspectRatio = new NodeAv3.Rational(this.pixelAspectRatio.num, this.pixelAspectRatio.den);
+      codecContext.threadCount = 0;
+      const options = NodeAv3.Dictionary.fromObject({ strict: NodeAv3.FF_COMPLIANCE_STRICT });
+      const ret = await codecContext.open2(codec, options);
+      NodeAv3.FFmpegError.throwIfError(ret, "Open codec context");
+      this.codecContext = codecContext;
+    }
+    async decode(packet) {
+      if (this.codecContext === null) {
+        await this.initCodecContext(packet);
+      }
+      assert(this.codecContext);
+      this.packet.isKeyframe = packet.type === "key";
+      this.packet.data = Buffer.from(packet.data);
+      this.packet.timeBase = { num: 1, den: 1e6 };
+      this.packet.pts = BigInt(packet.microsecondTimestamp);
+      this.packet.dts = NodeAv3.AV_NOPTS_VALUE;
+      this.packet.duration = BigInt(packet.microsecondDuration);
+      if (packet.sideData.alpha) {
+        const matroskaBlockAdditional = Buffer.alloc(8 + packet.sideData.alpha.byteLength);
+        matroskaBlockAdditional[7] = 1;
+        matroskaBlockAdditional.set(packet.sideData.alpha, 8);
+        this.packet.addSideData(NodeAv3.AV_PKT_DATA_MATROSKA_BLOCKADDITIONAL, matroskaBlockAdditional);
+      }
+      const preciseTimingIndex = binarySearchLessOrEqual(
+        this.preciseTimings,
+        packet.microsecondTimestamp,
+        (x) => x.microsecondTimestamp
+      );
+      const existingEntry = preciseTimingIndex !== -1 ? this.preciseTimings[preciseTimingIndex] : null;
+      if (existingEntry && existingEntry.microsecondTimestamp === packet.microsecondTimestamp) {
+        if (existingEntry.timestamp !== packet.timestamp) {
+          existingEntry.timestampIsValid = false;
+        }
+        if (existingEntry.duration !== packet.duration) {
+          existingEntry.durationIsValid = false;
+        }
+      } else {
+        this.preciseTimings.splice(preciseTimingIndex + 1, 0, {
+          microsecondTimestamp: packet.microsecondTimestamp,
+          timestamp: packet.timestamp,
+          duration: packet.duration,
+          timestampIsValid: true,
+          durationIsValid: true
+        });
+        if (this.preciseTimings.length > 128) {
+          this.preciseTimings.shift();
+        }
+      }
+      const ret = await this.codecContext.sendPacket(this.packet);
+      NodeAv3.FFmpegError.throwIfError(ret, "Send packet");
+      this.packet.unref();
+      while (true) {
+        const receiveRet = await this.codecContext.receiveFrame(this.frame);
+        if (receiveRet === NodeAv3.AVERROR_EAGAIN || receiveRet === NodeAv3.AVERROR_EOF) {
+          break;
+        }
+        this.receiveFrame(receiveRet);
+      }
+    }
+    receiveFrame(ret) {
+      NodeAv3.FFmpegError.throwIfError(ret, "Receive frame");
+      this.frame.sampleAspectRatio = new NodeAv3.Rational(this.pixelAspectRatio.num, this.pixelAspectRatio.den);
+      let timestamp = Number(this.frame.pts) / 1e6;
+      let duration = Number(this.frame.duration) / 1e6;
+      const preciseTimingIndex = binarySearchLessOrEqual(
+        this.preciseTimings,
+        Number(this.frame.pts),
+        (x) => x.microsecondTimestamp
+      );
+      const entry = preciseTimingIndex !== -1 ? this.preciseTimings[preciseTimingIndex] : null;
+      if (entry && entry.microsecondTimestamp === Number(this.frame.pts)) {
+        if (entry.timestampIsValid) {
+          timestamp = entry.timestamp;
+        }
+        if (entry.durationIsValid) {
+          duration = entry.duration;
+        }
+      }
+      const clone = this.frame.clone();
+      if (!clone) {
+        throw new Error("Frame clone allocation failed.");
+      }
+      const ccData = this.frame.getSideData(NodeAv3.AV_FRAME_DATA_A53_CC);
+      this.onSample(new import_mediabunny2.VideoSample(new AvFrameVideoSampleResource(clone), {
+        timestamp,
+        duration,
+        // The captions ride in the bitstream, so the frame is where they surface
+        closedCaptions: ccData ? parseClosedCaptionBytePairs(toUint8Array(ccData)) : null
+      }));
+    }
+    async flush() {
+      if (!this.codecContext) {
+        return;
+      }
+      const ret = await this.codecContext.sendPacket(null);
+      NodeAv3.FFmpegError.throwIfError(ret, "Flush decoder");
+      while (true) {
+        const receiveRet = await this.codecContext.receiveFrame(this.frame);
+        if (receiveRet === NodeAv3.AVERROR_EAGAIN || receiveRet === NodeAv3.AVERROR_EOF) {
+          break;
+        }
+        this.receiveFrame(receiveRet);
+      }
+      this.codecContext.flushBuffers();
+    }
+    close() {
+      this.codecContext?.freeContext();
+      this.frame.free();
+      this.packet.free();
+    }
+  };
+
   // packages/server/src/video-encoder.ts
-  var ANNEX_B = { codec: "" };
-  var MAX_PENDING_CAPTION_FRAMES = 128;
+  var import_mediabunny3 = __require("mediabunny");
+  var NodeAv4 = __toESM(__require("node-av"));
   var PRORES_FOURCC_TO_PROFILE = {
     apco: NodeAv4.AV_PROFILE_PRORES_PROXY,
     apcs: NodeAv4.AV_PROFILE_PRORES_LT,
@@ -3437,11 +3228,7 @@ var MediabunnyServer = (() => {
   };
   var getQualityCap = (config) => {
     const { _maxBitrate, _bufferSize } = config;
-    return _maxBitrate === void 0 ? null : {
-      maxBitrate: _maxBitrate,
-      bufferSize: _bufferSize ?? 2 * _maxBitrate,
-      statesBufferSize: _bufferSize !== void 0
-    };
+    return _maxBitrate === void 0 ? null : { maxBitrate: _maxBitrate, bufferSize: _bufferSize };
   };
   var NodeAvVideoEncoder = class extends import_mediabunny3.CustomVideoEncoder {
     constructor() {
@@ -3454,7 +3241,6 @@ var MediabunnyServer = (() => {
       this.quantizer = null;
       // Bookkeeping to restore the original timing information
       this.preciseTimings = [];
-      this.pendingCaptions = /* @__PURE__ */ new Map();
     }
     static supports(codec, config) {
       const codecSupported = config.bitrateMode === "quantizer" ? codec === "avc" || codec === "hevc" || codec === "vp9" || codec === "av1" : codec === "avc" || codec === "hevc" || codec === "vp8" || codec === "vp9" || codec === "av1" || codec === "prores";
@@ -3626,22 +3412,21 @@ var MediabunnyServer = (() => {
         }
         const cap = getQualityCap(this.config);
         if (cap) {
+          const bufferSize = cap.bufferSize ?? 2 * cap.maxBitrate;
+          let honoursBufferSize = false;
           if (this.avCodec.name === "libx264" || this.avCodec.name === "libx265") {
             codecContext.setOption("crf", String(mapped));
             codecContext.rcMaxRate = BigInt(cap.maxBitrate);
-            codecContext.rcBufferSize = cap.bufferSize;
+            codecContext.rcBufferSize = bufferSize;
+            honoursBufferSize = true;
           } else if (this.avCodec.name === "libsvtav1") {
             codecContext.setOption("crf", String(Math.max(mapped, 1)));
             codecContext.rcMaxRate = BigInt(cap.maxBitrate);
-            codecContext.rcBufferSize = cap.bufferSize;
+            codecContext.rcBufferSize = bufferSize;
+            honoursBufferSize = true;
           } else if (this.avCodec.name === "libvpx-vp9" || this.avCodec.name === "libaom-av1") {
             codecContext.setOption("crf", String(mapped));
             codecContext.bitRate = BigInt(cap.maxBitrate);
-            if (this.avCodec.name === "libvpx-vp9" && cap.statesBufferSize) {
-              import_mediabunny3.Logging._warn(
-                "bufferSize has no effect with libvpx-vp9: its capped mode bounds the bitrate over the stream, not over a buffer, so maxBitrate alone is the cap. Encode avc, hevc or av1 for a buffer model."
-              );
-            }
           } else if (this.avCodec.name?.endsWith("_nvenc")) {
             codecContext.setOption("rc", "vbr");
             codecContext.setOption("cq", String(mapped));
@@ -3651,6 +3436,11 @@ var MediabunnyServer = (() => {
               `Encoder '${this.avCodec.name}' cannot be used for capped quantizer-based encoding.`
             );
           }
+          if (cap.bufferSize !== void 0 && !honoursBufferSize) {
+            import_mediabunny3.Logging._warn(
+              `bufferSize has no effect with '${this.avCodec.name}', whose capped mode bounds the bitrate rather than a buffer, so maxBitrate alone is the cap. Encode avc, hevc or av1 with their software encoders for a buffer model.`
+            );
+          }
         } else if (this.avCodec.name === "libx264" || this.avCodec.name === "libx265") {
           codecContext.setOption("crf", String(mapped));
         } else if (this.avCodec.name === "libvpx-vp9" || this.avCodec.name === "libaom-av1") {
@@ -3778,7 +3568,7 @@ var MediabunnyServer = (() => {
           this.preciseTimings.shift();
         }
       }
-      this.takeCaptionsFromFrame(frameToEncode);
+      frameToEncode.removeSideData(NodeAv4.AV_FRAME_DATA_A53_CC);
       const ret = await this.codecContext.sendFrame(frameToEncode);
       NodeAv4.FFmpegError.throwIfError(ret, "Send frame");
       while (true) {
@@ -3789,49 +3579,6 @@ var MediabunnyServer = (() => {
         this.receivePacket(receiveRet);
       }
     }
-    // Every encoder is relieved of the captions and they are written onto the packet instead: VideoToolbox
-    // errors out of sendFrame on a frame that has them ("Unexpected end of SEI NAL Unit parsing type") and
-    // libx265 silently drops them, so leaving it to the encoder means the same input loses captions, keeps
-    // them or fails outright depending on which one was selected.
-    takeCaptionsFromFrame(frame) {
-      if (this.codec !== "avc" && this.codec !== "hevc") {
-        return;
-      }
-      const sideData = frame.getSideData(NodeAv4.AV_FRAME_DATA_A53_CC);
-      if (!sideData) {
-        return;
-      }
-      frame.removeSideData(NodeAv4.AV_FRAME_DATA_A53_CC);
-      const bytes = toUint8Array(sideData);
-      const bytePairs = [];
-      for (let i = 0; i + 2 < bytes.length; i += 3) {
-        const header = bytes[i];
-        if ((header & 4) === 0) {
-          continue;
-        }
-        bytePairs.push({
-          type: header & 3,
-          byte0: bytes[i + 1],
-          byte1: bytes[i + 2]
-        });
-      }
-      if (bytePairs.length === 0) {
-        return;
-      }
-      this.pendingCaptions.set(Number(frame.pts), bytePairs);
-      while (this.pendingCaptions.size > MAX_PENDING_CAPTION_FRAMES) {
-        this.pendingCaptions.delete(this.pendingCaptions.keys().next().value);
-      }
-    }
-    /** The captions of the frame this packet was encoded from, matched on the presentation time they share. */
-    claimCaptionsFor(packetPts) {
-      const bytePairs = this.pendingCaptions.get(packetPts);
-      if (!bytePairs) {
-        return null;
-      }
-      this.pendingCaptions.delete(packetPts);
-      return bytePairs;
-    }
     receivePacket(ret) {
       assert(this.codecContext);
       NodeAv4.FFmpegError.throwIfError(ret, "Receive packet");
@@ -3839,15 +3586,16 @@ var MediabunnyServer = (() => {
         return;
       }
       let packetData = toUint8Array(this.packet.data);
-      let timestamp = Number(this.packet.pts) / 1e6;
+      const pts = Number(this.packet.pts);
+      let timestamp = pts / 1e6;
       let duration = Number(this.packet.duration) / 1e6;
       const preciseTimingIndex = binarySearchLessOrEqual(
         this.preciseTimings,
-        Number(this.packet.pts),
+        pts,
         (x) => x.microsecondTimestamp
       );
       const entry = preciseTimingIndex !== -1 ? this.preciseTimings[preciseTimingIndex] : null;
-      if (entry && entry.microsecondTimestamp === Number(this.packet.pts)) {
+      if (entry && entry.microsecondTimestamp === pts) {
         if (entry.timestampIsValid) {
           timestamp = entry.timestamp;
         }
@@ -3900,12 +3648,6 @@ var MediabunnyServer = (() => {
             decoderConfigDescription = serializedRecord;
           }
         }
-        const captions = this.claimCaptionsFor(Number(this.packet.pts));
-        if (captions) {
-          const seiNalUnit = buildSeiNalUnit([buildClosedCaptionSeiMessage(captions)], this.codec);
-          assert(seiNalUnit);
-          packetData = spliceSeiNalUnit(packetData, ANNEX_B, this.codec, seiNalUnit);
-        }
         if (!expectsAnnexB) {
           const NAL_UNIT_LENGTH_SIZE = 4;
           const nalUnits = [];