bugfix: Accept HEVC streams with temporal sub-layers

media/codec_signaling/src/codec_signaling.c

@@ -548,8 +548,12 @@ static moq_result_t build_avcc(const moq_codec_init_data_cfg_t *cfg,
 /*
  * Parse the fields an hvcC header needs from an HEVC SPS NAL (payload after
  * the 2-byte NAL header), per ISO/IEC 23008-2 7.3.2.2.1. The 12-byte general
- * profile_tier_level maps directly onto hvcC bytes 1..12. Sub-layer PTL
- * (sps_max_sub_layers_minus1 > 0) is not parsed.
+ * profile_tier_level maps directly onto hvcC bytes 1..12.
+ *
+ * When sps_max_sub_layers_minus1 > 0 the SPS carries sub-layer PTL syntax
+ * between the general PTL and sps_seq_parameter_set_id. hvcC needs none of it,
+ * but it is variable-length and has to be stepped over to reach
+ * chroma_format_idc and the bit depths.
  */
 static moq_result_t parse_hevc_sps(const uint8_t *sps, size_t sps_len,
                                    uint8_t ptl[12],
@@ -562,7 +566,9 @@ static moq_result_t parse_hevc_sps(const uint8_t *sps, size_t sps_len,
     if (sps_len < 3) {
         return MOQ_ERR_PROTO;
     }
-    uint8_t rbsp[64];
+    /* Large enough for the general PTL plus a full set of sub-layer PTLs
+     * (7 x 96 bits) and the fields we read after them. */
+    uint8_t rbsp[256];
     size_t n = deemulate(sps + 2, sps_len - 2, rbsp, sizeof(rbsp));
     if (n < 13) {
         return MOQ_ERR_PROTO;
@@ -574,10 +580,37 @@ static moq_result_t parse_hevc_sps(const uint8_t *sps, size_t sps_len,
     uint32_t tid_nest = br_u(&br, 1);   /* sps_temporal_id_nesting     */
 
     memcpy(ptl, rbsp + 1, 12);          /* general profile_tier_level  */
+    br.bitpos = 8 + 12 * 8;
+
+    /* profile_tier_level() sub-layer part, 23008-2 7.3.3. The general half is
+     * the fixed 12 bytes already copied above; what follows is present only
+     * when there is more than one temporal sub-layer. max_sub is 3 bits, so
+     * it never exceeds 7 and the arrays below are always in range. */
     if (max_sub > 0) {
-        return MOQ_ERR_UNSUPPORTED;     /* sub-layer PTL not handled   */
+        bool profile_present[7] = { false };
+        bool level_present[7] = { false };
+        for (uint32_t i = 0; i < max_sub; i++) {
+            profile_present[i] = br_u(&br, 1) != 0;
+            level_present[i] = br_u(&br, 1) != 0;
+        }
+        for (uint32_t i = max_sub; i < 8; i++) {
+            br_u(&br, 2);               /* reserved_zero_2bits         */
+        }
+        for (uint32_t i = 0; i < max_sub; i++) {
+            if (profile_present[i]) {
+                /* Same layout as the general PTL minus its 8-bit level. */
+                br_u(&br, 32);
+                br_u(&br, 32);
+                br_u(&br, 24);
+            }
+            if (level_present[i]) {
+                br_u(&br, 8);           /* sub_layer_level_idc         */
+            }
+        }
+        if (br.overrun) {
+            return MOQ_ERR_PROTO;
+        }
     }
-    br.bitpos = 8 + 12 * 8;
 
     br_ue(&br);                         /* sps_seq_parameter_set_id    */
     uint32_t chroma = br_ue(&br);

media/codec_signaling/tests/test_codec_signaling.c

@@ -665,6 +665,46 @@ int main(void)
         CHECK(out_len == sizeof(expect) && memcmp(out, expect, sizeof(expect)) == 0);
     }
 
+    /* -- HEVC Annex B with temporal sub-layers -> hvcC -------------- */
+    {
+        /* VPS + SPS + PPS captured from Apple's low-latency HEVC encoder
+         * (com.apple.videotoolbox.videoencoder.hevc.rtvc) at 1280x720. Its SPS
+         * carries sps_max_sub_layers_minus1 = 1, which the parser used to
+         * reject outright with MOQ_ERR_UNSUPPORTED -- no hvcC meant no video
+         * track was ever announced, so a stream encoded this way came out
+         * audio-only. The sub-layer PTL is skipped, not interpreted. */
+        static const uint8_t annexb[] = {
+            0x00, 0x00, 0x00, 0x01, 0x40, 0x01, 0x0c, 0x03, 0xff, 0xff, 0x01, 0x60,
+            0x00, 0x00, 0x03, 0x00, 0xb0, 0x00, 0x00, 0x03, 0x00, 0x00, 0x03, 0x00,
+            0x5d, 0x00, 0x00, 0x1b, 0x02, 0x40, 0x00, 0x00, 0x00, 0x01, 0x42, 0x01,
+            0x03, 0x01, 0x60, 0x00, 0x00, 0x03, 0x00, 0xb0, 0x00, 0x00, 0x03, 0x00,
+            0x00, 0x03, 0x00, 0x5d, 0x00, 0x00, 0xa0, 0x02, 0x80, 0x80, 0x2d, 0x16,
+            0x20, 0x6e, 0xe4, 0x52, 0x32, 0xe7, 0xe1, 0x3d, 0x0b, 0xea, 0x1b, 0xd5,
+            0x29, 0xa8, 0x10, 0x10, 0x10, 0x1f, 0xc2, 0x01, 0x04, 0x00, 0x00, 0x00,
+            0x01, 0x44, 0x01, 0xc0, 0x72, 0xf0, 0x5b, 0x24,
+        };
+        /* general profile_tier_level, hvcC bytes 1..12: Main profile, Main
+         * tier, general_level_idc 0x5d (level 4.0). */
+        static const uint8_t ptl[] = {
+            0x01, 0x60, 0x00, 0x00, 0x00, 0xb0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x5d,
+        };
+        moq_codec_init_data_cfg_t cfg;
+        moq_codec_init_data_cfg_init(&cfg);
+        cfg.source_format = MOQ_CODEC_SOURCE_HEVC_ANNEXB;
+        cfg.source = bytes(annexb, sizeof(annexb));
+
+        uint8_t out[256];
+        size_t out_len = 0;
+        moq_result_t rc = moq_codec_init_data_build(&cfg, out, sizeof(out), &out_len);
+        CHECK(rc == MOQ_OK);
+        CHECK(out_len >= 23);
+        CHECK(out[0] == 0x01);                            /* configurationVersion */
+        CHECK(memcmp(out + 1, ptl, sizeof(ptl)) == 0);
+        /* byte 21: constantFrameRate(2)=0 | numTemporalLayers(3)=2 |
+         * temporalIdNested(1)=1 | lengthSizeMinusOne(2)=3 */
+        CHECK(out[21] == 0x17);
+    }
+
     /* -- AV1 OBU -> av1C (header vs ffmpeg; seq header verbatim) ---- */
     {
         /* Temporal delimiter + sequence header OBU from SVT-AV1. */