[FFmpeg-devel] libavformat/mov: fix multiple stsd handling of MOV files with edit list

Submitted by zhangjiejun1992@gmail.com on Dec. 2, 2016, 6:25 a.m.

Details

Message ID 20161202062507.83287-1-zhangjiejun1992@gmail.com
State New
Headers show

Commit Message

zhangjiejun1992@gmail.com Dec. 2, 2016, 6:25 a.m.
From: Jiejun Zhang <zhangjiejun1992@gmail.com>

When an edit list exists in a MOV file, counting by stscs no longer works because stscs' order is different from the actual timeline. This commit adds stsd-change markers to the actual timeline and changes stsd according to these markers.

Test sample: https://www.dropbox.com/s/qz0ort4znqq9jdy/attachment_video.mov?dl=0

Run `ffmpeg -i attachment_video.mov -f image2 frames%03d.png`. The git-master version cannot decode frame 327 to 330 because a wrong stsd is used. After applying this patch they are decoded correctly.

Signed-off-by: Jiejun Zhang <zhangjiejun1992@gmail.com>
---
 libavformat/isom.h |  13 +++++-
 libavformat/mov.c  | 125 +++++++++++++++++++++++++++++++++++++++++------------
 2 files changed, 109 insertions(+), 29 deletions(-)

Patch hide | download patch | download mbox

diff --git a/libavformat/isom.h b/libavformat/isom.h
index 02bfedd..141237a 100644
--- a/libavformat/isom.h
+++ b/libavformat/isom.h
@@ -115,6 +115,14 @@  typedef struct MOVFragmentIndex {
     MOVFragmentIndexItem *items;
 } MOVFragmentIndex;
 
+/**
+ * An entry for: stsd changes to `stsd_id` at sample `sample_index` (inclusive)
+ */
+typedef struct MOVStsdChangeEntry {
+    int64_t sample_index;
+    int stsd_id;
+} MOVStsdChangeEntry;
+
 typedef struct MOVStreamContext {
     AVIOContext *pb;
     int pb_is_copied;
@@ -128,12 +136,13 @@  typedef struct MOVStreamContext {
     MOVStts *ctts_data;
     unsigned int stsc_count;
     MOVStsc *stsc_data;
-    int stsc_index;
-    int stsc_sample;
     unsigned int stps_count;
     unsigned *stps_data;  ///< partial sync sample for mpeg-2 open gop
     MOVElst *elst_data;
     unsigned int elst_count;
+    MOVStsdChangeEntry *stsd_change_data;
+    int stsd_change_count;
+    int stsd_change_index;
     int ctts_index;
     int ctts_sample;
     unsigned int sample_size; ///< may contain value calculated from stsd or value from stsz atom
diff --git a/libavformat/mov.c b/libavformat/mov.c
index 88a79da..e5f7f72 100644
--- a/libavformat/mov.c
+++ b/libavformat/mov.c
@@ -2901,6 +2901,35 @@  static int64_t add_index_entry(AVStream *st, int64_t pos, int64_t timestamp,
     return index;
 }
 
+static int add_stsd_change_entry(MOVStreamContext *msc, int64_t sample_index, int stsd_id,
+        unsigned int* allocated_size)
+{
+    MOVStsdChangeEntry *entries;
+    const size_t min_size_needed = (msc->stsd_change_count + 1) * sizeof(MOVStsdChangeEntry);
+
+    const size_t requested_size =
+        min_size_needed > *allocated_size ?
+        FFMAX(min_size_needed, 2 * (*allocated_size)) :
+        min_size_needed;
+
+    if ((unsigned)msc->stsd_change_count + 1 >= UINT_MAX / sizeof(MOVStsdChangeEntry))
+        return -1;
+
+    entries = av_fast_realloc(msc->stsd_change_data,
+                              allocated_size,
+                              requested_size);
+
+    if (!entries)
+        return -1;
+
+    msc->stsd_change_data = entries;
+    entries[msc->stsd_change_count].sample_index = sample_index;
+    entries[msc->stsd_change_count].stsd_id = stsd_id;
+    msc->stsd_change_count++;
+
+    return 0;
+}
+
 /**
  * Rewrite timestamps of index entries in the range [end_index - frame_duration_buffer_size, end_index)
  * by subtracting end_ts successively by the amounts given in frame_duration_buffer.
@@ -2984,6 +3013,8 @@  static void mov_fix_index(MOVContext *mov, AVStream *st)
     int num_discarded_begin = 0;
     int first_non_zero_audio_edit = -1;
     int packet_skip_samples = 0;
+    int stsc_index = 0;
+    int stsc_sample = 0;
 
     if (!msc->elst_data || msc->elst_count <= 0 || nb_old <= 0) {
         return;
@@ -3012,6 +3043,11 @@  static void mov_fix_index(MOVContext *mov, AVStream *st)
 
     start_dts = edit_list_dts_entry_end;
 
+    msc->last_stsd_index = -1;
+    unsigned int stsd_change_data_allocated_size = 0;
+    msc->stsd_change_data = NULL;
+    msc->stsd_change_count = 0;
+
     while (get_edit_list_entry(mov, msc, edit_list_index, &edit_list_media_time,
                                &edit_list_duration, mov->time_scale)) {
         av_log(mov->fc, AV_LOG_DEBUG, "Processing st: %d, edit list %"PRId64" - media time: %"PRId64", duration: %"PRId64"\n",
@@ -3073,6 +3109,20 @@  static void mov_fix_index(MOVContext *mov, AVStream *st)
         }
         current = e_old + index;
 
+        // adjust stsd index
+        int time_sample = 0;
+        for (int i = 0; i < msc->stsc_count; i++) {
+            int next = time_sample + mov_get_stsc_samples(msc, i);
+            if (next > index) {
+                stsc_index = i;
+                stsc_sample = index - time_sample;
+                break;
+            }
+            time_sample = next;
+        }
+
+        av_log(mov->fc, AV_LOG_INFO, "stream %d, adjusted stsc_index: %d, stsc_sample: %d\n", st->index, stsc_index, stsc_sample);
+
         ctts_index_old = 0;
         ctts_sample_old = 0;
 
@@ -3176,12 +3226,36 @@  static void mov_fix_index(MOVContext *mov, AVStream *st)
                 }
             }
 
-            if (add_index_entry(st, current->pos, edit_list_dts_counter, current->size,
-                                current->min_distance, flags) == -1) {
+            int index_entry_index = add_index_entry(st, current->pos, edit_list_dts_counter, current->size,
+                    current->min_distance, flags);
+            if (index_entry_index == -1) {
                 av_log(mov->fc, AV_LOG_ERROR, "Cannot add index entry\n");
                 break;
             }
 
+            if (msc->stsc_data) {
+                if (msc->stsc_data[stsc_index].id > 0 &&
+                   msc->stsc_data[stsc_index].id - 1 < msc->stsd_count &&
+                   msc->stsc_data[stsc_index].id - 1 != msc->last_stsd_index) {
+                    msc->last_stsd_index = msc->stsc_data[stsc_index].id - 1;
+                    int ret = add_stsd_change_entry(msc, index_entry_index, msc->last_stsd_index, &stsd_change_data_allocated_size);
+                    if (ret < 0) {
+                        av_log(mov->fc, AV_LOG_ERROR, "Cannot add stsd change entry\n");
+                        break;
+                    }
+                    av_log(mov->fc, AV_LOG_INFO, "added a stsd change entry sample_index: %d, stsd_index:%d\n",
+                            index_entry_index,
+                            msc->last_stsd_index);
+                }
+
+                stsc_sample++;
+                if (stsc_index < msc->stsc_count - 1 &&
+                    mov_get_stsc_samples(msc, stsc_index) == stsc_sample) {
+                    stsc_index++;
+                    stsc_sample = 0;
+                }
+            }
+
             // Only start incrementing DTS in frame_duration amounts, when we encounter a frame in edit list.
             if (edit_list_start_encountered > 0) {
                 edit_list_dts_counter = edit_list_dts_counter + frame_duration;
@@ -3206,6 +3280,7 @@  static void mov_fix_index(MOVContext *mov, AVStream *st)
             }
         }
     }
+    msc->last_stsd_index = 0;
     // Update av stream length
     st->duration = edit_list_dts_entry_end - start_dts;
 
@@ -5385,6 +5460,7 @@  static int mov_read_close(AVFormatContext *s)
         av_freep(&sc->stts_data);
         av_freep(&sc->stps_data);
         av_freep(&sc->elst_data);
+        av_freep(&sc->stsd_change_data);
         av_freep(&sc->rap_group);
         av_freep(&sc->display_matrix);
 
@@ -5807,9 +5883,6 @@  static int mov_change_extradata(MOVStreamContext *sc, AVPacket *pkt)
     uint8_t *side, *extradata;
     int extradata_size;
 
-    /* Save the current index. */
-    sc->last_stsd_index = sc->stsc_data[sc->stsc_index].id - 1;
-
     /* Notify the decoder that extradata changed. */
     extradata_size = sc->extradata_size[sc->last_stsd_index];
     extradata = sc->extradata[sc->last_stsd_index];
@@ -5923,21 +5996,22 @@  static int mov_read_packet(AVFormatContext *s, AVPacket *pkt)
     pkt->pos = sample->pos;
 
     /* Multiple stsd handling. */
-    if (sc->stsc_data) {
-        /* Keep track of the stsc index for the given sample, then check
-        * if the stsd index is different from the last used one. */
-        sc->stsc_sample++;
-        if (sc->stsc_index < sc->stsc_count - 1 &&
-            mov_get_stsc_samples(sc, sc->stsc_index) == sc->stsc_sample) {
-            sc->stsc_index++;
-            sc->stsc_sample = 0;
-        /* Do not check indexes after a switch. */
-        } else if (sc->stsc_data[sc->stsc_index].id > 0 &&
-                   sc->stsc_data[sc->stsc_index].id - 1 < sc->stsd_count &&
-                   sc->stsc_data[sc->stsc_index].id - 1 != sc->last_stsd_index) {
+    if (sc->stsd_change_data) {
+        if (sc->stsd_change_index + 1 < sc->stsd_change_count &&
+                // sc->current_sample is already the next sample, so subtract 1
+                sc->stsd_change_data[sc->stsd_change_index + 1].sample_index == sc->current_sample - 1) {
+            ++sc->stsd_change_index;
+        }
+        if (sc->last_stsd_index != sc->stsd_change_data[sc->stsd_change_index].stsd_id) {
+            av_log(mov->fc, AV_LOG_INFO, "stsd id changed from %d to %d, at sample %d\n",
+                    sc->last_stsd_index,
+                    sc->stsd_change_data[sc->stsd_change_index].stsd_id,
+                    sc->current_sample - 1);
+            sc->last_stsd_index = sc->stsd_change_data[sc->stsd_change_index].stsd_id;
             ret = mov_change_extradata(sc, pkt);
-            if (ret < 0)
+            if (ret < 0) {
                 return ret;
+            }
         }
     }
 
@@ -6012,16 +6086,13 @@  static int mov_seek_stream(AVFormatContext *s, AVStream *st, int64_t timestamp,
         }
     }
 
-    /* adjust stsd index */
-    time_sample = 0;
-    for (i = 0; i < sc->stsc_count; i++) {
-        int next = time_sample + mov_get_stsc_samples(sc, i);
-        if (next > sc->current_sample) {
-            sc->stsc_index = i;
-            sc->stsc_sample = sc->current_sample - time_sample;
-            break;
+    /* adjust stsd change index */
+    for (i = 0; i < sc->stsd_change_count; i++) {
+        if (sc->current_sample >= sc->stsd_change_data[i].sample_index &&
+                (i + 1 == sc->stsd_change_count ||
+                 sc->current_sample < sc->stsd_change_data[i + 1].sample_index)) {
+            sc->stsd_change_index = i;
         }
-        time_sample = next;
     }
 
     ret = mov_seek_auxiliary_info(s, sc);