summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorRobert Swain <robert.swain@collabora.co.uk>2009-10-26 22:42:36 +0000
committerRobert Swain <robert.swain@collabora.co.uk>2009-11-23 16:29:14 +0100
commit6a6d2c4970d3b030cd440d3d5b3a3016a0b8ff99 (patch)
treebf69e0d184aba556ee50c1a6519996ac4f9b044f
parentab61fb22f6fb2ddf491e3e1637abb8c101640314 (diff)
qtdemux: Move variables into context in preparation for refactorisation
-rw-r--r--gst/qtdemux/qtdemux.c559
1 files changed, 261 insertions, 298 deletions
diff --git a/gst/qtdemux/qtdemux.c b/gst/qtdemux/qtdemux.c
index 30d836569..eb9fbe92e 100644
--- a/gst/qtdemux/qtdemux.c
+++ b/gst/qtdemux/qtdemux.c
@@ -252,6 +252,28 @@ struct _QtDemuxStream
gboolean send_global_tags;
GstEvent *pending_event;
+
+ GstByteReader stco;
+ GstByteReader stsz;
+ GstByteReader stsc;
+ GstByteReader stts;
+ GstByteReader stss;
+ GstByteReader stps;
+ GstByteReader ctts;
+
+ gboolean chunks_are_chunks;
+ /* stco */
+ guint co_size;
+ /* stsz */
+ guint32 sample_size; /* 0 means variable sizes are stored in stsz */
+ /* stsc */
+ guint32 n_samples_per_chunk;
+ /* stts */
+ guint32 n_sample_times;
+ /* stss */
+ guint32 n_sample_syncs;
+ /* stps */
+ guint32 n_sample_partial_syncs;
};
enum QtDemuxState
@@ -3718,409 +3740,349 @@ static gboolean
qtdemux_parse_samples (GstQTDemux * qtdemux, QtDemuxStream * stream,
GNode * stbl)
{
- GstByteReader co_reader;
- GstByteReader stsz;
- GstByteReader stsc;
- GstByteReader stts;
- GstByteReader ctts;
- guint32 sample_size = 0;
- guint32 n_samples = 0;
- guint32 n_samples_per_chunk = 0;
- int sample_index;
- QtDemuxSample *samples;
+ int sample_index = 0;
gint i, j, k;
- int index;
- guint64 timestamp;
- guint co_size;
+ int index = 0;
+ guint64 timestamp = 0;
/* sample to chunk */
- if (!qtdemux_tree_get_child_by_type_full (stbl, FOURCC_stsc, &stsc))
+ if (!qtdemux_tree_get_child_by_type_full (stbl, FOURCC_stsc, &stream->stsc))
goto corrupt_file;
/* sample size */
- if (!qtdemux_tree_get_child_by_type_full (stbl, FOURCC_stsz, &stsz))
+ if (!qtdemux_tree_get_child_by_type_full (stbl, FOURCC_stsz, &stream->stsz))
goto corrupt_file;
/* chunk offsets */
- if (qtdemux_tree_get_child_by_type_full (stbl, FOURCC_stco, &co_reader))
- co_size = sizeof (guint32);
- else if (qtdemux_tree_get_child_by_type_full (stbl, FOURCC_co64, &co_reader))
- co_size = sizeof (guint64);
+ if (qtdemux_tree_get_child_by_type_full (stbl, FOURCC_stco, &stream->stco))
+ stream->co_size = sizeof (guint32);
+ else if (qtdemux_tree_get_child_by_type_full (stbl, FOURCC_co64,
+ &stream->stco))
+ stream->co_size = sizeof (guint64);
else
goto corrupt_file;
/* sample time */
- if (!qtdemux_tree_get_child_by_type_full (stbl, FOURCC_stts, &stts))
+ if (!qtdemux_tree_get_child_by_type_full (stbl, FOURCC_stts, &stream->stts))
+ goto corrupt_file;
+
+ if (!gst_byte_reader_skip (&stream->stsz, 1 + 3) ||
+ !gst_byte_reader_get_uint32_be (&stream->stsz, &stream->sample_size))
goto corrupt_file;
- if (!gst_byte_reader_skip (&stsz, 1 + 3) ||
- !gst_byte_reader_get_uint32_be (&stsz, &sample_size))
+ /* skip version + flags */
+ if (!gst_byte_reader_skip (&stream->stco, 1 + 3))
goto corrupt_file;
- if (sample_size == 0 || stream->sampled) {
- /* skip version, flags, number of entries */
- if (!gst_byte_reader_skip (&co_reader, 1 + 3 + 4))
+ stream->chunks_are_chunks = !stream->sample_size || stream->sampled;
+ if (stream->chunks_are_chunks) {
+ /* skip number of entries */
+ if (!gst_byte_reader_skip (&stream->stco, 4))
goto corrupt_file;
- if (!gst_byte_reader_get_uint32_be (&stsz, &n_samples))
+ if (!gst_byte_reader_get_uint32_be (&stream->stsz, &stream->n_samples))
goto corrupt_file;
+ } else {
+ /* treat chunks as samples */
+ if (!gst_byte_reader_get_uint32_be (&stream->stco, &stream->n_samples))
+ goto corrupt_file;
+ }
- if (n_samples == 0)
- goto no_samples;
+ if (!stream->n_samples)
+ goto no_samples;
- GST_DEBUG_OBJECT (qtdemux, "stsz sample_size 0, allocating n_samples %u "
- "(%u MB)", n_samples,
- (guint) (n_samples * sizeof (QtDemuxSample)) >> 20);
+ if (stream->chunks_are_chunks)
+ GST_DEBUG_OBJECT (qtdemux, "stsz sample_size 0");
- if (n_samples >= QTDEMUX_MAX_SAMPLE_INDEX_SIZE / sizeof (QtDemuxSample))
- goto index_too_big;
+ GST_DEBUG_OBJECT (qtdemux, "allocating n_samples %u (%u MB)",
+ stream->n_samples,
+ (guint) (stream->n_samples * sizeof (QtDemuxSample)) >> 20);
- samples = g_try_new0 (QtDemuxSample, n_samples);
- if (samples == NULL)
- goto out_of_memory;
+ if (stream->n_samples >=
+ QTDEMUX_MAX_SAMPLE_INDEX_SIZE / sizeof (QtDemuxSample))
+ goto index_too_big;
- stream->n_samples = n_samples;
- stream->samples = samples;
+ stream->samples = g_try_new0 (QtDemuxSample, stream->n_samples);
+ if (!stream->samples)
+ goto out_of_memory;
+ if (stream->chunks_are_chunks) {
/* set the sample sizes */
- if (sample_size == 0) {
+ if (stream->sample_size == 0) {
/* different sizes for each sample */
- if (!qt_atom_parser_has_chunks (&stsz, n_samples, 4))
+ if (!qt_atom_parser_has_chunks (&stream->stsz, stream->n_samples, 4))
goto corrupt_file;
- for (i = 0; i < n_samples; i++) {
- samples[i].size = gst_byte_reader_get_uint32_be_unchecked (&stsz);
- GST_LOG_OBJECT (qtdemux, "sample %d has size %u", i, samples[i].size);
+ for (i = 0; i < stream->n_samples; i++) {
+ stream->samples[i].size =
+ gst_byte_reader_get_uint32_be_unchecked (&stream->stsz);
+ GST_LOG_OBJECT (qtdemux, "sample %d has size %u", i,
+ stream->samples[i].size);
}
} else {
/* samples have the same size */
- GST_LOG_OBJECT (qtdemux, "all samples have size %u", sample_size);
- for (i = 0; i < n_samples; i++)
- samples[i].size = sample_size;
+ GST_LOG_OBJECT (qtdemux, "all samples have size %u", stream->sample_size);
+ for (i = 0; i < stream->n_samples; i++)
+ stream->samples[i].size = stream->sample_size;
}
+ }
- /* set the sample offsets in the file */
- if (!gst_byte_reader_skip (&stsc, 1 + 3) ||
- !gst_byte_reader_get_uint32_be (&stsc, &n_samples_per_chunk))
- goto corrupt_file;
+ /* set the sample offsets in the file */
+ if (!gst_byte_reader_skip (&stream->stsc, 1 + 3) ||
+ !gst_byte_reader_get_uint32_be (&stream->stsc,
+ &stream->n_samples_per_chunk))
+ goto corrupt_file;
- if (!qt_atom_parser_has_chunks (&stsc, n_samples_per_chunk, 12))
- goto corrupt_file;
+ if (!stream->chunks_are_chunks) {
+ GST_DEBUG_OBJECT (qtdemux, "n_samples_per_chunk %u",
+ stream->n_samples_per_chunk);
+ }
- index = 0;
- for (i = 0; i < n_samples_per_chunk; i++) {
- GstByteReader co_chunk;
- guint32 first_chunk, last_chunk;
- guint32 samples_per_chunk;
+ if (!qt_atom_parser_has_chunks (&stream->stsc, stream->n_samples_per_chunk,
+ 12))
+ goto corrupt_file;
+
+ for (i = 0; i < stream->n_samples_per_chunk; i++) {
+ GstByteReader co_chunk;
+ guint32 first_chunk, last_chunk;
+ guint32 samples_per_chunk;
+
+ first_chunk = gst_byte_reader_get_uint32_be_unchecked (&stream->stsc);
+ samples_per_chunk = gst_byte_reader_get_uint32_be_unchecked (&stream->stsc);
+ gst_byte_reader_skip_unchecked (&stream->stsc, 4);
- first_chunk = gst_byte_reader_get_uint32_be_unchecked (&stsc);
- samples_per_chunk = gst_byte_reader_get_uint32_be_unchecked (&stsc);
- gst_byte_reader_skip_unchecked (&stsc, 4);
+ /* chunk numbers are counted from 1 it seems */
+ if (G_UNLIKELY (first_chunk == 0))
+ goto corrupt_file;
+ else
+ --first_chunk;
- /* chunk numbers are counted from 1 it seems */
- if (G_UNLIKELY (first_chunk == 0))
+ /* the last chunk of each entry is calculated by taking the first chunk
+ * of the next entry; except if there is no next, where we fake it with
+ * INT_MAX */
+ if (G_UNLIKELY (i == (stream->n_samples_per_chunk - 1))) {
+ last_chunk = G_MAXUINT32;
+ } else {
+ last_chunk = gst_byte_reader_peek_uint32_be_unchecked (&stream->stsc);
+ if (G_UNLIKELY (last_chunk == 0))
goto corrupt_file;
else
- --first_chunk;
+ --last_chunk;
+ }
- /* the last chunk of each entry is calculated by taking the first chunk
- * of the next entry; except if there is no next, where we fake it with
- * INT_MAX */
- if (G_UNLIKELY (i == n_samples_per_chunk - 1)) {
- last_chunk = G_MAXUINT32;
- } else {
- last_chunk = gst_byte_reader_peek_uint32_be_unchecked (&stsc);
- if (G_UNLIKELY (last_chunk == 0))
- goto corrupt_file;
- else
- --last_chunk;
- }
+ GST_LOG_OBJECT (qtdemux,
+ "entry %d has first_chunk %d, last_chunk %d, samples_per_chunk %d", i,
+ first_chunk, last_chunk, samples_per_chunk);
- if (G_UNLIKELY (last_chunk < first_chunk))
- goto corrupt_file;
+ if (G_UNLIKELY (last_chunk < first_chunk))
+ goto corrupt_file;
- if (last_chunk != G_MAXUINT32) {
- if (!qt_atom_parser_peek_sub (&co_reader, first_chunk * co_size,
- (last_chunk - first_chunk) * co_size, &co_chunk))
- goto corrupt_file;
- } else {
- co_chunk = co_reader;
- if (!gst_byte_reader_skip (&co_chunk, first_chunk * co_size))
- goto corrupt_file;
- }
+ if (last_chunk != G_MAXUINT32) {
+ if (!qt_atom_parser_peek_sub (&stream->stco,
+ first_chunk * stream->co_size,
+ (last_chunk - first_chunk) * stream->co_size, &co_chunk))
+ goto corrupt_file;
+ } else {
+ co_chunk = stream->stco;
+ if (!gst_byte_reader_skip (&co_chunk, first_chunk * stream->co_size))
+ goto corrupt_file;
+ }
+ if (stream->chunks_are_chunks) {
for (j = first_chunk; j < last_chunk; j++) {
guint64 chunk_offset;
- if (!qt_atom_parser_get_offset (&co_chunk, co_size, &chunk_offset))
+ if (!qt_atom_parser_get_offset (&co_chunk, stream->co_size,
+ &chunk_offset))
goto corrupt_file;
for (k = 0; k < samples_per_chunk; k++) {
GST_LOG_OBJECT (qtdemux, "Creating entry %d with offset %"
G_GUINT64_FORMAT, index, chunk_offset);
- samples[index].offset = chunk_offset;
- chunk_offset += samples[index].size;
+ stream->samples[index].offset = chunk_offset;
+ chunk_offset += stream->samples[index].size;
index++;
- if (G_UNLIKELY (index >= n_samples))
+ if (G_UNLIKELY (index >= stream->n_samples))
goto done2;
}
}
- }
- done2:
- {
- guint32 n_sample_times = 0;
- guint32 time;
-
- if (!gst_byte_reader_skip (&stts, 4))
- goto corrupt_file;
- if (!gst_byte_reader_get_uint32_be (&stts, &n_sample_times))
- goto corrupt_file;
- GST_LOG_OBJECT (qtdemux, "%u timestamp blocks", n_sample_times);
-
- /* make sure there's enough data */
- if (!qt_atom_parser_has_chunks (&stts, n_sample_times, 2 * 4))
- goto corrupt_file;
-
- timestamp = 0;
- stream->min_duration = 0;
- time = 0;
- index = 0;
- for (i = 0; i < n_sample_times; i++) {
- guint32 n;
- guint32 duration;
-
- n = gst_byte_reader_get_uint32_be_unchecked (&stts);
- duration = gst_byte_reader_get_uint32_be_unchecked (&stts);
- GST_LOG_OBJECT (qtdemux, "block %d, %u timestamps, duration %u ", i, n,
- duration);
-
- /* take first duration for fps */
- if (G_UNLIKELY (stream->min_duration == 0))
- stream->min_duration = duration;
+ } else {
+ for (j = first_chunk; j < last_chunk; j++) {
+ if (j >= stream->n_samples)
+ goto done;
- for (j = 0; j < n; j++) {
- GST_DEBUG_OBJECT (qtdemux,
- "sample %d: index %d, timestamp %" GST_TIME_FORMAT, index, j,
- GST_TIME_ARGS (timestamp));
+ stream->samples[j].offset =
+ qt_atom_parser_get_offset_unchecked (&co_chunk, stream->co_size);
- samples[index].timestamp = timestamp;
- /* add non-scaled values to avoid rounding errors */
- time += duration;
- timestamp =
- gst_util_uint64_scale (time, GST_SECOND, stream->timescale);
- samples[index].duration = timestamp - samples[index].timestamp;
+ GST_LOG_OBJECT (qtdemux, "Created entry %d with offset "
+ "%" G_GUINT64_FORMAT, j, stream->samples[j].offset);
- index++;
- if (G_UNLIKELY (index >= n_samples))
- goto done3;
+ if (stream->samples_per_frame * stream->bytes_per_frame) {
+ stream->samples[j].size = (samples_per_chunk * stream->n_channels) /
+ stream->samples_per_frame * stream->bytes_per_frame;
+ } else {
+ stream->samples[j].size = samples_per_chunk;
}
- }
- /* fill up empty timestamps with the last timestamp, this can happen when
- * the last samples do not decode and so we don't have timestamps for them.
- * We however look at the last timestamp to estimate the track length so we
- * need something in here. */
- for (; index < n_samples; index++) {
- GST_DEBUG_OBJECT (qtdemux,
- "fill sample %d: timestamp %" GST_TIME_FORMAT, index,
- GST_TIME_ARGS (timestamp));
- samples[index].timestamp = timestamp;
- samples[index].duration = -1;
- }
- }
- done3:
- {
- /* FIXME: split this block out into a separate function */
- GstByteReader stss, stps;
- /* sample sync, can be NULL */
- if (qtdemux_tree_get_child_by_type_full (stbl, FOURCC_stss, &stss)) {
- guint32 n_sample_syncs = 0;
-
- /* mark keyframes */
- if (!gst_byte_reader_skip (&stss, 4))
- goto corrupt_file;
- if (!gst_byte_reader_get_uint32_be (&stss, &n_sample_syncs))
- goto corrupt_file;
+ GST_DEBUG_OBJECT (qtdemux, "sample %d: timestamp %" GST_TIME_FORMAT
+ ", size %u", j, GST_TIME_ARGS (timestamp), stream->samples[j].size);
- if (n_sample_syncs == 0) {
- stream->all_keyframe = TRUE;
- } else {
- /* make sure there's enough data */
- if (!qt_atom_parser_has_chunks (&stss, n_sample_syncs, 4))
- goto corrupt_file;
- for (i = 0; i < n_sample_syncs; i++) {
- /* note that the first sample is index 1, not 0 */
- index = gst_byte_reader_get_uint32_be_unchecked (&stss);
- if (G_LIKELY (index > 0 && index <= n_samples))
- samples[index - 1].keyframe = TRUE;
- }
- }
- /* stps marks partial sync frames like open GOP I-Frames */
- if (qtdemux_tree_get_child_by_type_full (stbl, FOURCC_stps, &stps)) {
- guint32 n_sample_syncs = 0;
+ stream->samples[j].timestamp = timestamp;
+ sample_index += samples_per_chunk;
- if (!gst_byte_reader_skip (&stps, 4))
- goto corrupt_file;
- if (!gst_byte_reader_get_uint32_be (&stps, &n_sample_syncs))
- goto corrupt_file;
+ timestamp = gst_util_uint64_scale (sample_index,
+ GST_SECOND, stream->timescale);
+ stream->samples[j].duration = timestamp - stream->samples[j].timestamp;
- /* if there are no entries, the stss table contains the real
- * sync samples */
- if (n_sample_syncs != 0) {
- /* make sure there's enough data */
- if (!qt_atom_parser_has_chunks (&stps, n_sample_syncs, 4))
- goto corrupt_file;
- for (i = 0; i < n_sample_syncs; i++) {
- /* note that the first sample is index 1, not 0 */
- index = gst_byte_reader_get_uint32_be_unchecked (&stps);
- if (G_LIKELY (index > 0 && index <= n_samples))
- samples[index - 1].keyframe = TRUE;
- }
- }
- }
- } else {
- /* no stss, all samples are keyframes */
- stream->all_keyframe = TRUE;
+ stream->samples[j].keyframe = TRUE;
}
}
- } else {
- GST_DEBUG_OBJECT (qtdemux,
- "stsz sample_size %d != 0, treating chunks as samples", sample_size);
+ }
- /* skip version + flags */
- if (!gst_byte_reader_skip (&co_reader, 1 + 3))
- goto corrupt_file;
+ if (!stream->chunks_are_chunks)
+ goto ctts;
+done2:
+ {
+ guint32 time;
- /* treat chunks as samples */
- if (!gst_byte_reader_get_uint32_be (&co_reader, &n_samples))
+ if (!gst_byte_reader_skip (&stream->stts, 4))
goto corrupt_file;
+ if (!gst_byte_reader_get_uint32_be (&stream->stts, &stream->n_sample_times))
+ goto corrupt_file;
+ GST_LOG_OBJECT (qtdemux, "%u timestamp blocks", stream->n_sample_times);
- if (n_samples == 0)
- goto no_samples;
-
- GST_DEBUG_OBJECT (qtdemux, "allocating n_samples %u (%u MB)", n_samples,
- (guint) (n_samples * sizeof (QtDemuxSample)) >> 20);
-
- if (n_samples >= QTDEMUX_MAX_SAMPLE_INDEX_SIZE / sizeof (QtDemuxSample))
- goto index_too_big;
-
- samples = g_try_new0 (QtDemuxSample, n_samples);
- if (samples == NULL)
- goto out_of_memory;
-
- stream->n_samples = n_samples;
- stream->samples = samples;
-
- if (!gst_byte_reader_skip (&stsc, 1 + 3) ||
- !gst_byte_reader_get_uint32_be (&stsc, &n_samples_per_chunk))
+ /* make sure there's enough data */
+ if (!qt_atom_parser_has_chunks (&stream->stts, stream->n_sample_times,
+ 2 * 4))
goto corrupt_file;
- GST_DEBUG_OBJECT (qtdemux, "n_samples_per_chunk %u", n_samples_per_chunk);
- sample_index = 0;
timestamp = 0;
+ stream->min_duration = 0;
+ time = 0;
+ index = 0;
+ for (i = 0; i < stream->n_sample_times; i++) {
+ guint32 n;
+ guint32 duration;
- if (!qt_atom_parser_has_chunks (&stsc, n_samples_per_chunk, 12))
- goto corrupt_file;
+ n = gst_byte_reader_get_uint32_be_unchecked (&stream->stts);
+ duration = gst_byte_reader_get_uint32_be_unchecked (&stream->stts);
+ GST_LOG_OBJECT (qtdemux, "block %d, %u timestamps, duration %u ", i, n,
+ duration);
- for (i = 0; i < n_samples_per_chunk; i++) {
- GstByteReader co_chunk;
- guint32 first_chunk, last_chunk;
- guint32 samples_per_chunk;
+ /* take first duration for fps */
+ if (G_UNLIKELY (stream->min_duration == 0))
+ stream->min_duration = duration;
- first_chunk = gst_byte_reader_get_uint32_be_unchecked (&stsc);
- samples_per_chunk = gst_byte_reader_get_uint32_be_unchecked (&stsc);
- gst_byte_reader_skip_unchecked (&stsc, 4);
+ for (j = 0; j < n; j++) {
+ GST_DEBUG_OBJECT (qtdemux,
+ "sample %d: index %d, timestamp %" GST_TIME_FORMAT, index, j,
+ GST_TIME_ARGS (timestamp));
- /* chunk numbers are counted from 1 it seems */
- if (G_UNLIKELY (first_chunk == 0))
- goto corrupt_file;
- else
- --first_chunk;
+ stream->samples[index].timestamp = timestamp;
+ /* add non-scaled values to avoid rounding errors */
+ time += duration;
+ timestamp = gst_util_uint64_scale (time, GST_SECOND, stream->timescale);
+ stream->samples[index].duration =
+ timestamp - stream->samples[index].timestamp;
- /* the last chunk of each entry is calculated by taking the first chunk
- * of the next entry; except if there is no next, where we fake it with
- * INT_MAX */
- if (G_UNLIKELY (i == (n_samples_per_chunk - 1))) {
- last_chunk = G_MAXUINT32;
- } else {
- last_chunk = gst_byte_reader_peek_uint32_be_unchecked (&stsc);
- if (G_UNLIKELY (last_chunk == 0))
- goto corrupt_file;
- else
- --last_chunk;
+ index++;
+ if (G_UNLIKELY (index >= stream->n_samples))
+ goto done3;
}
+ }
+ /* fill up empty timestamps with the last timestamp, this can happen when
+ * the last samples do not decode and so we don't have timestamps for them.
+ * We however look at the last timestamp to estimate the track length so we
+ * need something in here. */
+ for (; index < stream->n_samples; index++) {
+ GST_DEBUG_OBJECT (qtdemux,
+ "fill sample %d: timestamp %" GST_TIME_FORMAT, index,
+ GST_TIME_ARGS (timestamp));
+ stream->samples[index].timestamp = timestamp;
+ stream->samples[index].duration = -1;
+ }
+ }
+done3:
+ {
+ /* sample sync, can be NULL */
+ if (qtdemux_tree_get_child_by_type_full (stbl, FOURCC_stss, &stream->stss)) {
- GST_LOG_OBJECT (qtdemux,
- "entry %d has first_chunk %d, last_chunk %d, samples_per_chunk %d", i,
- first_chunk, last_chunk, samples_per_chunk);
-
- if (G_UNLIKELY (last_chunk < first_chunk))
+ /* mark keyframes */
+ if (!gst_byte_reader_skip (&stream->stss, 4))
+ goto corrupt_file;
+ if (!gst_byte_reader_get_uint32_be (&stream->stss,
+ &stream->n_sample_syncs))
goto corrupt_file;
- if (last_chunk != G_MAXUINT32) {
- if (!qt_atom_parser_peek_sub (&co_reader, first_chunk * co_size,
- (last_chunk - first_chunk) * co_size, &co_chunk))
- goto corrupt_file;
+ if (!stream->n_sample_syncs) {
+ stream->all_keyframe = TRUE;
} else {
- co_chunk = co_reader;
- if (!gst_byte_reader_skip (&co_chunk, first_chunk * co_size))
+ /* make sure there's enough data */
+ if (!qt_atom_parser_has_chunks (&stream->stss, stream->n_sample_syncs,
+ 4))
goto corrupt_file;
+ for (i = 0; i < stream->n_sample_syncs; i++) {
+ /* note that the first sample is index 1, not 0 */
+ index = gst_byte_reader_get_uint32_be_unchecked (&stream->stss);
+ if (G_LIKELY (index > 0 && index <= stream->n_samples))
+ stream->samples[index - 1].keyframe = TRUE;
+ }
}
+ /* stps marks partial sync frames like open GOP I-Frames */
+ if (qtdemux_tree_get_child_by_type_full (stbl, FOURCC_stps,
+ &stream->stps)) {
- for (j = first_chunk; j < last_chunk; j++) {
- if (j >= n_samples)
- goto done;
-
- samples[j].offset =
- qt_atom_parser_get_offset_unchecked (&co_chunk, co_size);
-
- GST_LOG_OBJECT (qtdemux, "Created entry %d with offset "
- "%" G_GUINT64_FORMAT, j, samples[j].offset);
+ if (!gst_byte_reader_skip (&stream->stps, 4))
+ goto corrupt_file;
+ if (!gst_byte_reader_get_uint32_be (&stream->stps,
+ &stream->n_sample_partial_syncs))
+ goto corrupt_file;
- if (stream->samples_per_frame * stream->bytes_per_frame) {
- samples[j].size = (samples_per_chunk * stream->n_channels) /
- stream->samples_per_frame * stream->bytes_per_frame;
- } else {
- samples[j].size = samples_per_chunk;
+ /* if there are no entries, the stss table contains the real
+ * sync samples */
+ if (stream->n_sample_partial_syncs) {
+ /* make sure there's enough data */
+ if (!qt_atom_parser_has_chunks (&stream->stps,
+ stream->n_sample_partial_syncs, 4))
+ goto corrupt_file;
+ for (i = 0; i < stream->n_sample_partial_syncs; i++) {
+ /* note that the first sample is index 1, not 0 */
+ index = gst_byte_reader_get_uint32_be_unchecked (&stream->stps);
+ if (G_LIKELY (index > 0 && index <= stream->n_samples))
+ stream->samples[index - 1].keyframe = TRUE;
+ }
}
-
- GST_DEBUG_OBJECT (qtdemux, "sample %d: timestamp %" GST_TIME_FORMAT
- ", size %u", j, GST_TIME_ARGS (timestamp), samples[j].size);
-
- samples[j].timestamp = timestamp;
- sample_index += samples_per_chunk;
-
- timestamp = gst_util_uint64_scale (sample_index,
- GST_SECOND, stream->timescale);
- samples[j].duration = timestamp - samples[j].timestamp;
-
- samples[j].keyframe = TRUE;
}
+ } else {
+ /* no stss, all samples are keyframes */
+ stream->all_keyframe = TRUE;
}
}
+ctts:
/* composition time to sample */
- if (qtdemux_tree_get_child_by_type_full (stbl, FOURCC_ctts, &ctts)) {
+ if (qtdemux_tree_get_child_by_type_full (stbl, FOURCC_ctts, &stream->ctts)) {
guint32 n_entries;
guint32 count;
gint32 soffset;
- gst_byte_reader_skip (&ctts, 1 + 3);
- n_entries = gst_byte_reader_get_uint32_be_unchecked (&ctts);
+ gst_byte_reader_skip (&stream->ctts, 1 + 3);
+ n_entries = gst_byte_reader_get_uint32_be_unchecked (&stream->ctts);
- if (!qt_atom_parser_has_chunks (&ctts, n_entries, 4 + 4))
+ if (!qt_atom_parser_has_chunks (&stream->ctts, n_entries, 4 + 4))
goto corrupt_file;
/* Fill in the pts_offsets */
index = 0;
for (i = 0; i < n_entries; i++) {
- count = gst_byte_reader_get_uint32_be_unchecked (&ctts);
- soffset = gst_byte_reader_get_uint32_be_unchecked (&ctts);
+ count = gst_byte_reader_get_uint32_be_unchecked (&stream->ctts);
+ soffset = gst_byte_reader_get_uint32_be_unchecked (&stream->ctts);
for (j = 0; j < count; j++) {
/* we operate with very small soffset values here, it shouldn't overflow */
- samples[index].pts_offset = soffset * GST_SECOND / stream->timescale;
+ stream->samples[index].pts_offset =
+ soffset * GST_SECOND / stream->timescale;
index++;
- if (G_UNLIKELY (index >= n_samples))
+ if (G_UNLIKELY (index >= stream->n_samples))
goto done;
}
}
@@ -4142,13 +4104,14 @@ no_samples:
}
out_of_memory:
{
- GST_WARNING_OBJECT (qtdemux, "failed to allocate %d samples", n_samples);
+ GST_WARNING_OBJECT (qtdemux, "failed to allocate %d samples",
+ stream->n_samples);
return FALSE;
}
index_too_big:
{
GST_WARNING_OBJECT (qtdemux, "not allocating index of %d samples, would "
- "be larger than %uMB (broken file?)", n_samples,
+ "be larger than %uMB (broken file?)", stream->n_samples,
QTDEMUX_MAX_SAMPLE_INDEX_SIZE >> 20);
return FALSE;
}