movenc: Add an API unit test for fragmenting options/calls

Contrary to the normal fate tests that run via avconv, this tests
nontrivial call sequences that are only doable via the API
(mainly for different corner cases when using the muxer for
segmenting).

The test muxes fake packet data (with extradata that looks
enough like proper data to make the file be viewable with e.g.
boxdumper) and checks the hash of the produced files. The test also
verifies that fragments produced via different call sequences remain
identical (to avoid e.g. updating the output hashes and suddenly
having fragments that used to be identical suddenly diverging), for
fragments written with frag_discont and/or delay_moov.

Signed-off-by: Martin Storsjö <martin@martin.st>

1 files changed, 588 insertions, 0 deletions

diff --git a/libavformat/movenc-test.c b/libavformat/movenc-test.c
new file mode 100644
index 0000000000..20043b295f
--- /dev/null
+++ b/libavformat/movenc-test.c
@@ -0,0 +1,588 @@
+/*
+ * Copyright (c) 2015 Martin Storsjo
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "config.h"
+
+#include "libavutil/intreadwrite.h"
+#include "libavutil/mathematics.h"
+#include "libavutil/md5.h"
+
+#include "avformat.h"
+
+#if HAVE_UNISTD_H
+#include <unistd.h>
+#endif
+
+#if !HAVE_GETOPT
+#include "compat/getopt.c"
+#endif
+
+#define HASH_SIZE 16
+
+static const uint8_t h264_extradata[] = {
+    0x01, 0x4d, 0x40, 0x1e, 0xff, 0xe1, 0x00, 0x02, 0x67, 0x4d, 0x01, 0x00, 0x02, 0x68, 0xef
+};
+static const uint8_t aac_extradata[] = {
+    0x12, 0x10
+};
+
+
+const char *format = "mp4";
+AVFormatContext *ctx;
+uint8_t iobuf[32768];
+AVDictionary *opts;
+
+int write_file;
+const char *cur_name;
+FILE* out;
+int out_size;
+struct AVMD5* md5;
+uint8_t hash[HASH_SIZE];
+
+AVStream *video_st, *audio_st;
+int64_t audio_dts, video_dts;
+
+int bframes;
+int duration;
+int audio_duration;
+int frames;
+int gop_size;
+int64_t next_p_pts;
+enum AVPictureType last_picture;
+int skip_write;
+int skip_write_audio;
+int clear_duration;
+
+int num_warnings;
+
+int check_faults;
+
+
+static void count_warnings(void *avcl, int level, const char *fmt, va_list vl)
+{
+    if (level == AV_LOG_WARNING)
+        num_warnings++;
+}
+
+static void init_count_warnings(void)
+{
+    av_log_set_callback(count_warnings);
+    num_warnings = 0;
+}
+
+static void reset_count_warnings(void)
+{
+    av_log_set_callback(av_log_default_callback);
+}
+
+static int io_write(void *opaque, uint8_t *buf, int size)
+{
+    out_size += size;
+    av_md5_update(md5, buf, size);
+    if (out)
+        fwrite(buf, 1, size, out);
+    return size;
+}
+
+static void init_out(const char *name)
+{
+    char buf[100];
+    cur_name = name;
+    snprintf(buf, sizeof(buf), "%s.%s", cur_name, format);
+
+    av_md5_init(md5);
+    if (write_file) {
+        out = fopen(buf, "wb");
+        if (!out)
+            perror(buf);
+    }
+    out_size = 0;
+}
+
+static void close_out(void)
+{
+    int i;
+    av_md5_final(md5, hash);
+    for (i = 0; i < HASH_SIZE; i++)
+        printf("%02x", hash[i]);
+    printf(" %d %s\n", out_size, cur_name);
+    if (out)
+        fclose(out);
+    out = NULL;
+}
+
+static void check_func(int value, int line, const char *msg, ...)
+{
+    if (!value) {
+        va_list ap;
+        va_start(ap, msg);
+        printf("%d: ", line);
+        vprintf(msg, ap);
+        printf("\n");
+        check_faults++;
+    }
+}
+#define check(value, ...) check_func(value, __LINE__, __VA_ARGS__)
+
+static void init_fps(int bf, int audio_preroll, int fps)
+{
+    AVStream *st;
+    ctx = avformat_alloc_context();
+    if (!ctx)
+        exit(1);
+    ctx->oformat = av_guess_format(format, NULL, NULL);
+    if (!ctx->oformat)
+        exit(1);
+    ctx->pb = avio_alloc_context(iobuf, sizeof(iobuf), AVIO_FLAG_WRITE, NULL, NULL, io_write, NULL);
+    if (!ctx->pb)
+        exit(1);
+    ctx->flags |= AVFMT_FLAG_BITEXACT;
+
+    st = avformat_new_stream(ctx, NULL);
+    if (!st)
+        exit(1);
+    st->codec->codec_type = AVMEDIA_TYPE_VIDEO;
+    st->codec->codec_id = AV_CODEC_ID_H264;
+    st->codec->width = 640;
+    st->codec->height = 480;
+    st->time_base.num = 1;
+    st->time_base.den = 30;
+    st->codec->extradata_size = sizeof(h264_extradata);
+    st->codec->extradata = av_mallocz(st->codec->extradata_size + AV_INPUT_BUFFER_PADDING_SIZE);
+    if (!st->codec->extradata)
+        exit(1);
+    memcpy(st->codec->extradata, h264_extradata, sizeof(h264_extradata));
+    st->codec->flags |= AV_CODEC_FLAG_GLOBAL_HEADER;
+    video_st = st;
+
+    st = avformat_new_stream(ctx, NULL);
+    if (!st)
+        exit(1);
+    st->codec->codec_type = AVMEDIA_TYPE_AUDIO;
+    st->codec->codec_id = AV_CODEC_ID_AAC;
+    st->codec->sample_rate = 44100;
+    st->codec->channels = 2;
+    st->time_base.num = 1;
+    st->time_base.den = 44100;
+    st->codec->extradata_size = sizeof(aac_extradata);
+    st->codec->extradata = av_mallocz(st->codec->extradata_size + AV_INPUT_BUFFER_PADDING_SIZE);
+    if (!st->codec->extradata)
+        exit(1);
+    memcpy(st->codec->extradata, aac_extradata, sizeof(aac_extradata));
+    st->codec->flags |= AV_CODEC_FLAG_GLOBAL_HEADER;
+    audio_st = st;
+
+    if (avformat_write_header(ctx, &opts) < 0)
+        exit(1);
+    av_dict_free(&opts);
+
+    frames = 0;
+    gop_size = 30;
+    duration = video_st->time_base.den / fps;
+    audio_duration = 1024 * audio_st->time_base.den / audio_st->codec->sample_rate;
+    if (audio_preroll)
+        audio_preroll = 2048 * audio_st->time_base.den / audio_st->codec->sample_rate;
+
+    bframes = bf;
+    video_dts = bframes ? -duration : 0;
+    audio_dts = -audio_preroll;
+}
+
+static void init(int bf, int audio_preroll)
+{
+    init_fps(bf, audio_preroll, 30);
+}
+
+static void mux_frames(int n)
+{
+    int end_frames = frames + n;
+    while (1) {
+        AVPacket pkt;
+        uint8_t pktdata[4];
+        av_init_packet(&pkt);
+
+        if (av_compare_ts(audio_dts, audio_st->time_base, video_dts, video_st->time_base) < 0) {
+            pkt.dts = pkt.pts = audio_dts;
+            pkt.stream_index = 1;
+            pkt.duration = audio_duration;
+            audio_dts += audio_duration;
+        } else {
+            if (frames == end_frames)
+                break;
+            pkt.dts = video_dts;
+            pkt.stream_index = 0;
+            pkt.duration = duration;
+            if ((frames % gop_size) == 0) {
+                pkt.flags |= AV_PKT_FLAG_KEY;
+                last_picture = AV_PICTURE_TYPE_I;
+                pkt.pts = pkt.dts + duration;
+                video_dts = pkt.pts;
+            } else {
+                if (last_picture == AV_PICTURE_TYPE_P) {
+                    last_picture = AV_PICTURE_TYPE_B;
+                    pkt.pts = pkt.dts;
+                    video_dts = next_p_pts;
+                } else {
+                    last_picture = AV_PICTURE_TYPE_P;
+                    if (((frames + 1) % gop_size) == 0) {
+                        pkt.pts = pkt.dts + duration;
+                        video_dts = pkt.pts;
+                    } else {
+                        next_p_pts = pkt.pts = pkt.dts + 2 * duration;
+                        video_dts += duration;
+                    }
+                }
+            }
+            if (!bframes)
+                pkt.pts = pkt.dts;
+            frames++;
+        }
+
+        if (clear_duration)
+            pkt.duration = 0;
+        AV_WB32(pktdata, pkt.pts);
+        pkt.data = pktdata;
+        pkt.size = 4;
+        if (skip_write)
+            continue;
+        if (skip_write_audio && pkt.stream_index == 1)
+            continue;
+        av_write_frame(ctx, &pkt);
+    }
+}
+
+static void mux_gops(int n)
+{
+    mux_frames(gop_size * n);
+}
+
+static void skip_gops(int n)
+{
+    skip_write = 1;
+    mux_gops(n);
+    skip_write = 0;
+}
+
+static void finish(void)
+{
+    av_write_trailer(ctx);
+    av_free(ctx->pb);
+    avformat_free_context(ctx);
+    ctx = NULL;
+}
+
+static void help(void)
+{
+    printf("movenc-test [-w]\n"
+           "-w          write output into files\n");
+}
+
+int main(int argc, char **argv)
+{
+    int c;
+    uint8_t header[HASH_SIZE];
+    uint8_t content[HASH_SIZE];
+    int empty_moov_pos;
+    int prev_pos;
+
+    for (;;) {
+        c = getopt(argc, argv, "wh");
+        if (c == -1)
+            break;
+        switch (c) {
+        case 'w':
+            write_file = 1;
+            break;
+        default:
+        case 'h':
+            help();
+            return 0;
+        }
+    }
+
+    av_register_all();
+
+    md5 = av_md5_alloc();
+    if (!md5)
+        return 1;
+
+    // Write a fragmented file with an initial moov that actually contains some
+    // samples. One moov+mdat with 1 second of data and one moof+mdat with 1
+    // second of data.
+    init_out("non-empty-moov");
+    av_dict_set(&opts, "movflags", "frag_keyframe", 0);
+    init(0, 0);
+    mux_gops(2);
+    finish();
+    close_out();
+
+    // Write a similar file, but with b-frames and audio preroll, handled
+    // via an edit list.
+    init_out("non-empty-moov-elst");
+    av_dict_set(&opts, "movflags", "frag_keyframe", 0);
+    av_dict_set(&opts, "use_editlist", "1", 0);
+    init(1, 1);
+    mux_gops(2);
+    finish();
+    close_out();
+
+    // Use b-frames but no audio-preroll, but without an edit list.
+    // Due to avoid_negative_ts == AVFMT_AVOID_NEG_TS_MAKE_ZERO, the dts
+    // of the first audio packet is > 0, but it is set to zero since edit
+    // lists aren't used, increasing the duration of the first packet instead.
+    init_out("non-empty-moov-no-elst");
+    av_dict_set(&opts, "movflags", "frag_keyframe", 0);
+    av_dict_set(&opts, "use_editlist", "0", 0);
+    init(1, 0);
+    mux_gops(2);
+    finish();
+    close_out();
+
+    format = "ismv";
+    // Write an ISMV, with b-frames and audio preroll.
+    init_out("ismv");
+    av_dict_set(&opts, "movflags", "frag_keyframe", 0);
+    init(1, 1);
+    mux_gops(2);
+    finish();
+    close_out();
+    format = "mp4";
+
+    // An initial moov that doesn't contain any samples, followed by two
+    // moof+mdat pairs.
+    init_out("empty-moov");
+    av_dict_set(&opts, "movflags", "frag_keyframe+empty_moov", 0);
+    init(0, 0);
+    mux_gops(2);
+    finish();
+    close_out();
+    memcpy(content, hash, HASH_SIZE);
+
+    // Similar to the previous one, but with input that doesn't start at
+    // pts/dts 0. avoid_negative_ts behaves in the same way as
+    // in non-empty-moov-no-elst above.
+    init_out("empty-moov-no-elst");
+    av_dict_set(&opts, "movflags", "frag_keyframe+empty_moov", 0);
+    init(1, 0);
+    mux_gops(2);
+    finish();
+    close_out();
+
+    // Same as the previous one, but disable avoid_negative_ts (which
+    // would require using an edit list, but with empty_moov, one can't
+    // write a sensible edit list, when the start timestamps aren't known).
+    // This should trigger a warning - we check that the warning is produced.
+    init_count_warnings();
+    init_out("empty-moov-no-elst-no-adjust");
+    av_dict_set(&opts, "movflags", "frag_keyframe+empty_moov", 0);
+    av_dict_set(&opts, "avoid_negative_ts", "0", 0);
+    init(1, 0);
+    mux_gops(2);
+    finish();
+    close_out();
+
+    reset_count_warnings();
+    check(num_warnings > 0, "No warnings printed for unhandled start offset");
+
+    // Verify that delay_moov produces the same as empty_moov for
+    // simple input
+    init_out("delay-moov");
+    av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov", 0);
+    init(0, 0);
+    mux_gops(2);
+    finish();
+    close_out();
+    check(!memcmp(hash, content, HASH_SIZE), "delay_moov differs from empty_moov");
+
+    // Test writing content that requires an edit list using delay_moov
+    init_out("delay-moov-elst");
+    av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov", 0);
+    init(1, 1);
+    mux_gops(2);
+    finish();
+    close_out();
+
+    // Test writing a file with one track lacking packets, with delay_moov.
+    skip_write_audio = 1;
+    init_out("delay-moov-empty-track");
+    av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov", 0);
+    init(0, 0);
+    mux_gops(2);
+    // The automatic flushing shouldn't output anything, since we're still
+    // waiting for data for some tracks
+    check(out_size == 0, "delay_moov flushed prematurely");
+    // When closed (or manually flushed), all the written data should still
+    // be output.
+    finish();
+    close_out();
+    check(out_size > 0, "delay_moov didn't output anything");
+
+    // Check that manually flushing still outputs things as expected. This
+    // produces two fragments, while the one above produces only one.
+    init_out("delay-moov-empty-track-flush");
+    av_dict_set(&opts, "movflags", "frag_custom+delay_moov", 0);
+    init(0, 0);
+    mux_gops(1);
+    av_write_frame(ctx, NULL); // Force writing the moov
+    check(out_size > 0, "No moov written");
+    av_write_frame(ctx, NULL);
+    mux_gops(1);
+    av_write_frame(ctx, NULL);
+    finish();
+    close_out();
+
+    skip_write_audio = 0;
+
+
+
+    // Verify that the header written by delay_moov when manually flushed
+    // is identical to the one by empty_moov.
+    init_out("empty-moov-header");
+    av_dict_set(&opts, "movflags", "frag_keyframe+empty_moov", 0);
+    init(0, 0);
+    close_out();
+    memcpy(header, hash, HASH_SIZE);
+    init_out("empty-moov-content");
+    mux_gops(2);
+    // Written 2 seconds of content, with an automatic flush after 1 second.
+    check(out_size > 0, "No automatic flush?");
+    empty_moov_pos = prev_pos = out_size;
+    // Manually flush the second fragment
+    av_write_frame(ctx, NULL);
+    check(out_size > prev_pos, "No second fragment flushed?");
+    prev_pos = out_size;
+    // Check that an extra flush doesn't output any more data
+    av_write_frame(ctx, NULL);
+    check(out_size == prev_pos, "More data written?");
+    close_out();
+    memcpy(content, hash, HASH_SIZE);
+    // Ignore the trailer written here
+    finish();
+
+    init_out("delay-moov-header");
+    av_dict_set(&opts, "movflags", "frag_custom+delay_moov", 0);
+    init(0, 0);
+    check(out_size == 0, "Output written during init with delay_moov");
+    mux_gops(1); // Write 1 second of content
+    av_write_frame(ctx, NULL); // Force writing the moov
+    close_out();
+    check(!memcmp(hash, header, HASH_SIZE), "delay_moov header differs from empty_moov");
+    init_out("delay-moov-content");
+    av_write_frame(ctx, NULL); // Flush the first fragment
+    check(out_size == empty_moov_pos, "Manually flushed content differs from automatically flushed, %d vs %d", out_size, empty_moov_pos);
+    mux_gops(1); // Write the rest of the content
+    av_write_frame(ctx, NULL); // Flush the second fragment
+    close_out();
+    check(!memcmp(hash, content, HASH_SIZE), "delay_moov content differs from empty_moov");
+    finish();
+
+
+    // Verify that we can produce an identical second fragment without
+    // writing the first one. First write the reference fragments that
+    // we want to reproduce.
+    av_dict_set(&opts, "movflags", "frag_custom+empty_moov+dash", 0);
+    init(0, 0);
+    mux_gops(1);
+    av_write_frame(ctx, NULL); // Output the first fragment
+    init_out("empty-moov-second-frag");
+    mux_gops(1);
+    av_write_frame(ctx, NULL); // Output the second fragment
+    close_out();
+    memcpy(content, hash, HASH_SIZE);
+    finish();
+
+    // Produce the same second fragment without actually writing the first
+    // one before.
+    av_dict_set(&opts, "movflags", "frag_custom+empty_moov+dash+frag_discont", 0);
+    av_dict_set(&opts, "fragment_index", "2", 0);
+    av_dict_set(&opts, "avoid_negative_ts", "0", 0);
+    av_dict_set(&opts, "use_editlist", "0", 0);
+    init(0, 0);
+    skip_gops(1);
+    init_out("empty-moov-second-frag-discont");
+    mux_gops(1);
+    av_write_frame(ctx, NULL); // Output the second fragment
+    close_out();
+    check(!memcmp(hash, content, HASH_SIZE), "discontinuously written fragment differs");
+    finish();
+
+    // Produce the same thing by using delay_moov, which requires a slightly
+    // different call sequence.
+    av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash+frag_discont", 0);
+    av_dict_set(&opts, "fragment_index", "2", 0);
+    init(0, 0);
+    skip_gops(1);
+    mux_gops(1);
+    av_write_frame(ctx, NULL); // Output the moov
+    init_out("delay-moov-second-frag-discont");
+    av_write_frame(ctx, NULL); // Output the second fragment
+    close_out();
+    check(!memcmp(hash, content, HASH_SIZE), "discontinuously written fragment differs");
+    finish();
+
+
+    // Test VFR content, with sidx atoms (which declare the pts duration
+    // of a fragment, forcing overriding the start pts of the next one).
+    // Here, the fragment duration in pts is significantly different from
+    // the duration in dts. The video stream starts at dts=-10,pts=0, and
+    // the second fragment starts at dts=155,pts=156. The trun duration sum
+    // of the first fragment is 165, which also is written as
+    // baseMediaDecodeTime in the tfdt in the second fragment. The sidx for
+    // the first fragment says earliest_presentation_time = 0 and
+    // subsegment_duration = 156, which also matches the sidx in the second
+    // fragment. For the audio stream, the pts and dts durations also don't
+    // match - the input stream starts at pts=-2048, but that part is excluded
+    // by the edit list.
+    init_out("vfr");
+    av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov+dash", 0);
+    init_fps(1, 1, 3);
+    mux_frames(gop_size/2);
+    duration /= 10;
+    mux_frames(gop_size/2);
+    mux_gops(1);
+    finish();
+    close_out();
+
+    // Test VFR content, with cleared duration fields. In these cases,
+    // the muxer must guess the duration of the last packet of each
+    // fragment. As long as the framerate doesn't vary (too much) at the
+    // fragment edge, it works just fine. Additionally, when automatically
+    // cutting fragments, the muxer already know the timestamps of the next
+    // packet for one stream (in most cases the video stream), avoiding
+    // having to use guesses for that one.
+    init_count_warnings();
+    clear_duration = 1;
+    init_out("vfr-noduration");
+    av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov+dash", 0);
+    init_fps(1, 1, 3);
+    mux_frames(gop_size/2);
+    duration /= 10;
+    mux_frames(gop_size/2);
+    mux_gops(1);
+    finish();
+    close_out();
+    clear_duration = 0;
+    reset_count_warnings();
+    check(num_warnings > 0, "No warnings printed for filled in durations");
+
+    av_free(md5);
+
+    return check_faults > 0 ? 1 : 0;
+}