This should be more correct. This also should give more sensible
switching between video streams with different amount of b-frame
delay.
The current dash.js release (1.2.0) fails to start playback of
such files from the start (if the start pts is > 0), but this has
been fixed in the current git version of dash.js.
Also enable the use of edit lists, so that streams in many cases
start at pts=0.
Signed-off-by: Martin Storsjö <martin@martin.st>
Use the more generic approach with the delay_moov flag, instead of
having a update mechanism specific to this one single atom.
Signed-off-by: Martin Storsjö <martin@martin.st>
This delays writing the moov until the first fragment is written,
or can be flushed by the caller explicitly when wanted. If the first
sample in all streams is available at this point, we can write
a proper editlist at this point, allowing streams to start at
something else than dts=0. For AC3 and DNXHD, a packet is
needed in order to write the moov header properly.
This isn't added to the normal behaviour for empty_moov, since
the behaviour that ftyp+moov is written during avformat_write_header
would be changed. Callers that split the output stream into header+segments
(either by flushing manually, with the custom_frag flag set, or by
just differentiating between data written during avformat_write_header
and the rest) will need to be adjusted to take this option into use.
For handling streams that start at something else than dts=0, an
alternative would be to use different kinds of heuristics for
guessing the start dts (using AVCodecContext delay or has_b_frames
together with the frame rate), but this is not reliable and doesn't
necessarily work well with stream copy, and wouldn't work for getting
the right initialization data for AC3 or DNXHD either.
Signed-off-by: Martin Storsjö <martin@martin.st>
If fragments == 0 it means we haven't written any moov atom yet.
If the empty_moov flag is set, we already have written an empty moov
atom at startup. Thus, the check for empty_moov is redundant.
This is in preparation for allowing writing the moov atom later,
even when using the empty moov flag.
Signed-off-by: Martin Storsjö <martin@martin.st>
When writing an explicit time, reset the cur_time variable to this
value as well. This avoids writing excessive time attributes for each
segment in the timeline, as long as the segments are continuous.
Signed-off-by: Martin Storsjö <martin@martin.st>
The MoveFileExA is available in the headers regardless which API
subset is targeted, but it is missing in the Windows Phone link
libraries. When targeting Windows Store apps, the function is
available both in the headers and in the link libraries, and thus
there is no indication for the build system that this function
should be avoided - such an indication is only given by the
Windows App Certification Kit, which forbids using the MoveFileExA
function.
Therefore check the WINAPI_FAMILY defines instead, to figure out
which API subset is targeted.
Signed-off-by: Martin Storsjö <martin@martin.st>
Since the mpegts muxer now can handle being called with a NULL
AVIOContext, we don't need to try to allocate one before calling
write_trailer.
Signed-off-by: Martin Storsjö <martin@martin.st>
If opening and closing dynamic buffers as AVIOContext, we may
not have any AVIOContext available when wanting to close and
deallocate the muxer. Allow calling write_trailer despite this.
Signed-off-by: Martin Storsjö <martin@martin.st>
The pts and the corresponding duration is written in sidx
atoms, thus make sure these match up correctly.
Signed-off-by: Martin Storsjö <martin@martin.st>
Since this structurally is quite different from normal RTP
(multiple streams are muxed into one single mpegts stream,
which is packetized into one single RTP session), it is kept
as a separate muxer.
Since this structurally also behaves differently than normal
RTP, all of the other muxers that do chained RTP muxing
(rtsp, sap, mp4) would need to be updated similarly to handle
this - in particular, creating one single rtp_mpegts muxer
for the whole presentation instead of one rtp muxer per stream.
Signed-off-by: Martin Storsjö <martin@martin.st>
Instead explicitly jump to the default case in the cases where
it is wanted, and avoid fallthrough between different codecs,
which could easily introduce bugs if people editing the code
aren't careful.
Signed-off-by: Martin Storsjö <martin@martin.st>
This avoids trying to do sliced encoding, even if a slice/packet
size is requested (via the -ps option or the rtp_payload_size
field), since the encoder currently doesn't support it (or at least
our decoder can't decode it, even if the h261_encode_gob_header
function is hooked up to be called from the slicing part in
mpegvideo_enc.c).
Signed-off-by: Martin Storsjö <martin@martin.st>
If we throw away the buffered incomplete frame, make sure to also
throw away the buffered bits of an incomplete byte at the same
time.
CC: libav-stable@libav.org
Signed-off-by: Martin Storsjö <martin@martin.st>
In particular, when packetizing mpegts into rtp, the input packet
timestamp may come from more than one stream, which could cause
multiple packets be written with the same timestamp.
Signed-off-by: Martin Storsjö <martin@martin.st>
This is the same adjustment that the mp4 muxer does to the start
timestamp of fragments, since the timestamp of a sample in an mp4
file is implicit from the sum of earlier sample durations.
This avoids gaps in the timeline (which can stop dash.js from
playing it back), and makes sure the timestamp on the segmenter
level matches what the mp4 muxer actually writes into the segments.
This is only an issue if the AVPacket duration of the last
packet of a segment doesn't point to the actual start timestamp
of the next packet (the first in the next segment).
Signed-off-by: Martin Storsjö <martin@martin.st>
Write a new start time if the duration of the previous segment
didn't match the start of the next one. Check that segments
actually are continuous before writing a repeat count.
This makes sure timestamps deduced from the timeline actually
match the real start timestamp as written in filenames (if
using a template containing $Time$).
Signed-off-by: Martin Storsjö <martin@martin.st>
Since 3cec81f4d4, a zero-length metadata value would try to
allocate 2*0 bytes, where av_malloc() returns NULL.
Always add one to the allocated length, to allow space for
a null terminator in the zero-length case.
Incidentally, this fixes fate-alac on RVCT 4.0, where a compiler
bug seems to mess up the mov muxer to the point that it writes
the wrong sort of metadata. Previously this bug was undetected,
but since 3cec81f4d4 such mov files started returning
AVERROR(ENOMEM) in the mov demuxer.
Signed-off-by: Martin Storsjö <martin@martin.st>
For the last_duration field, it's mostly theoretical, but the
total_duration field more probably may need to actually be 64 bit.
Bug-Id: CID 1254944
Signed-off-by: Martin Storsjö <martin@martin.st>
This avoids potential out of bounds writes, with potential future
versions of the library.
Bug-Id: CID 1254945
CC: libav-stable@libav.org
Signed-off-by: Martin Storsjö <martin@martin.st>
Only the upper 2 bits of the first byte are known to be
a fixed value.
The lower bits in the first byte of a RTP packet could be set
if the input is from another RTP packetizers than libavformat's,
but for RTCP packets, they would also be set when sending RTCP RR
packets, triggering false warnings about incorrect input format
to the protocol.
Signed-off-by: Martin Storsjö <martin@martin.st>
The RTP muxer enables the actual codepaths within sdp.c,
which depend on hevc.o since e5cfc8fd.
This fixes builds with --disable-everything --enable-muxer=rtp.
Signed-off-by: Martin Storsjö <martin@martin.st>
Don't include the function pointer table in the code segment
in arm mode.
This shouldn't have any significant performance effect. It does
end up as a few more instructions than before, for ARM, but
only at the entry to this function, not within the fft functions
themselves.
Signed-off-by: Martin Storsjö <martin@martin.st>
Tell the user that the RTP muxer needs to be used to packetize
the data - using the RTP protocol on its own isn't enough.
Signed-off-by: Martin Storsjö <martin@martin.st>
This fixes the build on compilers that interpreted the earlier
code as a variable length array (which we intentionally disallow).
Signed-off-by: Martin Storsjö <martin@martin.st>
Currently, when streaming to an RTMP server, any time a packet of type
RTMP_PT_NOTIFY is encountered, the packet is prepended with @setDataFrame
before it gets sent to the server. This is incorrect; only packets for
onMetaData and |RtmpSampleAccess should invoke @setDataFrame on the RTMP
server. Specifically, the current bug manifests itself when trying to
stream onTextData or onCuePoint invocations.
This fix addresses that problem and ensures that the @setDataFrame is
only prepended for onMetaData and |RtmpSampleAccess.
Since data is fed to the rtmp_write function in smaller pieces (depending
on the calling IO buffer size), we can't generally assume that the
whole packet (or even the whole command string) is available at once,
therefore we can only check the command string once the full packet
has been transferred to us for sending.
Based on a patch by Jeffrey Wescott.
Signed-off-by: Martin Storsjö <martin@martin.st>
We try to avoid mixing av_malloc with av_realloc, since av_malloc
may be implemented with functions that can't (formally) be mixed
with the functions used in av_realloc.
Signed-off-by: Martin Storsjö <martin@martin.st>
This reverts commit b9d08c77a4.
After taking MoveFileEx into use, we can replace files with renames
on windows as well.
Signed-off-by: Martin Storsjö <martin@martin.st>
This allows getting the normal unix semantics, where a rename
allows replacing an existing file.
Based on a suggestion by Reimar Döffinger.
Signed-off-by: Martin Storsjö <martin@martin.st>
This doesn't add any dependency on library internals, since this
only is a static inline function that gets built into each of the
calling functions - this is only to reduce the code duplication.
Signed-off-by: Martin Storsjö <martin@martin.st>
This allows setting the right fragment number if doing
random-access writing of fragments, and also allows reading the
current sequence number.
Signed-off-by: Martin Storsjö <martin@martin.st>
This allows creating a later mp4 fragment without sequentially
writing the earlier ones before (when called from a segmenter).
Normally when writing a fragmented mp4 file sequentially, the
first timestamps of a fragment are adjusted to match the
end of the previous fragment, to make sure the timestamp is the
same, even if it is calculated as the sum of previous fragment
durations. (And for the first packet in a file, the offset of
the first packet is written using an edit list.)
When writing an individual mp4 fragment discontinuously like this
(with potentially writing the earlier fragments separately later),
there's a risk of getting a gap in the timeline if the duration
field of the last packet in the previous fragment doesn't match up
with the start time of the next fragment.
Using this requires setting -avoid_negative_ts make_non_negative
(or -avoid_negative_ts 0).
Signed-off-by: Martin Storsjö <martin@martin.st>
This makes sure that the internal utf8 path names are handled
properly - the normal file handling functions assume path names
are in the native codepage, which isn't utf8.
This assumes that the tools outside of lavf don't use the mkdir
definition. (The tools don't do the same reading of command line
parameters as wchar either - they probably won't handle all possible
unicode file parameters properly, but at least work more predictably
if no utf8/wchar conversion is involved.)
This is moved further down in os_support.h, since windows.h shouldn't
be included before winsock2.h, while io.h needs to be included before
the manual defines for lseek functions.
Signed-off-by: Martin Storsjö <martin@martin.st>
On windows, rename(2) will fail if the target file exists. On
unix this trick is used to make sure that people reading the file
either will get the full previous file, or the full new version
of the file, but no intermediate version.
Signed-off-by: Martin Storsjö <martin@martin.st>
This makes sure that segments actually start at a keyframe (and
makes sure we don't split segments twice in a row, with one segment
consisting of only a handful of packets), when one stream uses b-frames
while another one doesn't.
Signed-off-by: Martin Storsjö <martin@martin.st>
Don't write any bitrate attribute if it isn't known. As long as one
doesn't want automatic bitrate switching, playback can work just
fine even if it isn't set.
If strict standard compliance is requested, this is still considered
an error, since the attribute is mandatory according to the spec.
Based on a patch by Rodger Combs.
Signed-off-by: Martin Storsjö <martin@martin.st>
This makes sure the default behaviour of using the internal encoder
stays the same regardless if libtwolame is enabled or not (as for
any external library).
This fixes fate-lavf-mpg if libav is built with libtwolame enabled.
CC: libav-stable@libav.org
Signed-off-by: Vittorio Giovara <vittorio.giovara@gmail.com>
The chained flv muxer wants one set of tags - normally this set
could be signaled via the AVOutputFormat codec_tag field (as
smoothstreamingenc and dashenc do). hdsenc doesn't signal it, since
the FLV codec tag arrays aren't exported from flvenc.c. This can
lead to the caller keeping an original codec tag from the originating
container here, which would then be a mismatch for the FLV muxer.
Since we don't really care about what codec tag the caller might
have set, just clear it and let the lavf muxer layer set the right
one for the chained FLV muxer later instead.
Signed-off-by: Martin Storsjö <martin@martin.st>
simd_align_16 is a configure item that can be enabled or disabled,
it's not a variable containing a list of other configure items
as need_memalign previously. This was broken in eba2233b5.
CC: libav-stable@libav.org
Signed-off-by: Martin Storsjö <martin@martin.st>
When given a stream starting at dts=0, it would previously consider
s->offset as uninitialized and set an offset when the second packet
was written, ending up writing two packets with dts=0. By initializing
this field to AV_NOPTS_VALUE, we make sure that we only initialize it
once, on the first packet.
Signed-off-by: Martin Storsjö <martin@martin.st>
This is mapped to the faststart flag (which in this case
perhaps should be called "shift and write index at the
start of the file"), which for fragmented files will
write a sidx index at the start.
When segmenting DASH into files, there's usually one sidx
at the start of each segment (although it's not clear to me
whether that actually is necessary). When storing all of it
in one file, the MPD doesn't necessarily need to describe
the individual segments, but the offsets of the fragments can be
fetched from one large sidx atom at the start of the file. This
allows creating files for the DASH ISO BMFF on-demand profile.
Signed-off-by: Martin Storsjö <martin@martin.st>
Previously only tfra entries were added for the first track in each moof.
The frag_info array used for tfra can also be used for writing
other kinds of fragment indexes, where it's more important to
include all tracks.
When the separate_moof option is enabled (as in ismv), we write
a separate moof for each track, so this doesn't make any difference
in that case.
Signed-off-by: Martin Storsjö <martin@martin.st>
This is mostly to serve as a reference example on how to segment
the output from the mp4 muxer, capable of writing the segment
list in four different ways:
- SegmentTemplate with SegmentTimeline
- SegmentTemplate with implicit segments
- SegmentList with individual files
- SegmentList with one single file per track, and byte ranges
The muxer is able to serve live content (with optional windowing)
or create a static segmented MPD.
In advanced cases, users will probably want to do the segmenting
in their own application code.
Signed-off-by: Martin Storsjö <martin@martin.st>
A flag "dash" is added, which enables the necessary flags for
creating DASH compatible fragments.
When this is enabled, one sidx atom is written for each track
before every moof atom.
Signed-off-by: Martin Storsjö <martin@martin.st>
By calling this after writing the moof the first time (for
calculating the moof size), we can avoid intermediate storage
of tfrf_offset in MOVTrack.
Signed-off-by: Martin Storsjö <martin@martin.st>
When writing fragmented streams with an empty initial moov,
we won't have any samples in any tracks when writing the
moov atom, thus trust that any tracks that are added actually
will be present.
Signed-off-by: Martin Storsjö <martin@martin.st>
This allows running the code on android, where 64 bit binaries with
text relocations aren't allowed to be loaded.
Signed-off-by: Martin Storsjö <martin@martin.st>
vorbis_parser.o is built unconditionally since 5e80fb7ff, and the
unconditionally built parts of it depend on xiph.o.
This fixes builds with --disable-everything.
Signed-off-by: Martin Storsjö <martin@martin.st>
The latest fdk-aac code drop (from android 5.0) changed the channel
layout enums (changing the value of existing enum constants), and
renamed the option for downmixing.
The failsafe comparison between ctype and FF_ARRAY_ELEMS(channel_counts)
can trigger warnings (-Wtautological-constant-out-of-range-compare)
when building with the old FDK AAC releases, where it can't be
out of range with the enum values used there.
CC: libav-stable@libav.org
Signed-off-by: Martin Storsjö <martin@martin.st>
This avoids a potential crash if writing a fragmented psp mp4
(which probably is only a hypothetical scenario).
Signed-off-by: Martin Storsjö <martin@martin.st>
Previously we wrote decoding timestamps here, while the specs
say it should be presentation timestamps.
Signed-off-by: Martin Storsjö <martin@martin.st>
When using the new first_trun flag instead of checking the track id,
we don't need to have a special case for the separate_moof flag
any longer.
This simplifies the complicated codepath ever so slightly.
Signed-off-by: Martin Storsjö <martin@martin.st>
In this case, shift tracks to start from zero instead (potentially
stretching the first sample in tracks that start later than the
first one).
Some software does not support edit lists at all, the adobe flash
player seems to be one of these. This results in AV sync errors when
edit lists are used to adjust AV sync.
Some players, such as QuickTime, don't respect the duration for
audio packets, so if an audio track starts later than the video
track and the first audio sample gets a duration longer than the
actual amount of data in it, the result will be out of sync.
Based on patches by Michael Niedermayer.
Signed-off-by: Martin Storsjö <martin@martin.st>
It is never an error if this method failed. If rt->live was
explicitly set to 0 (known to be a recorded file), print it
as a warning, otherwise print it as a debug message.
Based on a patch by Michael Niedermayer.
Signed-off-by: Martin Storsjö <martin@martin.st>
The tfdt atom shouldn't be needed in those cases, we already
write tfxd atoms for ismv anyway, which is roughly equivalent.
This avoids having to declare the iso6 brand for ismv files.
Signed-off-by: Martin Storsjö <martin@martin.st>
ISO/IEC 14496-12:2012/Cor 1:2013 is explicit about how this should be
handled. All zeros doesn't mean that the full file has got a zero
duration, only that the track samples described within the initial moov
have got zero duration. An all ones duration means an indeterminate
duration.
Keep writing a duration consisting of all ones for the ISM mode -
older windows media player versions won't play a file if this is
zero. (Newer windows media player versions play either version fine.)
Signed-off-by: Martin Storsjö <martin@martin.st>
Similarly to the omit_tfhd_offset flag added in e7bf085b, this
avoids writing absolute byte positions to the file, making them
more easily streamable.
This is a new feature from 14496-12:2012, so application support
isn't necessarily too widespread yet (support for it in libav was
added in 20f95f21f in July 2014).
Signed-off-by: Martin Storsjö <martin@martin.st>
The custom IO flag actually never is set for muxers, only for
demuxers, so the check was pointless (unless a user intentionally
would set the flag to signal using custom IO).
Signed-off-by: Martin Storsjö <martin@martin.st>
If one track doesn't have any samples within a moof, no traf/trun
is written for it. When the omit_tfhd_offset flag is set, none
of the tfhd atoms have any base_data_offset set, and the implicit
offset (end of previous track fragment data, or start of the moof
for the first trun) is used.
Signed-off-by: Martin Storsjö <martin@martin.st>
While a standalone implementation is nice, we already depend on
gmtime and gmtime_r in a number of places.
Signed-off-by: Martin Storsjö <martin@martin.st>
If it isn't available in the system, we've got a fallback to
the normal localtime function, so normal code can assume it is
available as long as time_internal.h is included.
Signed-off-by: Martin Storsjö <martin@martin.st>
gmtime isn't thread safe in general. In msvcrt (which lacks gmtime_r),
the buffer used by gmtime is thread specific though.
One call to localtime is left in avconv_opt.c, where thread safety
shouldn't matter (instead of making avconv depend on the libavutil
internal header).
Signed-off-by: Martin Storsjö <martin@martin.st>
This allows writing most code as if they always are is available.
These are ok to use from other libraries even though it's not a
public header, since they only provide an inline declaration, and
doesn't add an actual dependency on lavu internals. (This can be
considered more a build system compatibility fallback than a
libavutil feature.)
Signed-off-by: Martin Storsjö <martin@martin.st>
If the buffer provided to strftime is too small, the buffer contents
are indeterminate - it does not guarantee actually null terminating
the buffer.
Signed-off-by: Martin Storsjö <martin@martin.st>
The ones left using av_gettime are NTP timestamps (for RTCP,
which is specified to send the actual current realtime clock
in RTCP SR packets), and the NUT muxer timestamper, which is
documented as using wallclock time.
Signed-off-by: Martin Storsjö <martin@martin.st>
Since av_gettime() is used in a number of places where actual
real time clock is required, the monotonic clock introduced in
ebef9f5a5 would have consequences that are hard to handle. Instead
split it into a separate function that can be used in the cases
where only relative time is desired.
On platform where no monotonic clock is available, the difference
between the two av_gettime functions is not clear, and one could
mistakenly use the relative clock where an absolute one is
required. Therefore add an offset, to make it evident that the
time returned from av_gettime_relative never is actual current
real time, even though it is based on av_gettime.
Based on a patch by Olivier Langlois.
Signed-off-by: Martin Storsjö <martin@martin.st>
Martin Storsjö
