Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
This commit print as AV_LOG_VERBOSE the jitter buffer
size. It might be the default value or the value set by application.
Signed-off-by: Eloi BAIL <eloi.bail@savoirfairelinux.com>
Signed-off-by: Martin Storsjö <martin@martin.st>
This commit adds a warning trace when jitter buffer
is full. It helps to understand leading decoding issues.
Signed-off-by: Eloi BAIL <eloi.bail@savoirfairelinux.com>
Signed-off-by: Martin Storsjö <martin@martin.st>
When receiving an RTCP packet, the difference between the last RTCP
timestamp and the base timestamp may be negative. As these timestamps
are of the uint32_t type, the result becomes a large integer. Cast
the difference to int32_t to avoid this issue.
The result of this issue is very large start times for RTSP
streams, and difficulty to restart correctly after a pause.
Signed-off-by: Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org>
Signed-off-by: Martin Storsjö <martin@martin.st>
The packetizer only supports splitting at GOB headers - if
such aren't available frequently enough, it splits at any
random byte offset (not at a macroblock boundary either, which
would be allowed by the spec) and sends a payload header pretend
that it starts with a GOB header.
As long as a receiver doesn't try to handle such cases cleverly
but just drops broken frames, this shouldn't matter too much
in practice.
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>
The RFC spec draft only specifies the "H265" name - there is no
specification saying how to interpret "HEVC" (if such a packet
format is specified it could be an entirely different format).
Since this is a very new standard (still a draft), there is little
need for compatibility with existing, broken implementations. Therefore
remove the extra alias, to avoid the risk of encouraging incorrect
usage.
Intentionally keeping the ff_hevc_dynamic_handler name for the
handler, to use "hevc" consistently as name for the codec instead
of "h265" within the library internals as long as there only is one
single variant in actual use.
Signed-off-by: Martin Storsjö <martin@martin.st>
The warnings are false positives, older gcc versions (such as 4.5)
think the variables can be used uninitialized while they in
practice can't, while newer (4.6) gets it right.
Signed-off-by: Martin Storsjö <martin@martin.st>
This only takes care of decrypting incoming packets; the outgoing
RTCP packets are not encrypted. This is enough for some use cases,
and signalling crypto keys for use with outgoing RTCP packets
doesn't fit as simply into the API. If the SDP demuxer is hooked
up with custom IO, the return packets can be encrypted e.g. via the
SRTP protocol.
If the SRTP keys aren't available within the SDP, the decryption
can be handled externally as well (when using custom IO).
Signed-off-by: Martin Storsjö <martin@martin.st>
Without this, we'd signal a huge loss rate (due to unsigned
wraparound) if we had received one packet more than expected (that
is, one seq number sent twice). The code has a check for lost_interval
<= 0, but that doesn't do what was intended as long as the variable is
unsigned.
Signed-off-by: Martin Storsjö <martin@martin.st>
The code below the comment does not at all relate to statistics,
and even if moved to the right place, the comment adds little
value.
Signed-off-by: Martin Storsjö <martin@martin.st>
Previously, we always signalled a zero time since the last RTCP
SR, which is dubious.
The code also suggested that this would be the difference in
RTP NTP time units (32.32 fixed point), while it actually is
in in 1/65536 second units. (RFC 3550 section 6.4.1)
Signed-off-by: Martin Storsjö <martin@martin.st>
This brings back some code that was added originally in 4a6cc061
but never was used, and was removed as unused in 4cc843fa. The
code is updated to actually work and is tested to return sane
values.
Signed-off-by: Martin Storsjö <martin@martin.st>
The base_seq variable is set to first_seq - 1 (in
rtp_init_sequence), so no + 1 is needed here.
This avoids reporting 1 lost packet from the start.
Signed-off-by: Martin Storsjö <martin@martin.st>
The question can be answered: No, we do not know the initial sequence
number from the SDP. In certain cases, it can be known from the
RTP-Info response header in RTSP though. (In that case, we use it as
timestamp origin, but not for rtp receiver statistics.)
Signed-off-by: Martin Storsjö <martin@martin.st>
This is required by RFC 3550 (section 6.5):
The list of items in each chunk MUST be terminated by one or more
null octets, the first of which is interpreted as an item type of
zero to denote the end of the list.
This was implicitly added as padding before, unless the host name
length matched up so no padding was added.
This makes wireshark parse the packets properly if other RTCP items
are appended to the same packet.
Signed-off-by: Martin Storsjö <martin@martin.st>
This sends NACK for missed packets and PLI (picture loss indication)
if a depacketizer indicates that it needs a new keyframe, according
to RFC 4585.
This is only enabled if the SDP indicated that feedback is supported
(via the AVPF or SAVPF profile names).
The feedback packets are throttled to a certain maximum interval
(currently 250 ms) to make sure the feedback packets don't eat up
too much bandwidth (which might be counterproductive). The RFC
specifies a more elaborate feedback packet scheduling.
The feedback packets are currently sent independently from normal
RTCP RR packets, which is not totally spec compliant, but works
fine in the environments I've tested it in. (RFC 5506 allows this,
but requires a SDP attribute for enabling it.)
Signed-off-by: Martin Storsjö <martin@martin.st>
To use this, set sdpflags=custom_io to the sdp demuxer. During
the avformat_open_input call, the SDP is read from the AVFormatContext
AVIOContext (ctx->pb) - after the avformat_open_input call,
during the av_read_frame() calls, the same ctx->pb is used for reading
packets (and sending back RTCP RR packets).
Normally, one would use this with a read-only AVIOContext for the
SDP during the avformat_open_input call, then close that one and
replace it with a read-write one for the packets after the
avformat_open_input call has returned.
This allows using the RTP depacketizers as "pure" demuxers, without
having them tied to the libavformat network IO.
Signed-off-by: Martin Storsjö <martin@martin.st>
This comment was added in e309128f, in 2002, and has been brought
along since then more or less unmodified.
The first point of the todo was implemented in dbf30963 in 2006,
the second one is not relevant to rtpdec.c (brought along from
rtp.c in 8eb793c4 in 2008) but would be more relevant to the
rtp muxer, although it isn't a good idea anyway.
Signed-off-by: Martin Storsjö <martin@martin.st>
Mainly clean up the RTP statistics code, plus a few other obviously
misindentend lines.
Remove some useless comments, de-doxygenize some comments,
add spacing around operators and fix a typo.
Signed-off-by: Martin Storsjö <martin@martin.st>
