According to the ffmpeg documentation, if the given packet is reference-
counted, the writting function will take ownership of this reference and
unreference it later when it sees fit. The caller must not access the
data through this reference after this function returns.

The flag can be usefull to have info about the content of the packet:
* AV_PKT_FLAG_KEY: The packet contains a keyframe
* AV_PKT_FLAG_CORRUPT: The packet content is corrupted.

* Cannot copy timing information if there is no case.
* Frame::getAVStream returns a pointer instead of a reference.

…into fix_timingOfPackets

…ration

* FFmpeg / libav already log a warning and manage this issue (by
creating new dts/pts/duration for the packet).
* These instructions fix nothing.

* Use private attributes:
    * _lastWrappedPacketDuration: to get duration of generated packets.
    * _wrappedPacketsDuration: to get the correct video stream duration.
* Temporary solution of OutputFile::wrap: copy duration of packet wrapped.

getOutputDuration and getStreamDuration methods return duration as
float.

* We do not rewrap all streams, so the global bit rate is different.
* But we check the bitrate of each stream wrapped.

…n testTranscoderRewrap

See issue avTranscoder#173

…ation to packet

Set default log level to WARNING for all tests.

If some data has already been wrapped in the stream.

The assert checks the duration of the stream to the duration of the
output expected.

* All audio packets have not the same size.
* Need to compute it at each time.
* See the ffmpeg way:
https://github.com/FFmpeg/FFmpeg/blob/master/libavformat/mux.c#L521
https://github.com/FFmpeg/FFmpeg/blob/7bf465abf3697dcddfb4021e6db9aeec1724f7ed/libavformat/utils.c#L769

The output audio stream has not the correct number of samples.

Because they may be NULL.

The generator hs to be allocated!

…ency

The macro to log (LOG_WARN...) already instanciate a 'std::stringstream
os'.

Use assert_almost_equals.