audio device input is now handle in its own package which resides in the new input directory
a list of codecs was added to codecutil package to help with multiple packages using the same codecs
Since adding the extra bufSize arg to Lex functions, the test functions using them needed to be updated.
NewAudioDevice was changed to accept a logger to log to instead of creating a new one.
It seems unnecessary to have the rtcp.Client.Done() func, considering that we could use
the rtcp.Client.err channel itself to determine if the RTCP client has been stopped.
We simple wait on a chan receive in revid in the error handling routine, and we check the
'ok' return - if it is false, then the err chan has been closed and we can get out of the
error handling loop. This should also reduce CPU usage significantly.
Removed the command line flags that were used to specifiy local and remote addresses for RTP and RTCP. These are now
derived from the initial RTSP connection and also from the RTSP SETUP method reply.
Now adopting an RTCP client so that the RTP stream from the RTSP server can be maintained past 1 minute.
This change involved some refactor.
The rtcp.NewClient signature has been simplified. There is now a default send interval and name for use
in the source description in the receiver reports. These can be customised if required with the new
SetSendInterval and SetName funcs. The rtcp.NewClient signature now takes an rtp.Client, so that it
can get information from the RTP stream, like most recent sequence number. As a result of this requirement
the rtp package parse file has been extended with some functions for parsing out the sequence number and
ssrc from RTP packets and the RTP client provides getters for these things.
Added H264ID and H265ID consts and added logic to select this const for use in encoder based on mediaType param in NewEncoder. Also now
declaring PMT in NewEncoder so that we can set streamID correctly based on mediaType.
Currently just connecting to RTSP server, requesting OPTIONS, DESCRIBE, SETUP and PLAY. Also creating RTP client and giving
this to process from for the lexer.
The audio lexers need to know how much data they will be receiving unlike video which has a fixed buffer size.
This means that all the lex function will need to be given a buffer size since they are used as a function pointer with the same signature.
audio.go will be used for recording sound from the sound card and mic
it is like audio-netsender but it is a package instead of a command
and without the netsender.
Now that we want buffered senders (as required), the ringBuffer that was after the
lexer has been removed. Instead, we now have an ioext.multiWriterCloser to which the
lexer writes to. This then writes to the encoders, and then encoders write to each of
their own multiWriteClosers, which write to the appropriate senders. We now call
close on the first multiWriteCloser to close down the entired pipeline, as this close
call propogates through each level.
We have removed the outputClips routine as it's not required anymore to get data
from the revid ringBuffer, and have removed other things that were used by this, like
the IsRunning function.
We have also updated tests to work with these changes - they are passing.
The mtsSender now has a ringBuffer and tests have been updated accordingly. The mtsSender now uses an output routine to get data from it's ringBuffer to send.
Revid now uses ioext.multiWriteClosers for encoders to write to so that senders can be closed and therefore any output routines.
We are now using an io.multiWriter rather than the multiSender. Code has been updated inside revid.go to account for this change, and tests have also been updated accordingly. Tests for
the multiSender have been removed. A dummyMultiWriter has been written to inject our own multiWriter during testing.
Now that we're removing the concept of a loadSender, there is no need to have a minimalHttpSender (did not implement loadSender) and a httpSender (implemented loadSender). So we can now have
a single httpSender that implements io.Writer just like every other sender.
mtsSender has been simplified such that load and send are no longer called in Write. Load and Send have removed and logic is now in Write. The logic has been simplified such that it does not
try to send again. On next PR when ringbuffers are added to senders, we will add logic to deal with this.
Added a Write method to senders such that they implement io.Writer. The multiSender now takes a slice of io.writers.
Also modified revid code and tests to account for this chance.
Started working on adding exposure, brightness, saturation and awb flags to revid-cli so that we can control raspivid parameters.
Updated revid logic to give these to raspivid and updated config to consider these parameters in config validation.
Send retry has been removed from the multiSender. This also means there is not need for the active func, because we simply wait until the send is complete or failed to exit the output clips
routine. Tests pertinent to retrying or the active function have been removed.
Added test to check that if a sender is set to retry on send fails,
it will keep trying to send until it is successful, and have all
data that was intended to be written sent off.
Added a test in senders_test.go to check that we can give a multiSender a few senders and have it
write to them correctly when we call multSender.Write(). This involved writing a dummy
loadSender implementation called dummyLoadSender that stores what has been written to it so that
we can check at a later time.
We wish to have a way to check that the 'owner' of the multi sender is still active while it may be doing continual send retries - therefore a function with bool return called active has been added
as a field to multiSender so that we can call this and check whether the owner of the multiSender is 'active' or not.
Added a retry field for the multiSender which will be used to decide whether to retry sending or not. This is being set true if we have a http sender and no other senders.
We are now only employing flv encoding for use with rtmp, and rtmp does not like the inclusion of the 'flv header' at the start of a
stream, so this functionality has been removed along with error handling of the invalidFlvTag error from rtmp in the rtmp sender.
Errors captured from encoding are now sent to the error handler.
We also made multiSender smart so that during a write if sendRetry is on, we first check
that revid is actually running before attempting to send again. If revid is not running we return.
We also now log any send errors inside multiSender.
Removed packetization flag for revid-cli as no longer required.
Packetization will be decided based on outputs.
Removed buffer type definition and Write receiver func in
mtsSender_test.go as this is now defined in revid.go.
Made ringbuffer size and element size consisten no matter the
output methods, as we're now going to only be putting h264 in there.
Modified H264 lex function to take an io.Writer rather than an
Encoder.
Removed destination []loadSender slice from revids fields and
added an encoder []stream.Encoder slice to hold encoders used
during a particular configuration. Each encoder will write to
the desired outputs.
Modified logic regarding encoder and sender setup. We now check
what outputs we have and add encoders to revid's encoder slice
depending on what each output requires.
Modified outputClips routine such that it ranges through revid's
encoders and encodes to them. They then write to the senders and
they handle logic regarding the amount of data they send out
and when. They also handle actions to perform on send failures.
Wrote multiSender struct which will be written to from encoders.
It will then use it's senders to distribute the data accordingly
to senders that work with the encoding from said encoders.
Modified senders so that their load methods no longer take ring
chunks, but rather slices.
Modified senders such that their release methods no longer
perform chunk closing.