Audio/video are in practice not always from the same clock and can have
different behaviours with regards to clock rate and jitter. Handling
them separately generally gives better results for the timestamps output
by the source element.
This is a follow-up to commit 7ee4afac.
This commit cleans up the `Pad{Sink,Src}Handler` by
- Keeping arguments which are strictly necessary.
- Passing arguments by value for the trait functions which return
a `Future`. The arguments which were previously passed by reference
were `clone`d internally and then `clone`d again in most
implementations.
There are unfortunate differences in trait function signatures
between those which return a `Future` and the sync functions. This
is due to the requirement for the arguments to be moved to the
resulting `Future`, whereas sync functions can rely on references.
One particular notable difference is the use of the `imp` in sync
functions instead of the `elem` in functions returning a `Future`.
Because the `imp` is not guaranteed to implement `Clone`, we can't
move it to the resulting `Future`, so the `elem` is used.
This is no longer available as this could lead to building a defined
value in Rust which could be interpreted as undefined in C due to
the sentinel `u64::MAX` for `None`.
Use the constants (e.g. `ONE`, `K`, `M`, ...) and operations to build
a value and deref (`*`) to get the quantity as an integer.
Refactor connect_input_stream in order to avoid an ABBA deadlock
altogether: where in other spots we lock settings then state, here we
were emitting encoder-setup with state held, then locking settings in
the default handler.
We could have changed the locking order in the other spots, but instead
we can also just release the state lock when emitting the signal, which
is good practice.
Fixes#108
Higher value might be too big (in particular with many VPNs `webrtcsink`
won't work with the default 1400 MTU).
1200 is the default value in libwebrtc and it is references in several
places there.
It is now guaranteed that each fragment is at most fragment-duration
long unless the one and only GOP of the fragment is longer than that.
The first (non-EOS) stream determines the duration of each fragment and
all other streams are drained to at most the fragment end timestamp
determined this way.
In addition the next fragment's target time is now at the end of the
previous fragment plus fragment-duration instead of using
first-fragment + N*fragment-duration
regardless of where fragments were split before.
That is, fmp4mux now uses the same strategy as used by splitmuxsink and
as is required e.g. by HLS with regards to the target duration.
This avoids full traversals of the sessions HashMap.
Also fixes accidental session teardown when a producer is stopped while
also in a session as a consumer.
Always first try draining queued data in the loop and only start waiting
if there's nothing to drain right now. Otherwise data might have to be
drained right now but we still wait and nothing is ever waking up the
source pad task again.
Also make sure to not wait multiple times on the same gst::ClockId but
instead unset it after waiting on it and no new one was scheduled in the
meantime. Future waits on the same ClockId will immediately return and
instead we should wait on the condvar if no new ClockId is available.
