They were already converted to the running time a few lines above and
updated inside the buffer. Converting another time for the timeout
causes the timeout to happen at the wrong time, usually much later than
it should.
Test was relying on 'blocksize' property on the source which can
only be used in push mode.
This change in baseparse broke it:
e906197c62
so ensure we are actually in push mode by using pushfilesrc.
When initializing Pad functions in `Pad{Src,Sink}`, we downgrade the
`Pad{Src,Sink}` and upgrade it when necessary. This was implemented
to avoid reference cycles:
`gst::Pad` -> pad function -> `Pad{Src,Sink}` -> `gst::Pad`.
Since `Pad{Src,Sink}` reset the pad functions when dropping, there is
no cycles, so we can use an `Arc<Pad{Src,Sink}>` in the pad functions,
thus saving an `upgrade`.
This is a wrapper around uridecodebin3 and fallbackswitch that allows to
switch to a still frame on errors and automatically retries the source
in the background on errors until a given retry timeout is reached.
- Report a latency:
By design, tttocea608 will output buffers in the "past" when
receiving an input buffer: we want the second to last buffer
in the buffer list that we output to have the same pts as the
input buffer, as it contains the end_of_caption control code
which determines when the current closed caption actually gets
displayed in pop_on mode. The previous buffers have timestamps
decreasing as a function of the framerate, for up to potentially
74 byte pairs (the breakdown is detailed in a comment).
The element thus has to report a latency, at 30 frames per second
it represents around 2.5 seconds.
- Refactor timestamping:
Stop using a frame duration, but rather base our timestamps on
a scaled frame index. This is to avoid rounding errors, and
allow for exactly one byte pair per buffer if the proper framerate
is set on the closed caption branch, and the video branch has
perfect timestamps, eg videorate. In practice, that one byte
pair per frame requirement should only matter for line 21 encoding,
but we have to think about this use case too.
- Splice in erase_display_memory:
When there is a gap between the end of a buffer and the start
of the next one, we want to erase the display memory (this
is unnecessary otherwise, as the end_of_caption control code
will in effect ensure that the display is erased when the
new caption is displayed). The previous implementation only
supported this imperfectly, as it could cause timestamps to
go backwards.
- Output last erase_display_memory:
The previous implementation was missing the final
erase_display_memory on EOS
- Output gaps
- Write more tests
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-rs/-/merge_requests/314>
This is similar to what the standard jitterbuffer does, and
is necessary to avoid deadlocks with the rtpbin session lock,
if the user calls onto any API that requires us to take the
state lock at the wrong time (eg setting the latency property,
clearing the pt map)
They're actually used as a hackish intrusive list around GQueue so using
a different allocator is rather dangerous. Or rather, this is generally
dangerous and shouldn't be done but ...
Also make sure to free items manually in finalize() so that any
contained buffers can also be unreffed.
Contrary to what one might believe, this actually reduces the size of
the structs due to alignment constraints. On Linux x86-64 clang/gcc it
reduces the size of the caption_frame_t struct from 7760 bytes to 6800
bytes, on Windows x86-64 MSVC from 11600 bytes to 6800 bytes.
It also causes simpler and potentially faster assembly to be generated
as the values can be directly accessed as uint8_t instead of having to
extract the corresponding bits with bitwise operations.
It also gives us the same ABI with clang/gcc and MSVC.
