Remove the source from the data_loop with invoke to make sure it
really is removed before we destroy the node.
Based on patch by David Svensson Fors <davidsf at axis.com>
Remove the per-port memory and only use per-node memory. This will
make it possible to share memory between ports in the future.
Keep refs to memory on the buffers and free (close) the memory
when no longer used.
Do not send the offset and size in the add_mem call, just send the
fd and the flags. The area that we need to map from this to find the
meta, chunk and data are sent in a separate call. This should make
it possible to truncate the memory to a larger size to dynamically
allocate more shared memory for a client.
Remove the Id data type, it's not needed.
Don't automatically map memory in remote.c
Pass the original memory type from server to client.
Handle DmaBuf mem in video-play now that the server passed it on.
Add control objects for all controllable properties on ports.
Try to link compatible control properties in autolink.
Allocate shared memory for the output property memory and configure
the io area on the ports when the controls are linked.
Send the shared memfd to clients when the io area is configured.
Add port_set_io support in remote.c, mmap the control io area and
set on the port.
Add some param helpers
Add volume control to export-source update the volume before
sending each buffer.
Reorganize the io parameter ids and objects. Make separate enumerations
for buffer, control, input and output properties.
Add a volume output property to export-source. This is still unused
but will eventually be routed to a PropsIn io area where it can
control the volume of a mixer, for example.
Make a method to find the memory block for the given ptr. We can
use this to find the memfd of the memory if there is any. We can
then remove the Shared metadata on buffers.
Add control properties to audiotestsrc for frequency and volume
and add a test program.
Add size parameter to set_io for additional checks and to make it
easy to know how much control data we can write.
Make it possible to configure multiple io areas on a port by giving
an id to set_io.
Add some types to enumerate the supported ids
Make an area to exchange buffers and one to specify pull ranges.
The idea is to make more area types for controlable properties.
Implement enumeration of IO areas in volume.
We can't allocate the shared meta in shared mem because then clients can
damage it for other clients. Place it instead right after the buffer
metadata array. Filter out the shared metadata for a client, we send it
as part of the client_buffer structure.
Remove pointer metadata, it's not so useful.
Document the layout of the allocated buffers and the shared memory.
Work on metadata to define control parameters
Reset the builder when the filter failed.
Make the plugins create their params in a temporary buffer, we
don't want to do in-place filtering because that is not always
possible. Because the filter now resets on error, we can remove
the reset from the plugins.
Add an argument to pass the result param to the caller instead of
having the caller have to pick it up from the builder.
Improve docs for node, clock and monitor
Pass spa_pod everywhere instead of spa_pod_object.
Pass result argument to spa_pod_filter to make things a little
nicer.
Make enum_params and set_param to configure properties, format
and other parameters. This allows us to remove some duplicate
code and make the properties and parameters much more extensible.
Use the object id to mark the id of the parameter.
Remove the spa_format and spa_props.
We can now make the client-node easier by merging the various
format methods into the params.
Make the stream API more powerful now that we can pass params
around.
Only send data to a client when it has sent a NEED_INPUT otherwise
recycle the buffers immediately.
Explicitly recycle buffers when the client is not going to do this.
We can initialize the transport as soon as we get async notify
from the client when setup completes. Node initialization happens
after that and then finaly the node is initialized and we can send
the transport and fds.
Small cleanups
Make port status SPA_RESULT_OK until events changes it and data
processing can start
Only start pulling on ports in the OK state
Change we way we handle client-nodes, handle them async and continue
processing after they signaled completion
Add a new scheduler that decouples push and pull. It pushes to peer
elements when all inputs are provided and pulls from nodes when all
peer outputs are processed.
Also use double buffering for the client-node input, we process the
output of the previous cycle.
Only process the reuse_buffer if the client is explicitly going to
send them otherwise, recycle right after sending.
Let the tee and mix pass the io status upstream.
Initialize io area with NEED_BUFFER for inputs.
Implement reuse_buffer for the remote.
When the client is explicitly going to send reuse_buffer messages,
set the consumed buffer to INVALID so it doesn't automatically get
reused.
ASYNC is for when the node emits events to signal input and output it
has nothing to do with reuse_buffer
Remove weird PROCESS_INPUT code.
Add a PW property "pipewire.client.reuse". If set, the client-node
doesn't immediately reuse a buffer after sending
PW_CLIENT_NODE_MESSAGE_PROCESS_INPUT to the client. Instead, it waits
for reuse-buffer from the client. The SPA_GRAPH_NODE_FLAG_ASYNC is
used for this, together with adapted logic in process_input().
In stream.c, if the property is set, the handling of incoming buffers
for PW_DIRECTION_INPUT streams is changed. Each buffer has to be
recycled, so we make sure new_buffer is emitted for each intermediate
buffer, if buffer_id in the IO area has moved past some buffers.
Change-Id: I137a12b702b857cc73369930d7029ecbd69d63ff
Add an extra input and output node for remote nodes. We link this
node to all remote inputs and outputs to make it easier to use
the scheduler.
Improve scheduler debug
