Instead of using string contents for type identification use the address
of a constant string array. This should speed up type verifications a
little sind we only need to compare one machine word instead of a full
string. Also, this saves a few strings.
To make clear that types must be compared via address and not string
contents 'type_name' is now called 'type_id'.
This also simplifies the macros for declaring and defining public and
private subclasses.
- We now implement a logic where the sink maintains two distinct
volumes: the 'reference' volume which is shown to the users, and the
'real' volume, which is configured to the hardware. The latter is
configured to the max of all streams. Volume changes on sinks are
propagated back to the streams proportional to the reference volume
change. Volume changes on sink inputs are forwarded to the sink by
'pushing' the volume if necessary.
This renames the old 'virtual_volume' to 'real_volume'. The
'reference_volume' is now the one exposed to users.
By this logic the sink volume visible to the user, will always be the
"upper" boundary for everything that is played. Saved/restored stream
volumes are measured relative to this boundary, the factor here is
always < 1.0.
- introduce accuracy for sink volumes, similar to the accuracy we
already have for source volumes.
- other cleanups.
This of course makes the name 'fixed' a bit of a misnomer. However the
definitions are now like this:
fixed latency: the latency may change during runtime, but is solely
controlled by the backend, the client has no influence.
dynamic latency: the latency may change during runtime, influenced by
the requests of the clients.
i.e. fixed vs. dynamic is from the perspective of the client.
This adds pa_assert_io_context() and pa_assert_ctl_context() in addition
to a few related macros. When called they will fail when the current execution
context is not IO resp. not control context. (aka 'thread' context vs.
'main' context)
In some situations a rewind request travelling downstream might be
optimized away on its way and an upstream rewind processing might never
come back. Hence, call _process_rewind() before each _render()just to
make sure we processed them all.
Completely rework mixer logic. This now allows controlling a full set of
elements from a single sink's volume slider/mute button.
This also introduces sink and source "ports" that can be used to choose
different input or output ports with the UI. (i.e. "mic"/"line-in" or
"speaker"/"headphones".
The mixer paths and device maps are now configered in external
configuration files and can be tweaked as necessary.
If the only stream to render from is muted take samples from the
silence cache. This should shrink memory/cache bandwidth. Again the
gain was not what I hoped for.
While flags should generally be initialized by passing them to
pa_{sink|source}_new() we make an exception for the volume related flags
which may be initilized afterwards, but before _put().
The reference volume is to be used as reference volume for stored stream
volumes. Previously if a new stream was created the relative volume was
taken relatively to the virtual device volume. Due to the flat volume
logic this could then be fed back to the virtual device volume.
Repeating the whole story over and over would result in a device volume
that would go lower, and lower and lower.
This patch introduces a 'reference' volume for each sink which stays
unmodified by stream volume changes even if flat volumes are used. It is
only modified if the sink volumes are modified directly by the user.
For further explanations see http://pulseaudio.org/wiki/InternalVolumes
