python-pulse-control (pulsectl module)

Python (3.x and 2.x) blocking high-level interface and ctypes-based bindings for PulseAudio (libpulse), to use in a simple synchronous code.

Wrappers are mostly for mixer-like controls and introspection-related operations, as opposed to e.g. submitting sound samples to play and player-like client.

For async version to use with asyncio, see pulsectl-asyncio project instead.

Originally forked from pulsemixer project, which had this code bundled.

Contents

Usage
Notes
Installation
Links

Repository URLs:

Usage

Simple example:

import pulsectl

with pulsectl.Pulse('volume-increaser') as pulse:
  for sink in pulse.sink_list():
    # Volume is usually in 0-1.0 range, with >1.0 being soft-boosted
    pulse.volume_change_all_chans(sink, 0.1)

Listening for server state change events:

import pulsectl

with pulsectl.Pulse('event-printer') as pulse:
  # print('Event types:', pulsectl.PulseEventTypeEnum)
  # print('Event facilities:', pulsectl.PulseEventFacilityEnum)
  # print('Event masks:', pulsectl.PulseEventMaskEnum)

  def print_events(ev):
    print('Pulse event:', ev)
    ### Raise PulseLoopStop for event_listen() to return before timeout (if any)
    # raise pulsectl.PulseLoopStop

  pulse.event_mask_set('all')
  pulse.event_callback_set(print_events)
  pulse.event_listen(timeout=10)

Misc other tinkering:

>>> import pulsectl
>>> pulse = pulsectl.Pulse('my-client-name')

>>> pulse.sink_list()
[<PulseSinkInfo at 7f85cfd053d0 - desc='Built-in Audio', index=0L, mute=0, name='alsa-speakers', channels=2, volumes='44.0%, 44.0%'>]

>>> pulse.sink_input_list()
[<PulseSinkInputInfo at 7fa06562d3d0 - index=181L, mute=0, name='mpv Media Player', channels=2, volumes='25.0%, 25.0%'>]

>>> pulse.sink_input_list()[0].proplist
{'application.icon_name': 'mpv',
 'application.language': 'C',
 'application.name': 'mpv Media Player',
 ...
 'native-protocol.version': '30',
 'window.x11.display': ':1.0'}

>>> pulse.source_list()
[<PulseSourceInfo at 7fcb0615d8d0 - desc='Monitor of Built-in Audio', index=0L, mute=0, name='alsa-speakers.monitor', channels=2, volumes='100.0%, 100.0%'>,
 <PulseSourceInfo at 7fcb0615da10 - desc='Built-in Audio', index=1L, mute=0, name='alsa-mic', channels=2, volumes='100.0%, 100.0%'>]

>>> sink = pulse.sink_list()[0]
>>> pulse.volume_change_all_chans(sink, -0.1)
>>> pulse.volume_set_all_chans(sink, 0.5)

>>> pulse.server_info().default_sink_name
'alsa_output.pci-0000_00_14.2.analog-stereo'
>>> pulse.default_set(sink)

>>> card = pulse.card_list()[0]
>>> card.profile_list
[<PulseCardProfileInfo at 7f02e7e88ac8 - description='Analog Stereo Input', n_sinks=0, n_sources=1, name='input:analog-stereo', priority=60>,
 <PulseCardProfileInfo at 7f02e7e88b70 - description='Analog Stereo Output', n_sinks=1, n_sources=0, name='output:analog-stereo', priority=6000>,
 ...
 <PulseCardProfileInfo at 7f02e7e9a4e0 - description='Off', n_sinks=0, n_sources=0, name='off', priority=0>]
>>> pulse.card_profile_set(card, 'output:hdmi-stereo')

>>> help(pulse)
...

>>> pulse.close()

Current code logic is that all methods are invoked through the Pulse instance, and everything returned from these are "Pulse-Something-Info" objects - thin wrappers around C structs that describe the thing, without any methods attached.

Aside from a few added convenience methods, most of them should have similar signature and do same thing as their C libpulse API counterparts, so see pulseaudio doxygen documentation for more information on them.

Pulse client can be integrated into existing eventloop (e.g. asyncio, twisted, etc) using Pulse.set_poll_func() or Pulse.event_listen() in a separate thread.

Somewhat extended usage example can be found in pulseaudio-mixer-cli project code, as well as tests here.

Notes

Some less obvious things are described in this section.

Things not yet wrapped/exposed in python

There are plenty of information, methods and other things in libpulse not yet wrapped/exposed by this module, as they weren't needed (yet) for author/devs use-case(s).

Making them accessible from python code can be as simple as adding an attribute name to the "c_struct_fields" value in PulseSomethingInfo objects.

See github #3 for a more concrete example of finding/adding such stuff.

For info and commands that are not available through libpulse introspection API, it is possible to use pulsectl.connect_to_cli() fallback function, which will open unix socket to server's "module-cli" (signaling to load it, if necessary), which can be used in exactly same way as "pacmd" tool (not to be confused with "pactl", which uses native protocol instead of module-cli) or pulseaudio startup files (e.g. "default.pa").

Probably a bad idea to parse string output from commands there though, as these are not only subject to change, but can also vary depending on system locale.

Volume

In PulseAudio, "volume" for anything is not a flat number, but essentially a list of numbers, one per channel (as in "left", "right", "front", "rear", etc), which should correspond to channel map of the object it relates/is-applied to.

In this module, such lists are represented by PulseVolumeInfo objects.

I.e. sink.volume is a PulseVolumeInfo instance, and all thin/simple wrappers that accept index of the object, expect such instance to be passed, e.g. pulse.sink_input_volume_set(sink.index, sink.volume).

There are convenience volume_get_all_chans, volume_set_all_chans and volume_change_all_chans methods to get/set/adjust volume as/by a single numeric value, which is also accessible on PulseVolumeInfo objects as a value_flat property.

PulseVolumeInfo can be constructed from a numeric volume value plus number of channels, or a python list of per-channel numbers.

All per-channel volume values in PulseVolumeInfo (and flat values in the wrapper funcs above), are float objects in 0-65536 range, with following meanings:

0.0 volume is "no sound" (corresponds to PA_VOLUME_MUTED).
1.0 value is "current sink volume level", 100% or PA_VOLUME_NORM.
>1.0 and up to 65536.0 (PA_VOLUME_MAX / PA_VOLUME_NORM) - software-boosted sound volume (higher values will negatively affect sound quality).

Probably a good idea to set volume only in 0-1.0 range and boost volume in hardware without quality loss, e.g. by tweaking sink volume (which corresponds to ALSA/hardware volume), if that option is available.

Note that flat-volumes=yes option ("yes" by default on some distros, "no" in e.g. Arch Linux) in pulseaudio daemon.conf already scales device-volume with the volume of the "loudest" application, so already does what's suggested above.

Fractional volume values used in the module get translated (in a linear fashion) to/from pa_volume_t integers for libpulse. See src/pulse/volume.h in pulseaudio sources for all the gory details on the latter (e.g. how it relates to sound level in dB).

Code example:

from pulsectl import Pulse, PulseVolumeInfo

with Pulse('volume-example') as pulse:
  sink_input = pulse.sink_input_list()[0] # first random sink-input stream

  volume = sink_input.volume
  print(volume.values) # list of per-channel values (floats)
  print(volume.value_flat) # average level across channels (float)

  time.sleep(1)

  volume.value_flat = 0.3 # sets all volume.values to 0.3
  pulse.volume_set(sink_input, volume) # applies the change

  time.sleep(1)

  n_channels = len(volume.values)
  new_volume = PulseVolumeInfo(0.5, n_channels) # 0.5 across all n_channels
  # new_volume = PulseVolumeInfo([0.15, 0.25]) # from a list of channel levels (stereo)
  pulse.volume_set(sink_input, new_volume)
  # pulse.sink_input_volume_set(sink_input.index, new_volume) # same as above

In most common cases, doing something like pulse.volume_set_all_chans(sink_input, 0.2) should do the trick though - no need to bother with specific channels in PulseVolumeInfo there.

String values

libpulse explicitly returns utf-8-encoded string values, which are always decoded to "abstract string" type in both python-2 (where it's called "unicode") and python-3 ("str"), for consistency.

It might be wise to avoid mixing these with encoded strings ("bytes") in the code, especially in python-2, where "bytes" is often used as a default string type.

Enumerated/named values (enums)

In place of C integers that correspond to some enum or constant (e.g. -1 for PA_SINK_INVALID_STATE), module returns EnumValue objects, which are comparable to strings ("str" type in py2/py3).

For example:

>>> pulsectl.PulseEventTypeEnum.change == 'change'
True
>>> pulsectl.PulseEventTypeEnum.change
<EnumValue event-type=change>
>>> pulsectl.PulseEventTypeEnum
<Enum event-type [change new remove]>

It might be preferrable to use enums instead of strings in the code so that interpreter can signal error on any typos or unknown values specified, as opposed to always silently failing checks with bogus strings.

Event-handling code, threads

libpulse clients always work as an event loop, though this module kinda hides it, presenting a more old-style blocking interface.

So what happens on any call (e.g. pulse.mute(...)) is:

Make a call to libpulse, specifying callback for when operation will be completed.
Run libpulse event loop until that callback gets called.
Return result passed to that callback call, if any (for various "get" methods).

event_callback_set() and event_listen() calls essentally do raw first and second step here.

Which means that any pulse calls from callback function can't be used when event_listen() (or any other pulse call through this module, for that matter) waits for return value and runs libpulse loop already.

One can raise PulseLoopStop exception there to make event_listen() return, run whatever pulse calls after that, then re-start the event_listen() thing.

This will not miss any events, as all blocking calls do same thing as event_listen() does (second step above), and can cause callable passed to event_callback_set() to be called (when loop is running).

Also, same instance of libpulse eventloop can't be run from different threads, naturally, so if threads are used, client can be initialized with threading_lock=True option (can also accept lock instance instead of True) to create a mutex around step-2 (run event loop) from the list above, so multiple threads won't do it at the same time.

For proper python eventloop integration (think twisted or asyncio), use pulsectl-asyncio module instead.

There are also some tricks mentioned in github #11 to shoehorn this module into async apps, but even with non-asyncio eventloop, starting from pulsectl-asyncio would probably be much easier.

Tests

Test code is packaged/installed with the module and can be useful to run when changing module code, or to check if current python, module and pulseudio versions all work fine together.

Commands to run tests from either checkout directory or installed module:

% python2 -m unittest discover
% python3 -m unittest discover

Note that if "pulsectl" module is available both in current directory (e.g. checkout dir) and user/system python module path, former should always take priority for commands above.

Add e.g. -k test_stream_move for commands above to match and run specific test(s), and when isolating specific failure, it might also be useful to run with PA_DEBUG=1 env-var to get full verbose pulseaudio log, for example:

% PA_DEBUG=1 python -m unittest discover -k test_module_funcs

Test suite runs ad-hoc isolated pulseaudio instance with null-sinks (not touching hardware), custom (non-default) startup script and environment, and interacts only with that instance, terminating it afterwards. Still uses system/user daemon.conf files though, so these can affect the tests.

Any test failures can indicate incompatibilities, bugs in the module code, issues with pulseaudio (or its daemon.conf) and underlying dependencies. There are no "expected" test case failures.

All tests can run for up to 10 seconds currently (v19.9.6), due to some involving playback (using paplay from /dev/urandom) being time-sensitive.

Changelog and versioning scheme

This package uses one-version-per-commit scheme (updated by pre-commit hook) and pretty much one release per git commit, unless more immediate follow-up commits are planned or too lazy to run py setup.py sdist bdist_wheel upload for some trivial README typo fix.

Version scheme: {year}.{month}.{git-commit-count-this-month}

I.e. "16.9.10" is "11th commit on Sep 2016".

There is a CHANGES.rst file with the list of any intentional breaking changes (should be exceptionally rare, if any) and new/added non-trivial functionality.

It can be a bit out of date though, as one has to remember to update it manually.

"Last synced/updated:" line there might give a hint as to by how much.

Installation

It's a regular package for Python (3.x or 2.x).

If a package is available for your distribution, using your package manager is the recommended way to install it.

Otherwise, using pip is the best way:

% pip install pulsectl

(add --user option to install into $HOME for current user only)

Be sure to use python3/python2, pip3/pip2, easy_install-... commands based on which python version you want to install the module for, if you are still using python2 (and likely have python3 on the system as well).

If you don't have "pip" command:

% python -m ensurepip
% python -m pip install --upgrade pip
% python -m pip install pulsectl

(same suggestion wrt "install --user" as above)

On a very old systems, one of these might work:

% curl https://bootstrap.pypa.io/get-pip.py | python
% pip install pulsectl

% easy_install pulsectl

% git clone --depth=1 https://github.com/mk-fg/python-pulse-control
% cd python-pulse-control
% python setup.py install

(all of install-commands here also have --user option)

Current-git version can be installed like this:

% pip install 'git+https://github.com/mk-fg/python-pulse-control#egg=pulsectl'

Note that to install stuff to system-wide PATH and site-packages (without --user), elevated privileges (i.e. root and su/sudo) are often required.

Use "...install --user", ~/.pydistutils.cfg or virtualenv to do unprivileged installs into custom paths.

More info on python packaging can be found at packaging.python.org.

mk-fg/python-pulse-control

mk-fg

Reviews

Repository Details