Skip to content

libpulse implementation that unconditionally fails to connect to the PulseAudio daemon and does nothing else

License

LGPL-2.1, Unknown licenses found

Licenses found

LGPL-2.1
LICENSE
Unknown
COPYING
Notifications You must be signed in to change notification settings

oxij/libcardiacarrest

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

What?

libcardiacarrest is a trivial implementation of libpulse* PulseAudio library API that unconditionally (but gracefully) fails to connect to the PulseAudio daemon and does **nothing** else.

apulse and pressureaudio (which uses apulse internally) are an inspiration for this but unlike those two projects libcardiacarrest is not an emulation layer, all it does is it gracefully fails to provide the requested PulseAudio service hoping the application would try something else (e.g. ALSA or JACK).

Why?

Many apps can be built with PulseAudio support, many people want that support, many other people (e.g. me) freak out at the single mention of linking against PulseAudio libraries (see below).

This project provides a way to produce software artifacts that work for both camps without building two different versions of everything.

The general framework is as follows:

  • You use original PulseAudio headers (bundled) but link against libcardiacarrest libraries at link time. Unlike the original libpulse, libcardiacarrest doesn’t permit static linking, so it’s impossible to f*ck it up. (You can skip this step if you don’t build your software yourself and you are not sufficiently paranoid. Note, however, that I found several instances where stuff was statically linked with libpulse while recompiling my system to use libcardiacarrest. You have been warned.)
  • If you want to use PulseAudio daemon you then simply run your apps in an environment where your dynamic loader can find the original libpulse and everything works as expected.
  • But if instead you point your dynamic loader to apulse/pressureaudio then playback will ignore the daemon and will go over pure ALSA (but many other features will not work).
  • Finally, if you point your dynamic loader to libcardiacarrest then an app that uses libpulse API will get “can’t connect to the daemon” error and (hopefully) will simply switch to another output plugin like ALSA or JACK (mpv does exactly this, YMMV) and not just crash.

The intended use of this is

  • you point your dynamic loader to libcardiacarrest by default,
  • you override LD_LIBRARY_PATH with apulse/pressureaudio and the original libs on app-by-app basis.

But you can also use it backwards by running with the original libs by default and overriding with libcardiacarrest for selected apps (e.g., the browser) or whatever.

Explanation by examples

So you linked your mpv with libcardiacarrest.

  • With original libpulse in LD_LIBRARY_PATH:
    • You start mpv
    • it tries its pulse output plugin, that loads original libpulse
    • libpulse tries the daemon, starts it if it isn’t up already
    • libpulse connects to it, sets everything up
    • mpv pushes decoded waves to libpulse
    • libpulse pushes them to the daemon
    • the daemon pushes them to ALSA
    • ALSA pushes them to the kernel
    • sound starts playing
  • With apulse libpulse in LD_LIBRARY_PATH:
    • You start mpv
    • it tries its pulse output plugin, that loads apulse libpulse
    • apulse starts emulating the libpulse API over ALSA
    • mpv pushes decoded waves to apulse
    • apulse pushes them to ALSA
    • ALSA pushes them to the kernel
    • sound starts playing
  • With libcardiacarrest libpulse in LD_LIBRARY_PATH:
    • You start mpv
    • it tries its pulse output plugin, that loads libcardiacarrest libpulse
    • libcardiacarrest pretends it tries the daemon, fails
    • mpv is like “hm, PulseAudio daemon is not available, let’s try something else”
    • mpv tries another output plugin, say alsa
    • it sets itself up
    • mpv pushes decoded waves to ALSA
    • ALSA pushes them to the kernel
    • sound starts playing

The point is that libcardiacarrest is not an emulation layer, it just gracefully fails to provide the requested service.

Why not just link against the original PulseAudio libs and run no daemon then?

I don’t want a single executable bit of PulseAudio on my machine.

  • PulseAudio is very anti-UNIX and anti-KISS (the opposite of “small, sharp tools” and “tools, not restrictions”).
  • The code is enormous (ENORMOUS), over-engineered and of overall pretty shitty quality. It’s impossible to tell how many zero-days it contains.
  • I have yet to meet a single person who still wanted to run PulseAudio on their machine after reading PA sources. All people I’ve shown the sources to were either laughing, crying or both at the end of the session. Yes, that includes the library (see src/pulse.c for some notes on the subject).
  • I don’t like the ethics of pushing over-engineered software onto under-informed users who are unable to resist the tyranny of free cycles (binary packages).
  • Hence, I consider any software that is deliberately PulseAudio-only to be unethical (if not malicious). There are equivalent libraries with similarly convenient APIs (e.g. openal, libao) that are similarly (actually, more) cross-platform that (transparently) provide a choice over audio backends. Shoving PulseAudio down everyone’s throats is not a service to users.

http://i.imgur.com/K6dAvXn.jpg

Well, okay, any technical arguments against it?

  • See comments in src/pulse.c.
  • Xorg is a daemon, Wayland is a library. Yay, Wayland! PulseAudio is a daemon, ALSA is a library. Yay, PulseAudio? PulseAudio is exactly X11 for audio. The library-to-daemon protocol is similarly complicated, protocol parsing code is written by hand (not autogenerated, like for xcb), when exposing PulseAudio to the network all of that code gets exposed to the network, to hide that shitty code from the scary net they introduce yet another complicated authentication protocol on top of it, exactly like X11.

    Imagine exposing init system’s administrative interface (e.g. systemd’s systemctl call) to an untrusted network. Sounds crazy, right? PulseAudio does pretty much this. PulseAudio is an audio daemon, but it’s also a plugin system, small init system (for its plugins) and authentication system at the same time. The libpulse library also includes its own alternatives to iconv, glib and a bunch of slightly-different functions from libc. This is pure crazyness (very typical of Lennart Poettering creations, you might have heard of systemd, dbus and avahi, all of that is him). Pushing uninformed people into running PulseAudio connected to untrusted data sources is unethical.

    Yep, I’m looking at you, Firefox developers, all your sandboxing means exactly nothing if the attacker gets to the PA socket.

  • PulseAudio clients can gather all kinds of information about your hardware setup and other connected clients (see pa_context API and pulse/introspect.h in PA headers).
  • PulseAudio clients can helpfully start/restart the daemon for you.
  • PulseAudio clients can reconfigure the daemon on the fly.
  • PulseAudio clients can dynamically load and unload modules into the daemon. And PulseAudio ships with a bunch of crazy modules by default.

Even if all the code paths in the original libpulse with no daemon available would have lead straight to errors (which is not the case, it does a lot of shit even without a daemon and tries to spawn a daemon at every corner) I would still prefer to link against libcardiacarrest in case I f*ck something up in my config.

Just read the sources (and then you, too, would try to stay away from it forever).

But an app I want doesn’t run without PulseAudio!

Use apulse and/or pressureaudio.

But I need an audio daemon!

Like with X11, most likely, you do not. ALSA does software mixing by default for a decade now (but a lot of documentation on the net is outdated). Primary sound cards, bluetooth headsets, per-application soft volume, loopback (aka “desktop sound”) capture (comes out-of-the-box with ALSA 1.1.6) and quasi-dynamic output switching can be configured with a couple of lines of asound.conf.

However, like with X11, sometimes you do need a daemon, and if you need an audio daemon you should use sndio or JACK (I prefer v1 as it doesn’t require dbus). Their library-to-daemon protocols are much simpler (sndio’s is simply trivial), library API is much cleaner (sndio’s one fits on a single page), they don’t lag, they introduce no latency, and they also do MIDI, not just PCM. Unfortunately, neither of them is a default choice in Ubuntu, so relatively few apps target them (but almost all pro-audio apps have JACK support).

For uncooperative apps you can use apulse to go PA->ALSA and then alsa-plugins to go ALSA->JACK. Works for me, YMMV.

If you have to run PulseAudio at least don’t simply attach it to untrusted data sources. Including the browser. I recommend you to run Firefox with apulse/pressureaudio to make it play over pure ALSA and then use pulse module of alsa-plugins to go from ALSA back to PA. Yes, I really do recommend doing apulse->ALSA->PA->ALSA instead of the normal libpulse->PA->ALSA. This way Firefox gets no direct access to PulseAudio daemon.

Implementation details

libcardiacarrest only implements PA mainloop and context creation, everything else either pretends to do something but does absolutely nothing, returns errors, or simply aborts the whole process (when it’s unclear how to do something saner without writing a lot of code).

libcardiacarrest wants to be the minimal amount of code required to make all the (sane) apps think that libpulse works, but the daemon is not available. Insane uses of the API like modifying lots of PA-structures before ever trying to connect to the daemon will cause the process to abort, it’s by design, fix your app (or report my misunderstanding of your sanity).

In total, libcardiacarrest logic is implemented in ~100 LOC of fairly trivial code, everything else is just PA cruft (yep, it’s a lot).

The only dependencies are libc and glib.

What builds with libcardiacarrest?

Everything. APIs are identical.

What works (gracefully fails) with libcardiacarrest?

I’m running with libcardiacarrest on all of my machines without any issues. Everything I ever tried running under this works. Xfce desktop, KDE5 desktop, firefox, chromium, mpv, mplayer, cmus, SDL1, SDL2, various official and unofficial pulseaudio tools, etc.

Everything that failed to work with this (pamixer, cubeb) was fixed and patches were accepted upstream.

License

LGPLv2.1+ (same as bundled PulseAudio headers). See COPYING.

Contributions

Accepted both via GitHub issues and PRs, and via email (including patches formatted with git-format-patch).

Contributor HOWTO/FAQ

  • Read the src_shell{git log}. Emulate it.
  • To update PulseAudio:
    • Run src_shell{./3rdparty/pulseaudio-headers/update.sh /path/to/pulseaudio-source} from the root of the repository.

      A better way is to actually build actual libpulse and steal its headers. Alternatively, you can also run src_shell{diff -Nuar pulseaudio-r1/src/pulse pulseaudio-r2/src/pulse} between two PA releases and check if any interesting .h were added or removed.

      The update.sh script should be good enough most of the time, though. PulseAudio doesn’t add/remove new headers that often.

    • Run src_shell{git diff ./3rdparty/pulseaudio-headers/pulse/version.h}.
    • Update version.mk file accordingly.
    • Run src_shell{git diff ./3rdparty/pulseaudio-headers/pulse}.
    • If something except version.h changed, update src accordingly.
    • If something non-trivial in src was changed, bump CA_VERION in version.mk.

About

libpulse implementation that unconditionally fails to connect to the PulseAudio daemon and does nothing else

Resources

License

LGPL-2.1, Unknown licenses found

Licenses found

LGPL-2.1
LICENSE
Unknown
COPYING

Stars

Watchers

Forks

Packages

No packages published