[LAD] LADI

PrevNext  Index
DateFri, 20 Nov 2009 22:15:20 +0200
From Nedko Arnaudov <[hidden] at arnaudov dot name>
To[hidden] at lists dot jackaudio dot org, [hidden] at lists dot linuxaudio dot org, [hidden] at lists dot linuxaudio dot org
Follow-UpRui Nuno Capela Re: [LAD] LADI
Follow-UpDanni Coy Re: [LAD] LADI
After recent discussion on IRC I'm loosing faith in whether it is worth
to contribute to linux audio session handling/management. Two reasons
were given why it does not get testing from users. One is that what I
did so far is not mature, has annoying bugs and I'm not wanting to fix
them. The other one is that ladish is not giving more than users already
have with qjackctl. Also it was mentioned that D-Bus is not what users
find acceptable for controlling jack server.

Given the almost missing feedback about LADI development from community
members that could benefit from it, I'm not sure whether I should
continue to contribute. Maybe I should give up on trying to make linux
audio usable for my needs. I could also stop using computers and make
music only by using my guitar. Because alternatives to Linux Audio
(windows/mac) don't suit my needs. Moreover they don't have the
potential to suit. This is why I'm contributing to Linux Audio and not
making VST plugins or something. This anti-dbus movement is getting too
far. If there is no user that accepts my point of view, there is no point
of me contributing.

Because it may be possible that someone has missed the whole point of my
jackdbus and session handling effort, I'll try to explain what I find
wrong/unacceptable in JACK (dbus-less) system as we have it now.

 * JACK server tries to kill clients that are suspected to misbehave and
   cause xruns or expose other kind of bad behaviour. This can result in
   qjackctl (or patchage for example) being killed. IMO, killing control
   apps is wrong. Apps that that don't process audio/midi should be
   treated differently.
 * When jackd is autolaunched, log messages are going to stdout/stderr of
   the app that launched them. This is wrong, unix daemons are supposed
   to have a log file, even if they are per-user. One of the reasons why
   log file is a good thing to have is that it allows to analyze problem
   post factum. This helps a lot if some misbehaviour is rarely
   reproducable.
 * Control apps that start the jack server through jackd know only about
   the parameters that were known at compile time. Somewhat recent
   example, IIRC, jack2 specific parameters (-S) and firewire options
   missing after upgrade of jack because qjackctl does not know about
   them. IMO, control apps should be able to query parameters for jack
   and display the available options to user.
 * Control apps that manage jack connections, are subject to realtime
   constraints. IMO, this complicates control apps development for no
   good reason. This is valid only for jack1. jack2 already uses
   non-realtime threads for port notifications.
 * Implementing control app requires C level program or use of specially
   crafted bindings. It will be good if control apps could be
   implemented with few lines of code in a scripting language as Python,
   Ruby, Perl, etc.
 * JACK graph (clients, ports and connections) API is badly designed and
   is prone to race conditions. Fons talked about this problem recently
   too.
 * Session handling capabilities are suboptimal. Various programs lurk
   here. I'll mention the two most popular ones: qjackctl cant
   save/restore internal state of the programs. It also  cant save and
   then relaunch them automatically. lash cannot manage jack
   settings and cannot restore connections for applications that are not
   linked against liblash. There are other problems but those are the most
   frustrating ones.
 * Hardware port virtualization is suboptimal. it is provided through
   the JACK "system" client. The only reliable ports are first ones, they
   are expected to be the main input/output. If applications wants to
   connect to phones for example it does not know on what ports they
   are. projects/session should be movable to other system, one with
   different hardware setup and [extensive] reconnecting should not be
   required.
 * Hardware port names are not human readble. Aliases exist but are not
   widely used for various reasons. Users should be able to name and
   group their ports to match their hardware cable setup.
 * JACK "system" client is used for non-hardware ports (-X seq).
 * There is no global list of JACK enabled applications.
 * JACK MIDI is not widely accepted. JACK AUDIO + ALSA seq appears to be
   acceptable solution. IMO, sample accurate audio+midi is very
   important.
 * There is no session handling for netjack LAN setups.
 * Session handling apps cannot restore apps to more than one X11 screen (do
   not mix with X11 display).
 * Patchage-like (flowcanvas based) patchbay interface is best what I've
   seen. Unfortunately Patchage does not integrate well with other parts
   of JACK infrastructure.

As you can see, I have collected enough problems to fight. Almost all of
these fixes need new software modules to be written or existing ones to
be rewritten. In past years I've tried to collaborate
with people behind JACK, LASH, Patchage and Qjackctl. At the end, I
think that this attempt was probably wrong, with the only excepton of
jack2 (Stephane Letz) who accepted my jackdbus development into jack2
and we worked toghether on improving it and on the more general
"control API" initiative. The jackdbus code, my first contribution, that
is supposed to improve this mess was rejected by Paul Davis whom
maintains jack1. I've got some patches accepted by Patchage`s
author and maintainer, Dave Robillard, but they were rather cosmetic
ones. This forced me to maintain a software branch (Dave calls it a
fork) called lpatchage. It was supposed to be dashboard for the JACK
system. I also actively participated in lash-0.6.x development. The only
other developper, Juuso Alasuutari, shared some of my ideas, but at the
end we ended with fundamental differences about how session handling
should behave and what lash should become.

As part of my LADI efforts, two people where very helpful. Marc-Olivier
Barre and I managed to create pyjackctl, later renamed to laditools. It
is set of minimalistic but very useful control apps for JACK, a2jmidid
and LASH/ladish. Krzysztof Foltman helped a lot with probably the most
complex app in the laditools suite, ladiconf.

I'd like to mention people with whom collaboration, even if attempted by
me was non-existing: Rui Nuno Capela and Bob Ham. They clearly declared
From start that wont help for various reasons and I appreciate
this. Because this saved my precious time of part-time contributor to
Linux Audio.

In May/June this year, Fons Andriaensen got hit by a forcibly packaged
jackdbus (i'd call it mispackaging of jack) and started a flame war
against D-Bus evilness. I tried to be constructive until I got message
From the packager that dbus was forced, even given that he earlier
stated that he explicitly enabled dbus support in that package for one
reason or another.

In June this year, it become evident that I'm not able to contribute to
LASH anymore and that I want something else. I left the LASH project and
started a new session manager, ladish (project page is at ladish.org).
The first preview was released not long after. It was not yet a real
session manager but it was able to save and restore multiple jack
configs, a foundation for what I call "studio" in ladish. Since then
I've implemented some more stuff and I was hoping that next preview that
will be able to relaunch applications and restore connections between
their jack ports will be ready by the end of this year.

In recent few months, I had less time to contribute to ladish and
development slowed. The anti-dbus statements from various people
continued almost always without real argumentation behind them. For
these that were complains about dbus problems in real setups, I've given
suggestions. Almost always I got ignorace and more anti-dbus statements
in return. Maybe what I did is really unusable by others, despite
it being obviously usable by me. Maybe I suck at trying to help & support
people who seem to think that ladi is not that bad. Or maybe D-Bus
is indeed evil and eats babies and I fail to understand why, even after
I've listen to so many "arguments". Or maybe there are happy people
using jackdbus and laditools (or even lash-0.6.x) and looking toward
ladish. But I dont see them. If community does not share my frustration
with problems of the JACK system I mentioned above and does not accept
my vision that D-Bus is just the most suitable (but not perfect)
technology for what I'm trying to achieve, then it would be better for
everybody if I don't contribute anymore. I can detach from the community
and I can even detach from linux and even computers.

So now is the time to give your positive feedback and constructive
critics. Don't troll and don't start another flame war unless your goal
is to alienate me to stage of me detaching from this community. I will
not respond to trolish and flamish mails, feel free to contact me
with private mails if you prefer so. As I'm cross posting this mail, if
you are subscribed to more than one of the mailing lists, please reply
only to one of them. In order of preference, lad, lau, jack-devel. This
should avoid discussions half-shared between lists. If they happen at
all.

This mail is not supposed to be offensive to anyone. If someone feels
so, I declare that offense is not intentional. I don't deny the right of
different opinion on any subject and thus I have no reason to burn
witches.

-- 
Nedko Arnaudov <GnuPG KeyID: DE1716B0>
PrevNext  Index

1258748219.1017_0.ltw:2,a <87aayhdkbr.fsf at usbix dot spacelabs dot org>