Hi Ryan, hi Matt! I read your discussion. I thought quite a while about possible options, so here are my thoughts:
(1) Start a global system esd. - esound was designed as a per-user process, not system-wide. Every user can and should be able to use her own default device, which is not possible with one instance. - Users cannot send signals to processes they don't own, so they cannot change the default device by sending SIGUSR1 (see the previous patch I sent you), nor restart the daemon, at least if you don't use some nasty setuid wrapper tricks. - This completely prohibits working with multiple sound cards. (2) Reuse an already existing esd. - First, the socket permissions are already 777, esd checks the owner of the file. With the -trust parameter this check can be disabled. - However, this breaks if the first user suddenly logs off and/or kills his esd. - Same problems as in (1), so this is even more nasty. (3) Don't use esd at all. - That's what Ryan suggested: Ryan Murray [2005-08-18 17:16 -0700]: > I think the best solution looks like this: > - fix dmix to work for everyone > - change libesd to play directly to the sound device (the code is > already there for the fallback case). > - stop starting/using the esound daemon. - We did this change for a while, we didn't use esd at all and had gstreamer output to ALSA directly. Nowadays, dmix is enabled by default on most cards, but it still breaks on too many cards. Also, dmix' mixing quality is way below what esd/polypaudio offer. We want to provide a working sound environment for everybody, so currently we cannot do without a sound daemon yet. - ALSA does not allow different users to open the sound device simultaneously. So you need to check every application that it releases the sound device if it doesn't need it. (4) Allow multiple esds with per-user sockets. - This is the solution I proposed in the patch I sent, and which is currently in Ubuntu. It is symmetric, every user has her own private esd instance, can signal, kill, and configure it. - Ryan said it may break commercial apps. I'm really not too worried about that. First, if an app uses the socket directly instead of using libesd, it is broken by design. Second, most commercial apps I know use OSS and/or ALSA; I doubt that there are many commercial apps which can *only* output to esd's raw socket; frankly, I don't care about them in Ubuntu. - The biggest disadvantage of this approach is the same mentioned in (3) (ALSA does not allow different users to open the sound device simultaneously). (1) and (2) solve this, but are differently broken. To mitigate this, I added the -as 5 parameter; if several users use the same default card, only one can use the soundcard at any given time, but -as 5 will make sure that the card is not blocked unnecessarily. - This works flawlessly if several users use different sound cards, which is not supported by (1) and (2). So after considering all the pros and cons, I decided for (4) since it adds some benefits without breaking anything existing. Your choice for Debian of course, I only want to offer what we have. :-) Thanks, Martin -- Martin Pitt http://www.piware.de Ubuntu Developer http://www.ubuntu.com Debian Developer http://www.debian.org
signature.asc
Description: Digital signature
