On Thursday, July 19, 2012 19:07:52, Ron wrote: > Ok, so I've just had a long overdue catch-up with Thorvald, and we think > we have a plan that really covers all the bases ... > > We can re-enable speex support in the client, which was only just recently > dropped (so only the client currently in unstable is affected by that), > and since all the clients well back in pre-history should support that > just fine, we can jigger things so that it will be the baseline interop > if celt is not present, and use the existing threshold setting on the > servers to let people select the point at which the number of users with > opus support triggers switching to that. > > Which means we basically get the best of all worlds, we have interop with > existing old clients, we get to drop celt support and so don't have to > worry about getting burned by it, and people will automatically switch to > the wonders of opus (and get lower bandwidth and better quality) as soon > as enough of the connected parties have support for it.
This sounds great. > There's a few things that need testing, but we're reasonably confident > this can fly, and meet all the concerns of almost everyone. > > > There are only two small catches: > > - catch 1. He's about to fly out and will be afk for a week, so he > won't be able to look at this until he gets back. (which is why > I'm writing this now instead of letting him do it) > > - catch 2. The version of murmur currently in testing is completely > broken again due to the zeroc-ice screwup. That wouldn't have > happened if the -2 upload of mumble had transitioned as planned, > but well, you all know the story there ... Let's just call it the results from a communication breakdown. I appologize for my side of that. > So ... Chris, since you're currently the major objector, and opened > this bug with the TC, this question is mainly for you ... This sounds like a very good compromise to me, and I thank both you and Thorvald for the effort in coming up with it. .. And thanks for talking to me again. > What we'd like to do in the meantime, is let the mumble version from > unstable transition to testing now. That will: > > - Unbreak the server for everyone, which currently won't work at all. ... And I can verify that this is the case, as I just attempted to upgrade the version of mumble-server I was running on a server, which results in an error message on every attempted startup: /usr/sbin/murmurd: symbol lookup error: /usr/sbin/murmurd: undefined symbol: _ZN3Ice6upCastEPNS_10ConnectionE Whereby I upgraded to mumble-server from Sid, which works fine. (Thank you for dealing with the zero-ice difficulties, and I'm glad that work will be put to good use.) > - Break the client for people using ancient servers, and who are > talking to people without opus support. ie. not everyone, but a > fair number of people who haven't yet moved, or who can't convince > their friends to move yet. You know the deal there. Minorly sucks but I see no reasonable alternative. > - Most importantly though: minimise the diff that -release need to > review when Thorvald gets back and we have a new upload to make > once again. > > We tossed up which way to go with this (the alternative being to not > let it migrate and inflict the bigger diff on -release and broken > server on everyone) - but this seems to be the lesser evil, since it > will let people get some more testing miles on opus, and people who > would really be put out by that can just put it on hold for a short > time. > > Once Thorvald gets back and we re-add speex, this should all work again > for everyone, and we don't have to kick it out of testing, don't have > to embed a suspect lib, and shouldn't have to leave anyone feeling > hard done by ... > > > Does that sound ok for you? Yes it does. Best suggestion I've heard yet. > If it does, I'll bump the severity of your bug back down to something > not RC (but not close it yet, we'll let the speex enabling upload do > that), and request the release team unblock it. And if there is no > further complaint to discuss, then I guess you can tell -ctte you > have the pound of flesh you sought :) I appreciate your humor. ;-) No objection to the plan. > Otherwise ... well, then I don't know what ... you'll have to > suggest a plan B all your own, because this is the best we have ... > > All Thorvald asked is that people stay calm, so he can actually worry > about working on the code rather than being stressed by the drama :) > > > I'll make this happen if I get your ACK that it works for you too. ACK. Thanks very, very much. I'll give the security researcher working to audit the celt 0.7.1 codec a heads up that Debian is planning to EOL celt altogether, and thank him very much for his offer of support. -- Chris -- Chris Knadle chris.kna...@coredump.us GPG Key: 4096R/0x1E759A726A9FDD74
signature.asc
Description: This is a digitally signed message part.