23/10/14 11:08, Georg Koppen wrote:
Hi,
anonym:
Hi,
For years Tails has shipped with an Iceweasel with the (relevant) Tor Browser patches applied. It has worked ok, but it's been high maintenance, and keeping the preferences synced with TBB was a bit of a chore, among other issues. Since Tails 1.2 we've migrated to installing the Tor Browser from the actual (32-bit Linux) TBB tarballs you distribute. We're very much interested in your comments on how we do this, so please have look at our design page: [1]
what does adding Adblock [...] (might be a separate discussion, though)
Indeed, so I'm splitting this into a separate thread.
So, now we assume that the "Tag date" above also is the time when the tarballs are made available for download (and preferable announced on tor-qa@). Given that we want at least a +24 hours, this history doesn't look super promising for our (Tails') plans. I'm not sure the above assumption is very sound, though; for instance, the initial 4.0 tarballs (before the rebuild for POODLE but we're ignoring such exceptions) was announced on tor-qa@ on 2014-10-13 10:19:08 (UTC), which was 7 hours before the tbb-4.0-build1 tag.
Yes, this assumption may not hold in the current model we have, where we build first to see whether we get matching builds and tag later. I am not sure what would be a better model as not getting matching builds is a blocker for us. Would it be possible for you to take the builds announced on tor-qa instead of waiting for an official tag?
Of course, that's also possible -- I wasn't sure *what* I was supposed to look at, so thanks for the clarification!
I re-recalculated the table and got this:
TBB release tor-qa@ (UTC) Timing vs ideal Tails release 4.0 2014-10-13 10:18:04 +31 hours 3.6.5 2014-09-01 06:45:46 +35 hours 3.6.3 2014-07-24 09:10:03 -39 hours 3.6.2 2014-06-09 06:49:20 +36 hours 3.5.3 2014-03-18 18:18:34 +0 hours 3.5.2 2014-02-07 18:14:41 -81 hours
So, it improves the situation slightly overall, but it still looks like half of the Tails releases would have been stalled.
3.6.1 is omitted from the table above since it wasn't announced at all. 3.6 was announced on 2014-04-27 14:18:00 (UTC), which would give a very good timing value, however. I'm not how we would have dealt with that situation. The corollary is, what will we do if it happens again? I'm not sure I have a good answer for that yet, but I suppose we either build the tarballs ourselves (at least if the delay is due to problems not affecting Tails), or postpone our release too, if possible (see below why this may be problematic). At least I'd like our communication to improve *when* such blockers appear for you, so we known that we are potentially blocked.
However, do you think you can become such an upstream for Tails by trying to provide the time window detailed above? If you believe that window is too narrow, I suppose Tails could drop the "same-day release" goal and adopt a "day-after release" goal or possible even later. What do you think is possible? How can we help?
We aim at the "same-day release" as well which means something like starting builds Thursday/Friday before Mozilla's release on Tuesday, getting builds to tor-qa by Monday and (given there are no serious issues popping up) releasing on Tuesday. Having them sent to tor-qa by 18:00:00 (UTC) should be doable.
Ok. The earlier the better, of course, but I assume you aim for that any way since it helps your QA. Still, if there's anything we can do to help this progress from "should be doable" to something more reliable, please let us know.
To get a more concrete understanding of what exactly I'm proposing, let's use the next Tails release, 1.2.1, as an example: It's aimed to be released at 2014-11-25 18:00:00 (UTC), the same day as Firefox 31.3.0esr will be officially released. We'd need the (final) TBB (32-bit Linux) tarballs based on that Firefox version at the latest 24 hours before that, i.e. 2014-11-24 18:00:00 (UTC). Does that seem reasonable? I'm actually genuinely interested in an answer to this specific question, since I'm writing the Tails 1.2.1 release schedule as we speak and may have to adjust it if this turns out difficult for you to "promise".
We can try. How hard would it be for you to promise the day-after-release and while trying to get Tails released the same day, though? Does this buy you anything?
It's not that easy, at least not now. Normally I prepare releases on my own, and I plan my time to be quite flexible around the release dates., but I'm not alone; the release has to be coordinated for the QA well in advance. To change that on short notice may not be compatible with the schedules of the other people involved in QA. Hence, we want a rigid schedule that we can set weeks before the actual date. If same-day releases seems like it will cause a lot of timing problems, we may end up trying day-after releases instead.
However, once our automated testing has replaced manual QA, then this may get more flexible.
Cheers!