Closed Bug 334707 Opened 15 years ago Closed 14 years ago
.5 to 1 .5 .0 .2 partials for thunderbird look too large
The linux partial MAR is 5.1 MB. http://ftp.mozilla.org/pub/mozilla.org/thunderbird/nightly/22.214.171.124-candidates/ I opened up the MAR file and most of this size is coming from thunderbird-bin 4089150 Apr 19 12:54 thunderbird-bin.patch Not sure why that's so large yet.
We thought this could be a regression of Bug #329047. Stefan re-tested the 126.96.36.199 linux candidate build and it looked ok to him though. But that issue may still be the culprit.
The Linux Thunderbird builds moved build machines between 1.5 and 188.8.131.52, no? It could be that changes due to moving the environment (glibc, etc.) caused enough changes in the way the binary was laid out that the diff is unreasonably large. We're not planning on moving machines again, so we may just have to take the hit on Linux for the 184.108.40.206 release; 220.127.116.11's diffs should be smaller again. Do you remember where 1.5's builds came from (which machine)?
(In reply to comment #2) > Do you remember where 1.5's builds came from (which machine)? It would have come off the Mozilla1.8 tinderbox, with a BuildID for Tb 1.5 of 20051201. The tinderbox has respins during the afternoon of the 1st ; crazyhorse is the linux tinderbox; build log  says "Compiler is -- gcc (gcc (GCC) 3.3.3" For comparision, the current Tb tinderbox (esx-test-vm1) said (3 days ago, was awol) "Compiler is -- gcc (gcc (GCC) 3.3.2 20031022 (Red Hat Linux 3.3.2-1)" and the Firefox box (moz180-linux-tbox) says now "Compiler is -- gcc (gcc (GCC) 3.3.2 20031022 (Red Hat Linux 3.3.2-1)"  http://tinderbox.mozilla.org/showbuilds.cgi?tree=Mozilla1.8&maxdate=1133481600&legend=0  http://tinderbox.mozilla.org/showlog.cgi?log=Mozilla1.8/1133471880.5984.gz&fulltext=1
(In reply to comment #3) > It would have come off the Mozilla1.8 tinderbox, with a BuildID for Tb 1.5 of > 20051201. Hrm... 3.3.2 vs. 3.3.3... I'm actually more curious what the linker versions were (since that's probably more likely to affect layout): [cltbld@crazyhorse cltbld]$ ld -v GNU ld version 18.104.22.168.2 20020802 [cltbld@esx-test-vm1 cltbld]$ ld -v GNU ld version 22.214.171.124.4 20030523 Great. Well, the real question is: if the updates work *and* we haven't broken binary compatibility in [a noticeable] way, are we going to respin/re-release all Linux bits? This is somewhat problematic since, the blogosphere seems to think we've already released... so a respin would be multiple levels of badness.
I'm trying to think about what other ramifications we might have by releasing 126.96.36.199 with a different linker for 1.5. I think we need to reach a wider audience on getting input on this this morning. I'm not too worried about folks who already saw the builds on stage if we do respin. It looks like only 3,000 people found the builds on the ftp directory last night and downloaded it. And linux would be a small percentage of that. For that handful of users, the next time they get updated to 188.8.131.52, would the partial fail, and would they get moved to a full update (which would be ok), or would they get stranded with this bad build.
Another thing to think about. what happens when 1.5.0.x users get updated to Thunderbird 2 (which comes off crazyhorse)? I guess they'd get another large partial patch to get them back to a build linked with 184.108.40.206.2.
won't 1.5 -> 2 be large no matter what linker is used?
Aren't linux users the most likely to have a high speed connection and not care so much about large partials? The issue that concerns me more is the potential binary incompatibility with extensions. Does enigmail work?
(In reply to comment #6) > Another thing to think about. what happens when 1.5.0.x users get updated to > Thunderbird 2 (which comes off crazyhorse)? I guess they'd get another large > partial patch to get them back to a build linked with 220.127.116.11.2. > I thought it had pretty much been decided that we weren't going to offer the 1.5 -> 2.0 update via the update service... We didn't offer 1.0 -> 1.5, and there was a huge debate about it.
Paul, I just downloaded 1.5 on linux and loaded up about:buildconfig in the default start page window. And I can confirm that it was built using gcc 3.3.3. I think this gives us a little more confidence that the build environment on crazyhorse hasn't changed since we shipped 1.5 (at least from a gcc point of view).
For those following at home, the linux en-US build is now available: http://ftp.mozilla.org/pub/mozilla.org/thunderbird/nightly/18.104.22.168-candidates/thunderbird-22.214.171.124.en-US.linux-i686.tar.gz preed rocks.
l10n bits are now available in that directory too. Also, encouraging news: I'm starting to generate partial updates. The bg locale partial update is 696k.
Hardly relevant anymore, closing as WFM.
Status: NEW → RESOLVED
Closed: 14 years ago
Resolution: --- → WORKSFORME
You need to log in before you can comment on or make changes to this bug.