18:00:44 <vagrantc> #startmeeting
18:00:44 <MeetBot> Meeting started Thu Jan  7 18:00:44 2021 UTC.  The chair is vagrantc. Information about MeetBot at http://wiki.debian.org/MeetBot.
18:00:44 <MeetBot> Useful Commands: #action #agreed #help #info #idea #link #topic.
18:00:54 <vagrantc> #topic introductions
18:01:10 <vagrantc> please say hello
18:10:22 <vagrantc> #topic open for questions / ask me anything
18:13:41 <Foxboron> :) hellooo vagrantc
18:13:56 <Foxboron> (So you don't feel alone)
18:15:13 <h01ger> hey ho!
18:18:47 <vagrantc> :)
18:19:20 <vagrantc> trying to see that arm/arm64 specific systemd reproducibility issue ...
18:19:42 <vagrantc> seems to be a symbols sorting issue specific to arm...
18:19:49 <vagrantc> but that's a wild guess
18:20:09 <h01ger> from #debian-release today:
18:20:10 * vagrantc wonders about parallelism
18:20:11 <h01ger> [17:23] < h01ger> 2205 sources processed so far (out of 3050), 2121 uploaded, currenly looking at packages with popcon <10
18:20:11 <h01ger> [17:24] < h01ger> guess i will be done by saturday evening. then sunday looking at the few unfiled ftbfs or missing build deps or diffs when building source packages. those are like 10 packages atm.
18:20:11 <h01ger> [17:25] < h01ger> plus there are still 10 key packages in bullseye which havent been built on buildds. (mostly migration issues)
18:20:48 <h01ger> (this was about the 3000 packages mentioned at the very end of http://layer-acht.org/thinking/blog/20201231-no-source-change-source-uploads/ )
18:21:27 <h01ger> quick summary: those 3500 src: package upload fix 3500 missing .buildinfo issues in Debian bullseye
18:23:26 <vagrantc> yay!
18:23:50 <bmwiedemann> so closer to verifiability in practice
18:24:00 <vagrantc> at least the ability to attempt it
18:24:07 <vagrantc> and know what's not verifiable
18:24:53 <bmwiedemann> our official builds dont normalize the hostname, but that is no problem, because I can use the same hostname during verification. Only mtimes are tricky
18:28:14 <vagrantc> i'm curious what variations are tested/normalized across all the various efforts ... i guess some are documented in jenkins for the QA type tests
18:33:15 <bmwiedemann> I explicitly vary cores (1,4) , hostname, date (+0, +15.1y), CPU (host,kvm64) and then there is the natural variations from filesystem, mtimes, randomness, PIDs, ASLR
18:34:39 <bmwiedemann> the 1-core scheduler behaves very differently and that breaks many testsuite assumtions
18:35:08 <bmwiedemann> but OTOH you find interesting race conditions
18:52:08 <vagrantc> bmwiedemann: do you find things that actually fail to build on single-core systems? or a difference between building with only a single thread on a multi-core system?
18:52:37 <vagrantc> i've definitely found reproducibility bugs fixed by building with only 1 thread
18:55:35 <vagrantc> i forget which package, but one bug i found where any degree of parallelism built reproducibly against any other degree of parallelism, but single-threaded built different ... concluded there must be a "fast" task that when run single-threaded would run second and a "slow" task that would run first, but when run in any degree of parallelism, the "fast" task would complete first
19:00:33 <vagrantc> pretty sure all the reproducible buidls testing for debian is done with 2+ cores (and most with more than that)
19:01:18 <vagrantc> bmwiedemann: varying the date by 15.1y ... getting awfully close to 2038 :)
19:22:27 <bmwiedemann> vagrantc: yes, some things fail to build with 1-core - e.g. when parallel make produces a dep that is not explicitly listed as a dep
19:23:46 <bmwiedemann> the python y2038 bug is still open :-(
19:24:02 <bmwiedemann> just 17y and 12 days to go
19:27:09 <vagrantc> python3 still has a y2038 problem?
19:28:19 <vagrantc> https://bugs.python.org/issue34990 ?
19:29:21 <bmwiedemann> yes. that is mine
19:30:52 <bmwiedemann> and because of that one, I deferred all other y2038 testing
19:31:03 <vagrantc> right, that would block a lot...
19:31:26 <bmwiedemann> maybe I should just test-apply that patch in our package
19:32:06 <bmwiedemann> I think, i586 kvm did not even want to start after 2038
19:32:49 <vagrantc> i586 is old, let it get some rest and retire :)
19:33:55 <bmwiedemann> the wine world still needs some of it
19:34:15 <vagrantc> hah
19:34:48 <vagrantc> "the wine was never quite as good once we switched to 64-bit"
19:36:18 <bmwiedemann> is that a pun to the alcoholic beverage?
20:00:22 <vagrantc> well, not too many questions, but thanks for showing yourselves :)
20:00:25 <vagrantc> #endmeeting