--- Log opened Tue Dec 07 00:00:54 2021
06:43 -!- ajonas_ [sid385278@id-385278.helmsley.irccloud.com] has quit []
06:44 -!- ajonas [sid385278@id-385278.helmsley.irccloud.com] has joined #secp256k1
07:06 -!- hg [~halosghos@user/halosghost] has joined #secp256k1
08:25 -!- hg [~halosghos@user/halosghost] has quit [Ping timeout: 240 seconds]
08:27 -!- hg [~halosghos@user/halosghost] has joined #secp256k1
10:45 < real_or_random> hm I really don't understand the CI timeout on master 
10:45 < real_or_random> one of the valgrind builds times out but this after sipa's PR reduced the TEST_ITERS from 16 to 2 (!) for this valgrind build
10:46 < real_or_random> and the 2 is correctly applied here in this other valgrind job:  https://cirrus-ci.com/task/5575801836929024?logs=cat_tests_log
10:47 < robot-dreams> Is the variance high enough that even with TEST_ITERS = 2 it could still time out?
10:47 < real_or_random> I think yes...
10:48 < real_or_random> but I restarted once and it didn't help
10:48 < robot-dreams> I vaguely remember that when I tried it locally, TEST_ITERS = 2 didn't make the build take 8 times shorter than TEST_ITERS = 16
10:48 < robot-dreams> I will try it again right now
10:48 < real_or_random> but this other job took  also 57 min... so for some reason valgrind (on CI) now takes much longer than before 
10:48 < robot-dreams> In the meantime, does it make sense to try rolling back the PR for now :( if even for the reason that it's unfortunate for all CI runs to take longer?
10:48 < real_or_random> I can imagine that we were just lucky for the PR build, or someone else had been merged on master which then made the difference  
10:49 < real_or_random> I think as a quick fix, we could raise the timeouts
10:50 < real_or_random> I can imagine that all of these scatch space creations and destructions are pretty heavy for valgrind https://github.com/bitcoin-core/secp256k1/commit/e2cf77328a07c9d972db6a4a65f65424634b54ab 
10:53 < real_or_random> hm no, it should be just 1, this is not done in a loop
10:56 < real_or_random> or maybe it's just the fact that the scratch space is large? but that sounds wrong.   
10:58 < sipa> there are lots of parts of the tests that run independent of TEST_ITERS
10:58 < sipa> and perhaps there is just too much of that now
11:27 < real_or_random> yes, and as far as I remember some of these can be made dependent on TEST_ITERS
11:29 < sipa> we should probably do a big recalibration of all tests anyway
11:30 < sipa> presumably there are wild differences in how much time is spent on each (even ignoribg O(1) ones)
11:33 < sipa> it's probably not unreasonable to disable some of the more expensive O(1) tests entirely if iters is too low
11:34 < sipa> also, while we're on the topic: we should probably split up the tests between public API tests and internal tests (like bench is), so the tests can run against the actual compiled library too
11:56 < real_or_random> indeed.
11:56 < real_or_random> can you open an issue? :P
11:57 < real_or_random> the unwritten "we should probably" list is already quite long
11:57 < real_or_random> I mean, so is the written list. but that's strictly better
11:58 < sipa> on it
12:23 < roconnor> [Friday, December 3, 2021] [1:38:43 PM EST] <roconnor> There are a couple of places where the infinity flag ends up set without z being 0.
12:23 < roconnor> [Friday, December 3, 2021] [1:39:08 PM EST] <sipa> is that still the case?
12:23 < roconnor> [Friday, December 3, 2021] [1:39:12 PM EST] <sipa> i thought we changed that
12:23 < roconnor> [Friday, December 3, 2021] [1:39:21 PM EST] <roconnor> I was looking at gej_set_ge.
12:23 < roconnor> [Friday, December 3, 2021] [1:40:03 PM EST] <roconnor> sounds like I should PR a change.
14:01 -!- hg [~halosghos@user/halosghost] has quit [Read error: Connection reset by peer]
14:02 -!- hg [~halosghos@user/halosghost] has joined #secp256k1
18:22 -!- luke-jr [~luke-jr@user/luke-jr] has quit [Ping timeout: 240 seconds]
18:22 -!- lukedashjr [~luke-jr@user/luke-jr] has joined #secp256k1
18:23 -!- lukedashjr is now known as luke-jr
18:52 -!- hg [~halosghos@user/halosghost] has quit [Quit: WeeChat 3.3]
--- Log closed Wed Dec 08 00:00:55 2021