[14:26] I need someone who's better at bash parsing than I am: [14:26] lxc-attach -n "$container_name" -- env -u LANG USER=buildbot $PWD/bin/with-xvfb $PWD/bin/test -vvv --shuffle --subunit "$@" | subunit-1to2 [14:26] how do I make that pipe be part of the subcommand? [14:26] currently it's attempting to pipe the output of lxc-attach into it [14:28] Does lxc-attach produce some extra stuff as well as the output of the underlying command then? [14:28] But also, that seems backwards [14:29] If you want v2 output you could just use --subunit-v2. I thought you needed to produce v2 for testrepository and then use subunit-2to1 to convert it into something the master understands, though [14:29] tomwardill: quote the bit after the --? [14:29] It gets a bit awful with "$@" though [14:30] Better to avoid that if we can [14:31] BTW the other thing I noticed is that lpbuildbot/bzrbuildbot/subunittest.py is probably going to need to gain v2 support [14:31] Oh, except that's only when we upgrade the master, never mind [14:31] yeah, that's my backup plan for if the subunit-2to1 doesn't work [14:32] (yes, that previous one was backwards) [14:33] If lxc-attach is producing extra output, can you get it to shut up by adding the -q option? [14:35] I think we want lp-setup-lxc-test to be producing a clean subunit stream without any extra guff anyway [14:35] it's more that my output looks something like this: https://usercontent.irccloud-cdn.com/file/h1zsmSsw/image.png [14:36] Right, I get that, I'm just wondering why you need to avoid the output of lxc-attach being passed to subunit-2to1 [14:36] It could only be a problem if lxc-attach is doing more than just passing through stdout from its inferior command [14:37] I don't know tbh, I've got myself confused [14:38] * tomwardill starts at the beginning again [14:38] * cjwatson makes an LXC container to test. Been ages since I had one of these ... [14:39] I could probably just ship you my worker lxdvm [14:40] Heh, I think lxc-create will still be faster [14:41] I suspect this might end in having to upgrade lpbuildbot/bzrbuildbot/subunittest.py with v2 [14:41] (as you suggested) [14:41] Tricky because praseodymium doesn't have the python-subunit version needed for that [14:41] I think it should be avoidable in this pass [14:41] (We can backport python-subunit if we have to, but let's try to avoid that work just now) [14:42] cjwatson: my current lp-setup-lxc-test https://pastebin.canonical.com/p/Q2ZDHtyf9G/ [14:42] based on a xenial worker/test-lxc [14:42] OK [16:43] oh hey, got postgres to behave [16:43] now just the subunit stuff to solve [16:44] oh no, it's only partially behaving [16:44] boo [16:59] nope, postgres still refusing to start when run via buildbot. Works fine by hand [16:59] * tomwardill stares at it