19:08:05 <ryansb> #startmeeting Core Team Meeting 19:08:05 <zodbot> Meeting started Tue Jan 10 19:08:05 2017 UTC. The chair is ryansb. Information about MeetBot at http://wiki.debian.org/MeetBot. 19:08:05 <zodbot> Useful Commands: #action #agreed #halp #info #idea #link #topic. 19:08:05 <zodbot> The meeting name has been set to 'core_team_meeting' 19:08:19 <ryansb> #chair abadger1999 alikins mattclay gundalow jimi|ansible bcoca 19:08:19 <zodbot> Current chairs: abadger1999 alikins bcoca gundalow jimi|ansible mattclay ryansb 19:08:31 <abadger1999> Óla 19:08:45 <ryansb> #agenda https://github.com/ansible/community/issues/148 19:08:56 <ryansb> ok, maybe that's not a #. 19:08:59 <ryansb> #link https://github.com/ansible/community/issues/148 agenda 19:09:27 <ryansb> is podlesh here? 19:09:36 <podlesh> yes 19:09:40 <allanice001> ryansb, !link I think 19:09:42 <ryansb> #topic Feature: Serialized group-by https://github.com/ansible/ansible/pull/19707 19:09:53 <gundalow> #help 19:10:04 <ryansb> it's # for bot commands 19:10:07 <ryansb> at least on this bot 19:10:16 <allanice001> ack 19:10:34 <bcoca> i thnk that is composing too much, a group_by for host list is fine, combine with serial then if you want 19:10:42 <bcoca> serial_group_by seems ultraspecial 19:10:51 <podlesh> it's not the same 19:10:54 <bcoca> i had plans for a sort: for hosts 19:11:05 * gundalow is only half here, so may not get an instant response from me 19:11:16 <bcoca> a grouping function seems fine, but we can probably do that via filter/lookups also 19:12:00 <podlesh> problem is, that I want to have N servers in each batch 19:12:09 <podlesh> I can do serial: N 19:12:26 <podlesh> but it just takes first N servers 19:13:17 <podlesh> so yes, sorting host list + serial would achieve the same, under some special circumstances 19:13:46 <bcoca> well, its what my first look at the code shows you doing, you resort the servers by the groups and then apply serial if set 19:13:54 <podlesh> yes 19:14:11 <podlesh> without this, I need to execute several plays 19:14:32 <podlesh> which takes quite a long time 19:14:39 <bcoca> 2 at least, but thinking of a sort: directive that allows you to do this (and more) 19:15:00 <bcoca> time taken would be same as you just change the logic from playbook to hardcoded .... 19:15:10 <podlesh> hmmm, but that would require that each cluster has exactly the same number of nodes 19:15:29 <podlesh> (sort + serial) 19:15:46 <ryansb> So you want to do serial, but stop on groups? 19:15:47 <bcoca> mostly, leaving 'last batches unpaired' 19:16:15 <bcoca> ryansb: wants to intersperes different groups in the order so each batch does 1 host from each group 19:16:24 <podlesh> yes, that 19:16:38 <podlesh> basically serial for each group, but all groups in parallel 19:16:39 <bcoca> g1, g2, g3, serial3: batch one is h1_g1, h1_g2, h1_g3 19:17:09 <bcoca> which you can do right now but you need a 'pre play' that does fancy group_by 19:17:48 <bcoca> podlesh: not against the functionality, but i would prefere a more generic 'sort' function and no need to tie to serial directly 19:18:15 <bcoca> currently we have sort: inventory 19:18:41 <podlesh> yes, but there is one issue I think: what if each group have different number of members? 19:19:02 <bcoca> podlesh: also, what if a host is in multiple groups? 19:19:24 <podlesh> so batch one is h1_g1, h1_g2, h1_g3 and batch two is h2_g1 and h1_g3 19:19:46 <bcoca> having diff number of members can easily be accounted for by just exauhsting them as the groups have them 19:19:53 <bcoca> yep 19:21:10 * MichaelBaydoun is late and lurking 19:21:31 <podlesh> host in multiple groups is not a problem - in my implementation, there is only single key computed for each host 19:21:47 <podlesh> it might be group, or list of groups, does not matter 19:22:54 <bcoca> it does as a host can only be run once 19:23:57 <bcoca> i dont see how you prevent appending the host multiple times 19:24:02 <podlesh> yes, it will 19:24:41 <podlesh> do you mean if there is single host multiple times in the original host list? 19:24:50 <bcoca> yes 19:25:06 <bcoca> i dont see where you unique them, it will runt he host mulitple times 19:25:47 <podlesh> in that case yes, it will be added multiple times 19:25:50 <bcoca> since you construct the list every time and avoid the all hsots 19:26:53 <podlesh> is it possible to have one host in multiple instances with different instances of Host with different data? 19:27:01 <bcoca> also you are evaluating the condition per host, but what if it includes hostvars? at this point you only have extra/play and host + groups 19:27:27 <bcoca> podlesh: yes, and they are always uniqued when the play references them, that is how a host can be in multiple groups and even inventories 19:27:44 <bcoca> conflicts get resolved by overwrrite (default, there is also merge option) 19:28:02 <podlesh> I was under impression this is done before _get_serialized_batches 19:28:18 <bcoca> yes and i believe your patch removes that 19:29:24 * abadger1999 rings the 30minute bell. 19:29:35 <podlesh> ok, that would be serious error, indeed 19:29:40 <bcoca> and the jinja expresion is limited and confusing, i would prefer to move that to future inventory plugin (constructed groups) 19:29:57 <bcoca> so i like the functionality, but im -1 for the current implementation 19:30:28 <bcoca> i think this belongs broken into smaller features: a) constructed groups by jinja expression, b) user controllable host sorting 19:31:16 <bcoca> they both can be currently worked around by creative use of play + group_by and filters 19:33:24 <abadger1999> Is there an immediate action here to enable podlesh to move forward? 19:33:26 <podlesh> that kind of workarounds are usually quite complicated and limited because it would require loop over play 19:34:58 <podlesh> breaking into smaller features is ok, although I am not still exactly sure what would that sorting look like 19:35:30 <bcoca> sort: invenotry|alpha|reverse|numeric|numeric_desc|expression 19:35:34 <abadger1999> bcoca: could you point podlesh at which files he should look at? 19:36:12 <bcoca> teh constrcuted groups, its an 'inventory plugin' i've been thinking of for a long time (group_by rules in inventory instead of play) 19:36:18 <bcoca> abadger1999: if they existed .... 19:36:45 <podlesh> yes, I can imagine sorting of host list, but still can't see how to combine that with serial 19:36:54 <bcoca> my issue with sort: expressoin (which this functionality is trying to do) is the limited availability of vars/attributes at this point 19:37:04 <bcoca> serial: 3 19:37:27 <podlesh> so, basically, matrix transposition 19:37:29 <bcoca> same way as above, you dont need to, as long as you sorted by group, set serial to group size 19:37:50 <podlesh> not groups size, but number of groups 19:38:01 <bcoca> yeah 19:38:04 <podlesh> problem is, all groups would have to have the same size 19:38:19 <podlesh> and smaller groups would have to be somehow "padded" 19:38:45 <bcoca> not really 19:38:55 <bcoca> serial is adjustable 19:39:48 <abadger1999> k, we're at 40 minutes. We should probably move on to another topic. 19:40:11 <podlesh> ok 19:40:13 <ryansb> Fair, further discussion can go on the issue 19:40:24 <bcoca> yeah, this is not a simple topic to discuss, also i would like to mull more options 19:40:28 <abadger1999> podlesh: can you ask bcoca questions about linking to serial outside the meeting? And if you don't get enough info we can discuss more next meeting. 19:40:47 <ryansb> Next up: jbaublitz is not around, but he talked to me about this beforehand 19:40:49 <podlesh> maybe as comments to that pull requests? 19:41:06 <ryansb> Yeah, I meant pull request when I said issue. Oops 19:41:15 <ryansb> #link https://github.com/ansible/ansible/pull/19707 19:41:30 <ryansb> Next up is https://github.com/ansible/community/issues/148#issuecomment-271029293 19:41:42 <bcoca> https://github.com/ansible/ansible/issues/19707 19:42:02 <ryansb> #topic Google Cloud modules google-api-python-client discussion 19:42:55 <bcoca> taht is common issues with cloud apis, i believe aws modules sovle this by doing a retry/backoff 19:43:13 <ryansb> Currently the GCE modules currently use libcloud, which provides for GCE and a bunch of other providers 19:43:39 <ryansb> newer Google services & features get into the GCP python client way before they go into libcloud 19:43:41 <bcoca> gce is mostly maintained by google staff itself, not sure i would push them to change underlying lib, they chose libcloud 19:44:53 <ryansb> Who is "them" in this case? 19:45:13 <bcoca> tomseomtnng and ericotherthing 19:45:24 * bcoca has issues with names 19:45:48 <ryansb> When they chose libcloud the google-api-python-client didn't actually exist 19:45:51 <bcoca> they basically created most of the modules and maintain them 19:46:24 <abadger1999> supertom and ericjohnso 19:46:25 <ryansb> So what I want to discuss here is: 1. Is there any opposition to having modules that use google-api-python-client? 19:46:31 <bcoca> i dont know enough of the ecosystem or penetration of the libs to decide, i think we should have the maintainers weigh in before we respond 19:46:51 <ryansb> Fair enough, I'll get Tom&Eric's feedback before next meeting 19:47:13 <bcoca> yeah, i dont think we should make any decisions w/o pulling them in 19:47:47 <ryansb> Fair enough. Let's table this until they can comment 19:48:18 <ryansb> Zhenech and docs-in-the-sdist 19:48:33 <Zhenech> yeah *wakes up* 19:48:38 <ryansb> #topic Including docs in Ansible's sdist 19:48:40 <ryansb> #link https://github.com/ansible/ansible/pull/20004 19:49:13 <Zhenech> docs and tests, actually, but this can be split if needed/wanted 19:49:21 <abadger1999> jimi|ansible: ^ 19:49:49 <bcoca> just thought that now github tarballs ARE usable .... we might not need to change anything and just point those that want 'moar' to github 19:49:56 <abadger1999> I'm moderately for including them. (I've requested tests before). Maybe jimi|ansible can say why he's been against it in the past. 19:50:12 <jimi|ansible> why would we want to? we have ansible-doc and tests generally require a large infrastructure to run now 19:50:13 <bcoca> abadger1999: i believe requiremetns for tests would have to be added 19:50:19 <jimi|ansible> i am not in favor of doing that at all 19:50:28 <abadger1999> bcoca: if the goal is to get these into the hands of the packagers then github tarballs won't achieve that. 19:50:31 <Zhenech> docs was a request from a debian user, who said that "ansible-doc <foo>" is not sufficient while offline as it only outputs module doc 19:50:37 <bcoca> abadger1999: how so? 19:51:04 <jimi|ansible> ahh i see, well i could see expanding ansible-doc to include the full docs, but still no to tests 19:51:12 <bcoca> Zhenech: i've had longstanding plan for that to cover all plugins .. but not a tutorial on how to use ansible 19:51:34 <abadger1999> bcoca: because packagers want to work off the released tarballs. they're more official and oftentimes tags on github are bad (projects move tags and thetagged source doesn't include things that are built as part of a release tarball) 19:51:45 <abadger1999> jimi|ansible: tests are very useful for packagers. 19:51:54 <Zhenech> bcoca, as a packager I prefer the tarball that was blessed by the developers as "this is our release" 19:52:02 <Zhenech> and that is that what "make sdist" generates 19:52:04 <bcoca> abadger1999: then we can just make sure that release tarballs/tags match, that is under our control and will be minor change 19:52:07 <abadger1999> jimi|ansible: I've been putting tests into the Fedora and epel packages since we shipped 2.0 19:52:28 <mattclay> abadger1999: Which tests? 19:52:32 <abadger1999> bcoca: but distros tell people not to use github tags when tarballs are present. Because that's the generally better rule. 19:52:35 <bcoca> abadger1999: give 2 optoins, the lean tarball (existing one in releases.ansible.com, fat one from github) 19:52:48 <abadger1999> mattclay: I ship the entire tests directory but only run the unittests. 19:52:55 <jimi|ansible> bcoca: that's not easy to do with sdist 19:53:09 <abadger1999> mattclay: With ansible-test I probably could start doing a subset of the integration tests too. 19:53:12 <bcoca> jimi|ansible: not saying we should do via sdist 19:53:18 <mattclay> abadger1999: I assume you're only running a subset of the unit tests, due to the lengthy requirements list for them? 19:53:24 <alikins> +1 for including that stuff in sdist and/or release tarball 19:53:29 <bcoca> im saying packagers CAN do it as long as we make sure releases on github match 19:54:16 <abadger1999> fat one isn't very much fatter. 19:54:26 <bcoca> but has diff requirements 19:54:36 <bcoca> taht is bigger issue by far, imo 19:54:53 <bcoca> +5mb i dont think is a big deal unless you are using sat phone 19:54:59 <abadger1999> Zhenech's measurment was a total of 1MD more for including both tests and docs. 19:55:01 <allanice001> bcoca, alikins, mattclay why not build a "blessed package" or "blessed rpm" from shippable ? 19:55:02 <bcoca> having +10 new requirements is 19:55:03 <Zhenech> optional requirements, no? 19:55:21 <Zhenech> nobody has to build the docs and run the tests 19:55:22 <abadger1999> bcoca: there are no more new requirements. 19:55:30 <Zhenech> (and the docs are actually readable in rst form too) 19:55:31 <mattclay> So there's really 3 things here: including docs, including tests, running tests as part of packaging 19:55:33 <abadger1999> requirements are runtime requirements, not test-time requirements. 19:55:34 <bcoca> allanice001: we dont build rpm, we build ppa + tarball (and push to pypy) 19:55:50 <abadger1999> so our deps do *not *increase. 19:55:54 <mattclay> Well, 4 if we count building docs 19:55:55 <allanice001> gotcha 19:56:34 <bcoca> abadger1999: depends on system, most packagers will require tests to run before package is accepted, now you ship makefile with tests .... now they need to deal with new reqs 19:56:59 <ryansb> mattclay: 4MB? 19:57:15 <bcoca> ^ but again, i think its mute point, now with unified repo, github tarballs CAN be used normally 19:57:16 <abadger1999> bcoca: People who want to run tests will add the deps *to the package build time* if they want. 19:57:28 <abadger1999> There's still no increase in runtime deps. 19:57:37 <Zhenech> in debian we even have test-deps 19:57:39 <bcoca> packaging deps 19:57:40 <abadger1999> And packagers are free not to run the tests as well. 19:57:43 <Zhenech> so test can run at whatever other time we want 19:57:58 <Zhenech> (and other infra, and everything) 19:58:12 <bcoca> -1 to change existing tarball, +1 to point them at github one 19:58:14 <mattclay> ryansb: No, I just meant there's 4 topics here that can be decided separately. 19:58:23 <abadger1999> bcoca: Only if you mean build-time/test-time packaging deps... those do *not* leak over into an rpm -i ansible, apt-get ansible, or pip install ansible. 19:58:43 <abadger1999> +1 to changing existing tarball 19:58:51 <bcoca> abadger1999: never said that, but those need to be maintained and specified or most packager workflows break 19:59:52 <abadger1999> bcoca: not at all. We manage to run the tests on our local systems without it and the fedora package manages to build without it. 20:00:00 <ryansb> ah, gotcha 20:00:03 <Zhenech> the tarball as generated by my pr will not break anything in the debian packaging (I did not try the rpms) 20:00:24 <Zhenech> as far as the debian build infra is concerned it only sees more files 20:00:41 <bcoca> https://github.com/ansible/ansible/archive/v2.1.4.0-0.1.rc1.tar.gz 20:00:44 <mattclay> If we're including the tests, we should be clear that we're not running most of them when packaging. 20:00:53 <Zhenech> the makefile is there already now, and its not used 20:00:55 <bcoca> ^ i really dont see need dto change existing when auto function already does this 20:01:04 <abadger1999> Zhenech: it won't break the rpms either. 20:01:18 <Zhenech> abadger1999, thought so 20:01:19 <bcoca> https://github.com/ansible/ansible/archive/v2.2.1.0-0.3.rc3.tar.gz 20:01:23 <bcoca> ^ latest rc 20:01:32 <abadger1999> bcoca: packagers won't use it. 20:01:43 <bcoca> then that is on them 20:01:49 <abadger1999> If there's an official tarball, packagers will use the tarball in preference to the github downloads. 20:02:00 <abadger1999> github downloads are just too broken in general for projects. 20:02:14 <bcoca> i'll even go so far to say ... we should stop pushing tarballs now that we have unified repo 20:02:37 <Zhenech> you have to push them to pypi anyways? 20:02:52 <bcoca> we can push the github one 20:02:57 <bcoca> same with ppa 20:03:04 <bcoca> well, we build deb, that is diff 20:03:08 <alikins> I tend to agree to making the github tarballs the official ones and the ones used by packagers if thats workable. (assuming some details like providing checksums and/or sigs for github release tarball) 20:03:15 <Zhenech> that would make the github one the official one 20:03:17 <Zhenech> thats also fine 20:03:24 <abadger1999> bcoca: You'd have to make a mnual script if you want to push non-sdist created tarballs to pypi. 20:03:46 <abadger1999> bcoca: but yeah, what Zhenech said is true about doing that and this issue. 20:04:07 <abadger1999> alikins: I don't think we can provide checksums and sigs. 20:04:07 <bcoca> not familiar with sdist so cannot speak to that 20:04:19 <abadger1999> alikins: Because git doesn't preserve file timestamps right? 20:04:26 <bcoca> we can, just need to add it to the github realease 20:05:08 <bcoca> github even allows us to host our own tarballs insted of autogenerated 20:05:19 <bcoca> ^ they didnt before, that is why we have releases 20:05:22 <bcoca> .ansible.com 20:05:26 <abadger1999> bcoca: do they? They stopped doing that a few years ago... didn't know they restored it. 20:05:30 <bcoca> https://github.com/ansible/ansible/releases/new 20:05:38 <bcoca> they do, i tested a while back 20:05:48 <bcoca> see 'attaching binaries' 20:05:54 <jimi|ansible> we don't push tars to github, they generate them on their own, and it's basically just the full repo 20:06:04 <bcoca> we dont, we could 20:06:15 <jimi|ansible> what does that buy us over pushing them to pypi? 20:06:23 <abadger1999> Neat. I'll have to tr that out on some repos of my own. 20:06:25 <jimi|ansible> we still have to do that, now we'd just have two places to push them? 20:06:32 <bcoca> people stop using autogenerated ones? 20:06:50 <bcoca> we already push to 2 places (releases site) 20:07:21 <bcoca> but im also fine with NOt pushing to github, just suggesting as alternative 20:07:26 <jimi|ansible> ok so three then :) 20:07:49 <bcoca> it was replacment suggestion, not addition, but both/either work for me 20:09:00 <alikins> using the github releases as official seems reasonable to me. And encourage packages to use that. If we need to gen a sdist from that for pypi convenience, that seems okay too. 20:09:03 <bcoca> https://github.com/ansible/ansible/releases/tag/v2.2.1.0-0.3.rc3 < =release happens automatically, the issue here is us taking control over it or just keep correcting people when they download this by mistake (but now they DO have modules) 20:09:37 <alikins> assuming github releases dont cause more pain for packages (re "Tue 15:02| <abadger1999> github downloads are just too broken in general for projects.") 20:09:44 <alikins> packagers 20:09:46 <bcoca> ^ i just made that 'pre release' 20:10:00 <abadger1999> packagers will just ignore github tarball downloads if an alternative exists. 20:10:12 <abadger1999> Since we want pip to work, pypi will always be that alternative. 20:10:46 <bcoca> so this was not about the tarball but about pypi specifically? 20:10:48 <abadger1999> So github tarballs will neither help nor hinder packagers... they'll simply ignore them. 20:11:06 <bcoca> pypi is not the best dealing with 'non python files' 20:11:31 <bcoca> dont they also automatically execute tests if present? 20:11:31 <abadger1999> bcoca: we should not be pushing a different tarball to pypi and to releases.ansible.com... that makes people very nervous when sigs and checksums don't match. 20:11:48 <abadger1999> bcoca: pypi? no. distros? no. 20:12:00 <bcoca> distros do, at least debian 'used to' 20:12:06 <abadger1999> bcoca: not "automatically" 20:12:07 <bcoca> i know gentoo also did 20:12:17 <abadger1999> "By policy" maybe but not automatically. 20:12:21 <Zhenech> bcoca, debian would if "python setup.py test" would work 20:12:27 <bcoca> it was part of teh process, upload deb-src and push through automated build and testing 20:12:36 <Zhenech> but your tests need "make tests" anyways 20:13:12 <bcoca> that would be if anislbe was a 'python library' iirc (been a long time) but not jsut for any application 20:13:27 <bcoca> ansible is an application, not a library 20:14:15 <Zhenech> it uses setuptools, thats all debian cares about :) 20:14:19 <abadger1999> bcoca: for packaging purposes, it's both. 20:14:20 <Zhenech> (debian tooling) 20:14:43 <abadger1999> ships a module in site-packages therefore it's a library. ships stuff in /usr/bin/ therefor it's an application 20:14:52 <bcoca> no, it is an app that has libraries 20:15:00 <bcoca> ^most apps do 20:15:02 <abadger1999> bcoca: not for packaging purpoes. 20:15:08 <abadger1999> there's no differentiator there. 20:15:15 <abadger1999> Only humans know the difference 20:15:38 <bcoca> well, humans etup the machines that do the testing, so depends on teh classification and detection that happens in those proceses 20:16:15 <bcoca> you have a specific one in mind, im' thinking about 3 ones i was fammiliar with debian/gentoo/freebsd .. they will run make tests on apps by default if makefile is present 20:16:26 <bcoca> or tests are present in any way they can recognize 20:17:20 <Zhenech> debian will detect which build system to use (setuptools) and then knows that for tests it has to call python setup.py test, that is a noop 20:17:27 <Zhenech> so debian tooling will cary on 20:17:35 <Zhenech> unless told otherwise in the packaging by the packager 20:17:45 <bcoca> IF your package installs ansible LIKE a python lib, via pip/setuptools, yes most will treat is as python libs, not all packagers do this, some will use tarball and makefile (i know i've created packages for all3 and have done it both ways) 20:18:48 <Zhenech> yeah 20:18:50 <bcoca> so a lot of this depends on what many 3rd parties are doing (i dont doubt most will use pip/ptyhon tools instead of make, but i also know others wont) 20:19:04 <Zhenech> but the currently shipped makefile has a tests target, and it will fail as the tests are not there 20:19:30 <bcoca> do we ship the makefile in tarball? 20:19:34 <Zhenech> yes 20:19:45 <abadger1999> So anyhow.. Vote count so far is alikins, abadger1999 +1, bcoca, -1, jimi-c +1 to docs, has yet to be official on tests although he was leaning against it. 20:20:10 <bcoca> im not -1, just saying there are more options that require less effort 20:20:13 <jimi|ansible> still -1 on tests in general 20:20:30 <abadger1999> bcoca: There are more options... but they all require more effort 20:20:43 <abadger1999> bcoca: this one is simply make a few changes in the MANIFEST.in and we're done. 20:20:43 <bcoca> 0 effort on our side to point people at github tarballs 20:20:58 <abadger1999> bcoca: 0 effort for something that won't get used. 20:21:01 <bcoca> IF they want docs + tests 20:21:04 <jimi|ansible> well, really i'm against different tarballs, if it's all in one and done with sdist i'm indifferent 20:21:08 <bcoca> that is on them 20:21:34 <Zhenech> jimi|ansible, right, different "official" tarballs is bad 20:21:37 <abadger1999> and then if we start shipping build resources in the tarball the github tarballs will be a bad thing to show people. 20:21:49 <jimi|ansible> different tarballs anywhere, i'd want the same on pypi, github, and releases 20:22:00 <abadger1999> (most likely there would be shipping built docs instead of requiring people to regenerate them themselves) 20:22:13 <bcoca> so the github are already different and automated, only way we can make them same is overriding by publishing our own to github 20:22:24 <bcoca> ^ existing problem 20:22:28 <jimi|ansible> github is a different matter, we don't control that 20:22:37 <abadger1999> jimi|ansible: I'm also -1 to different tarballs. I'd do everything with sdist. 20:22:39 <bcoca> we do, but only by pushing our own 'on top' 20:23:15 <abadger1999> jimi|ansible: which Zhenech's PR does: https://github.com/ansible/ansible/pull/20004/files 20:24:13 <bcoca> https://github.com/ansible/ansible/releases/tag/v2.2.1.0-0.3.rc3 <= github release with OUR tarball and sha (sadly we cannot remove the 'source ones') 20:24:38 <bcoca> ^ i'll remove, just wanted to show that we CAN use this if we want to 20:28:43 <abadger1999> We're almost to 1.5 hours... what do we want to do here? 20:29:07 <abadger1999> Lots of committers haven't voted. 20:29:20 <Zhenech> hit merge *cough* 20:29:22 <abadger1999> Looks like +3 -1 for docs, +2.5 -1 for tests 20:29:47 * allanice001 adds +1 docs, -1 tests fwiw 20:29:48 <bcoca> if we already afiling tests, i really have little to oppose, most packagers are ignoring make tests right now it seems 20:29:56 <alikins> any distro package maintainers here? (abadger1999?) 20:30:04 <abadger1999> yep. 20:30:08 <abadger1999> I'm fedora and epel. 20:30:09 * Zhenech partially for debian 20:30:21 <Zhenech> hlieberman is the main guy, he's not here 20:30:22 * ryansb RHEL 20:30:42 <abadger1999> I want tests -- I already have to use a script to download and include those as a separate tarball. 20:31:27 <abadger1999> nirik: main guy for fedora epel. Do you have an opinion about including tests in the tarball? 20:31:57 <bcoca> https://github.com/ansible/ansible/releases 20:32:38 <alikins> any official tooling building straight from git? 20:32:55 <bcoca> ^ im actually liking this, but i will leave it up to release manager, my only objections are now mute 20:33:49 <Zhenech> gentoo actually uses the github generated tarballs 20:33:50 <Zhenech> odd 20:33:52 <bcoca> gentoo package builds from pypi (looking at it now) 20:33:56 <Zhenech> https://gitweb.gentoo.org/repo/gentoo.git/tree/app-admin/ansible/ansible-2.2.1.0_rc1.ebuild 20:34:02 <bcoca> SRC_URI="https://pypi.python.org/packages/${BLAKE2B_HASH}/${P}.tar.gz" 20:34:17 <Zhenech> ah, but the rc is from github 20:34:28 <Zhenech> you don't push rcs to pypi? 20:34:33 <bcoca> we dont push rc to pypi 20:34:37 <Zhenech> ok, fair enough 20:34:39 <bcoca> we didnt know we could 20:34:50 <bcoca> weve had many issues with pypi numbering .... 20:35:01 <Zhenech> and there is also the version 9999 one with straight from git: https://gitweb.gentoo.org/repo/gentoo.git/tree/app-admin/ansible/ansible-9999.ebuild 20:35:10 <mattclay> I can go either way on including tests. If we do include them, I think we should be clear that we're running a subset of tests. We should also make sure that 'make tests' actually passes without having to install a lot of extra dependencies just for tests. 20:35:16 <bcoca> ^ i posted that form 2.2.0.0 stable 20:36:28 <abadger1999> mattclay: <nod> That would work for me. Although with my packager hat on, I often have to do the work of subsetting and determining which dependencies I care about and which I don't. 20:36:49 <bcoca> mattclay: look at gentoo ebuild, i believe theyhave most deps 20:36:55 <bcoca> line 34+ 20:37:00 <abadger1999> Which I suppsoe just means, hey, ansible is doing a better job of preparing its tets for packagers than other projects ;-) 20:37:02 <alikins> I know lots and lots of official packaging does so to me it makes sense that building from a git checkout and a release tarball should be the same. Assuming user local installs also work from that same tarball, if pypi needs a special case I'm okay with it using a sdist tarball. But it would be even better if we could push the general tarball to pypi as well. 20:37:14 <mattclay> abadger1999: Since you've got packaging experience, it sounds like you and I should talk separately about what we can do to make tests work better for packaging purposes. 20:37:36 <abadger1999> mattclay: sounds good. and Zhenech since he grabbed the bull by the horns here ;-) 20:37:40 <bcoca> mattclay: 99% of that is specifying deps for tests 20:38:35 <bcoca> he, funny both ebuilds have the test requirements, but only the RC ones actually get the tests 20:40:36 <abadger1999> jimi|ansible: so... I'm thinking it looks like I could hit merge on Zhenech's PR and people would go along with it at this point.. Do you want to say something to stop me from doing that? ;-) 20:40:54 <bcoca> his -1 on tests? 20:41:43 <jimi|ansible> i don't think raw docs will do people much good, i'd much rather see ansible-doc capabilities expanded to include the play/inventory/etc. docs instead 20:42:07 <bcoca> they are rst and require building, i dont see them that useful either 20:42:09 <abadger1999> bcoca: he last said: <jimi|ansible> well, really i'm against different tarballs, if it's all in one and done with sdist i'm indifferent 20:42:10 <jimi|ansible> but it's not something we can't fix later 20:42:29 <abadger1999> yeah... raw docs will help the offline usage case... it's just not ideal. 20:42:56 <bcoca> need to add plugins to ansible-doc .. do we really want 'play topics'? 20:43:59 * bcoca removes the github prerelease to avoid confusion 20:44:56 <abadger1999> okay 20:45:08 <abadger1999> #action abadger1999 to merge the PR. 20:45:09 <Zhenech> hmm, actually building the docs during packaging would also require to add the hacking folder (at least partially) 20:45:11 <bcoca> hmm, does nto allow me to undo all ... leaving description 20:45:27 <bcoca> ah, you want html docs, not the rst? 20:45:31 <abadger1999> #action mattclay abadger1999 and Zhenech to talk about making the tests good for the packager use case 20:45:43 <bcoca> that adds 40-60mins to release .... 20:45:44 <Zhenech> bcoca, html would be stept 2 I'd say 20:45:55 <Zhenech> rst is good enough for starters 20:47:15 <abadger1999> #action will look at integrating the docs with ansible-doc in the future for a better offline experience 20:47:54 <abadger1999> bcoca: It could also be built by downstreams instead of as part of the release process.... that's something we can look at later too. 20:48:10 <abadger1999> #topic Open floor 20:48:28 <abadger1999> Anyone have something to say in our remaining 11 minutes before the next meeting? 20:48:32 * bcoca feels for packager that tries to shove in sphinx as build requirement .... 20:48:40 <allanice001> yup 20:48:44 <allanice001> 5 minutes pls 20:48:53 <abadger1999> allanice001: Go for it 20:49:08 <allanice001> AnsiBot (Doc Bot) is running in #ansible-devel 20:49:11 <abadger1999> bcoca: Many fedora/rhel packages do exactly that. 20:49:21 <Zhenech> bcoca, a lot of packages in debian has sphinx as build-dep 20:49:23 <allanice001> Has 3 commands, !help; !search; !about 20:49:23 <abadger1999> #topic AnsiBot (Doc Bot) is running in #ansible-devel 20:49:25 <Zhenech> noone cried so far 20:49:32 <Zhenech> but EOD now :) 20:49:49 <abadger1999> #info AnsiBot has 3 commands !help !search !about 20:49:59 <allanice001> Please use / test it, and let me know if / when you want it in #ansible 20:50:11 <abadger1999> Cool. 20:50:26 <allanice001> gundalow suggested we change it's name, not to get confused with ansibullbot 20:50:28 <bcoca> packagers have always had a high tolerance for pain ... or they dont remain packagers long 20:50:43 <bcoca> nagbot! 20:50:47 <bcoca> nagmenot 20:50:50 <Zhenech> my debian clock is at 10 years now, nuff said :) 20:51:18 <allanice001> add comments / queries and criticism to https://github.com/ansible/proposals/issues/48 20:51:20 <bcoca> i didnt even get to become official, after 2yrs helping other packagers i decided to move on .. then gentoo .... 20:51:59 <allanice001> source is available on https://github.com/infrascloudy/AnsiBot 20:52:14 <abadger1999> #link https://github.com/ansible/proposals/issues/48 20:52:20 <abadger1999> #link https://github.com/infrascloudy/AnsiBot 20:52:29 <allanice001> thats it from me 20:52:38 <abadger1999> allanice001: Very nice work. I saw lots of testing yesterday ;-) 20:52:51 <allanice001> Yeah, some fine tuning 20:53:16 <allanice001> jtanner had some suggestions whi 20:53:22 <allanice001> Which I changed 20:54:09 <allanice001> But all in all, is small easily managed 20:54:09 <abadger1999> Cool. 20:54:27 <abadger1999> #topic Open floor 20:54:30 <abadger1999> Anyone else? 20:54:36 <abadger1999> If not I'll close in 60s 20:55:43 <abadger1999> #endmeeting