Meeting on irc.gimp.org:#build-brigade Meeting started November 20 2006 16:00 UTC Meeting finished November 20 2006 17:00 UTC In attendence: ------------------------------ Thomas Vander Stichele (thomasvs) Fernando Herrera (fer) Juan José Sánchez Penas (juanjo) José Dapena Paz (dape) Iago Toral Quiroga (iago) Meeting log ----------------------------- nov 20 17:00:40 hi all nov 20 17:00:54 hi nov 20 17:00:59 it is time to start the meeting :) nov 20 17:01:18 hi nov 20 17:01:23 I'll upload the minutes nov 20 17:01:45 to live.gnome.org in cade someone is missing nov 20 17:01:49 and for future reference nov 20 17:02:05 yes nov 20 17:02:08 first thing nov 20 17:02:16 I have to leave in 25-30 minutes nov 20 17:02:29 dape: ok nov 20 17:02:37 then I guess it is better you start nov 20 17:02:39 so I would beg we start with jhbuild+buildbot related things nov 20 17:03:04 do you know if F.Peters is coming? nov 20 17:03:24 no news about that :S nov 20 17:04:01 as some of you know, frederic doesn't agree with using buildbot nov 20 17:04:43 but no news appart from that nov 20 17:04:58 oki nov 20 17:05:31 dape: then let's start with buildbot issues nov 20 17:05:55 can you talk a little about its current state? nov 20 17:05:58 ok nov 20 17:06:01 so people can know... nov 20 17:06:03 yes nov 20 17:06:15 I've integrated jhbuild (with some custom patches) with buildbot nov 20 17:06:35 and now I've got a buildbot that configures itself from jhbuild modulesets nov 20 17:07:00 and made some changes to prevent buildbot getting all resources because of starting one server per gnome module nov 20 17:07:27 and I would like to deploy this with gnome 2.16, and, if possible, gnome 2.17 nov 20 17:07:52 thomasvs, what about the foundation machines? nov 20 17:09:03 in the email I sent to the list, there was this URL, with a little demo of that buildbot running to build gtk+ nov 20 17:09:06 http://gtktests-buildbot.igalia.com/ nov 20 17:09:27 oh ok nov 20 17:09:40 that's because the machine is running for other projects nov 20 17:09:52 and I considered I shouldn't run all gnome moduleset nov 20 17:10:24 dape, did you also extend jhbuild to be able to run tests? nov 20 17:10:29 anyway, I think Igalia has offered another machine for this, and I can try with it in case we can't get the foundation machines nov 20 17:10:40 fer, yes nov 20 17:10:50 cool nov 20 17:10:54 in fact I implemented a simple patch to add a "pathrun" command nov 20 17:11:03 that enables me to run an arbitrary command in a project path nov 20 17:11:15 then I can run standard make check or whatever I need to do nov 20 17:11:24 this gives me more flexibility nov 20 17:11:43 great, have you seen Xan's graphics of cairo performance over the gtk-theme-torturer? nov 20 17:11:59 http://blogs.gnome.org/xan nov 20 17:12:24 yes nov 20 17:12:28 we could hook that in some way that we can get an xml-ouput of every continous build nov 20 17:12:41 and then plot grapchics comparing arbitrary builds nov 20 17:12:54 any output, with my patch we can decide any bash/scripting magic we want nov 20 17:13:06 in fact, the code coverage reports are generated this way nov 20 17:13:27 great nov 20 17:13:50 we should add the code to generate the reports we want in the scripts at the end of compilation nov 20 17:13:55 currently there are two scripts nov 20 17:14:09 one is make-check.sh, that runs the tests, and clean coverage info nov 20 17:14:34 and other is module-reports.sh, that currently simply generate coverage info from tests logs nov 20 17:14:42 but this script could implement more stuff nov 20 17:15:45 there are a lot of things that can be done, and probably more after we talk to maintainers and they give us feedback nov 20 17:15:54 yes nov 20 17:15:58 how do you integrate that output with the html output? nov 20 17:16:06 but we would need some developer work on buildbot to achieve them nov 20 17:16:29 any report tool should generate html, but they should be case per case nov 20 17:16:51 for example, we can get projects with check getting specific html output, performance tools getting their one, etc nov 20 17:17:07 I wouldn't add more than linking this info in a structured way in a first stage nov 20 17:17:27 then we can begin to think in cumulative information, in order to get stats and time charts nov 20 17:17:48 but first, getting the output is a good task itself nov 20 17:17:55 dape nov 20 17:18:02 can you point out nov 20 17:18:11 some important nov 20 17:18:20 short-term issues nov 20 17:18:26 regarding buildbot nov 20 17:18:27 ? nov 20 17:18:38 I mean, tasks that need to be done in buildbot nov 20 17:18:41 oh ok nov 20 17:18:56 --- dmalcolm_away is now known as dmalcolm nov 20 17:18:56 I think that short-mid term tasks should be nov 20 17:19:24 try to make patchs in jhbuild for buildbot go upstream nov 20 17:19:35 as many as possible, in order to make maintenance easier nov 20 17:20:01 of course, deploy the buildbot with all the current gnome moduleset nov 20 17:20:12 mmm... nov 20 17:20:16 2.16 at least, and 2.17 also if we can get hardware for this nov 20 17:20:40 we would need a machine for that nov 20 17:20:40 integrate test results for some projects (I think we can try with the tests you get in gtk) nov 20 17:21:15 and cosmetic tasks: better css's in buildbot, maybe with gnome web look and feel nov 20 17:21:22 wedape: yes nov 20 17:21:34 we need to fix the view nov 20 17:21:39 and the other cosmetic task, adding the new rss feed support that's been sent to buildbot mailing list nov 20 17:21:42 so we need a volunteer for that :) nov 20 17:21:53 I guess thomas already has some expirience with that nov 20 17:21:58 or are you guys good with rss? :) nov 20 17:22:05 css I mean nov 20 17:22:29 mmm I have some ideas, but I'm far from being an expert O:) nov 20 17:22:32 maybe we can talk to the gnomeweb people for those design tasks nov 20 17:22:51 yes, and they're preparing the new web design nov 20 17:23:05 maybe there can be some ideas on what should we do to make the view kick the ass nov 20 17:23:30 I would like myself to collaborate in the deployment of the current work nov 20 17:24:14 and I would pray for some help to get jhbuild patches upstream nov 20 17:24:41 the ones I implemented need more work to avoid james henstridge get frightened nov 20 17:26:55 I think making the deployment is important nov 20 17:27:00 as soon as we do it nov 20 17:27:07 as sooner we get feedback nov 20 17:27:54 back, sorry nov 20 17:28:40 thomasvs, do you have expirience with the view of buildbot? nov 20 17:28:54 I mean CSS and so nov 20 17:29:09 I have to leave now nov 20 17:29:24 bye dape! nov 20 17:29:32 iago, I think we can use that new development machine here at igalia for deployment, I could deploy the buildbot scripts there nov 20 17:29:38 iago: a little, but my experience is from older buildbot code nov 20 17:29:47 dape: I could use some info on what exactly I should deploy on these build machines here nov 20 17:29:59 dape: if you have an easy recipe for me I can get those machines started nov 20 17:30:13 oh ok nov 20 17:30:14 mmm nov 20 17:30:29 then we can try to get this working nov 20 17:30:41 :( I have to leave nov 20 17:30:59 dape: fine, just send me a mail about what I can do and I will make some time for it nov 20 17:31:06 ok, iago, thomasvs, I can help thomasvs deploying the buildbot stuff in the foundation machines nov 20 17:31:24 dape: ok, that's fine nov 20 17:31:25 take this into account when you make the meeting conclusions O:) nov 20 17:31:42 dape: sure, I won't forget it ;) nov 20 17:32:33 ok, then I'll go nov 20 17:33:10 thomasvs, regarding the view, the buildbot demo that we set up is quite ugly nov 20 17:33:24 you can take a look at it here: nov 20 17:33:26 http://gtktests-buildbot.igalia.com/ nov 20 17:33:58 this is a different view from the waterfall view ? nov 20 17:34:19 this is our waterfall for example: http://build.fluendo.com:8080/ nov 20 17:34:49 thomasvs, if you click in any module you get the waterfall for that project nov 20 17:34:50 the one dape set up is with each project as its own master inside one common process right ? nov 20 17:34:55 yes nov 20 17:35:08 (still leaving O:) nov 20 17:35:20 heh nov 20 17:35:29 iago: well, someone needs to decide what would be a "nice" view nov 20 17:35:33 I am no HTML wizard nov 20 17:35:52 there are some ideas that would be simple to do nov 20 17:35:59 like, "idle" should go under the builder name nov 20 17:36:07 and the table should be outlined correctly for it nov 20 17:36:16 since this is a view that I assume dape created, it shouldn't be hard to make it look better nov 20 17:36:16 <-- dape has quit (arg late!!!) nov 20 17:37:03 thomasvs: yes nov 20 17:37:21 I will talk to dape nov 20 17:37:38 to know the details of the current view nov 20 17:37:58 and I send the conlusions to the list nov 20 17:38:13 so we can talk about it along with him nov 20 17:38:29 cause I don't know what he did to get that view nov 20 17:39:20 I think he was talking about a css nov 20 17:39:43 but I'm not sure if that's enough with the current design of the view nov 20 17:40:38 no, it probably needs some more splitting into the table nov 20 17:40:41 and after that a css nov 20 17:41:31 ok nov 20 17:41:41 there is another issue nov 20 17:42:10 which is the integration of the tests nov 20 17:42:28 currently nov 20 17:42:44 there is only a stdout flush nov 20 17:42:48 http://gtktests-buildbot.igalia.com/gtk%2B/gtk%2B-sarge/builds/92/step-gtk%2B%20make-check.sh/0 nov 20 17:43:18 what should I see ? nov 20 17:43:49 in that url? you should see the result of a "make check" nov 20 17:44:03 right, but what shows the problem you talk about ? nov 20 17:44:06 ah nov 20 17:44:15 I think it would be nice nov 20 17:44:24 to know in the waterfall view nov 20 17:44:31 if there are failed tests for a module nov 20 17:44:42 right nov 20 17:44:42 maybe the number of passed/failed tests nov 20 17:44:45 I think we do that for gst nov 20 17:44:52 and probably a better view of all that mess nov 20 17:44:57 ah, interesting nov 20 17:45:00 it's a matter of writing some code for the specific step that checks the output of the tests nov 20 17:45:09 yep, the problem nov 20 17:45:13 in the gnome case nov 20 17:45:17 is that probably nov 20 17:45:28 people use diferent tools for their unit tests nov 20 17:45:32 right nov 20 17:45:44 so the step would need to know how to do this. but we can push people in the direction of using check nov 20 17:45:57 that's one of the reasons why we chose to standardize on check for gst - so that we could use the same way of parsing the info at the end nov 20 17:46:17 I can't remember if I ended up writing the buildbot step to look at it, and I do not have one deployed that has it it seems nov 20 17:47:09 I did it for flumotion and coverage nov 20 17:47:11 http://build.fluendo.com:8071/ nov 20 17:47:19 <-- the orange blocks (some of them at least) show coverage status nov 20 17:47:48 and it parses the output to find the percentage and displays it as part of the waterfall. we can do this too for gnome, for failed tests and coverage nov 20 17:47:55 I see nov 20 17:48:09 the problem is that some people will not use check :/ nov 20 17:48:19 currently I'm trying to convince people in gtk+ to use it nov 20 17:48:34 but seems they prefer doing something from scratch nov 20 17:48:41 to avoid a new dependency nov 20 17:48:42 which I think is silly, but hey nov 20 17:48:53 it's a devel-time dependency, so it shouldn't be an issue nov 20 17:48:56 :) nov 20 17:49:11 but my plan of attack was to improve the gstreamer use of it and add these things to our buildbots and use the output from it to convince other people nov 20 17:49:16 just haven't had much time to do that nov 20 17:49:24 I want to show graphs over time of coverage data and test result data nov 20 17:49:27 all from our buildbot nov 20 17:49:45 yes, we also added coverage to our demo: nov 20 17:49:45 but it takes work :/ nov 20 17:49:59 thomasvs, it would be nice to get your opinion in the gtk-list thread, cause Tim Janick is proposing to do something from scratch and it is difficult to go on with check if some of the maintainers are not convinced nov 20 17:50:06 http://gtktests-buildbot.igalia.com/gnomeslave/gtk+/lcov/gtk/index.html nov 20 17:50:13 juanjo: hm, I'm not on that list nov 20 17:50:38 iago: yeah, I have something like that for gst, I think dape used my template to get to that point nov 20 17:50:51 iago: I still need to integrate it to make sure it gets run on every build and then uploaded somewhere though nov 20 17:50:54 thomasvs, we can CC you if you want to get into the discussion anyway ;-) nov 20 17:51:04 juanjo: yes, sure nov 20 17:51:12 or subscribe me to the list :) nov 20 17:52:12 thomasvs: I guess we can make the tests run under Check look nice under buildbot nov 20 17:52:31 that would be good point to push maintainers to use check nov 20 17:52:35 iago: yes, exactly nov 20 17:54:09 ok, so I think we talked about all nov 20 17:54:12 summarizing nov 20 17:54:26 dape and you will work on the buildbot deployment nov 20 17:55:14 I can try to play a little bit with the view nov 20 17:55:45 and meanwhile I'll try to keep unit tests for gtk+ rolling nov 20 17:56:19 thomasvs: btw, I think that your expirience could be valuable in the gtk-devel debate on unit tests nov 20 17:56:53 I'll send you an email nov 20 17:57:04 pointing to the thread in case you cant to participate nov 20 17:57:11 ok nov 20 17:58:10 once we have a cool demo working, we could focus in other things like getting the collaboration of maintainers, doing some more marketing about the build brigade work (gnomejournal,etc.), and all the social side of the work... but for now, I think the main goal is to get in a couple of months a cool demo of what can be done nov 20 17:59:01 juanjo: yes, definitely nov 20 17:59:49 ok, so I guess this is all, not bad for a first meeting :) nov 20 18:00:16 I'll upload the minutes to l.g.o and send an email to the list to keep all people informed nov 20 18:00:17 should we schedule a meeting every month for tracking the work? nov 20 18:00:23 is this day and hour ok for everyone? nov 20 18:00:54 <-- Casanova has quit (http://prashblog.com -- One day I am going to find this `peer' and reset his connection) nov 20 18:01:23 juanjo: it is ok for me and dape nov 20 18:01:28 I think it would be nice nov 20 18:02:44 btw, i will talk with josh about handling the channel with burrito, in order to be able to keep a topic and gather people interested here :) nov 20 18:03:14 juanjo: yes, good point :) nov 20 18:05:54 ok, it seems the meeting is over :) I think we can have the next one by mid-december to see how things are going nov 20 18:05:57 juanjo: mondays in the middle of a working day is very hard for me nov 20 18:06:08 but I will adapt if that's what other people want nov 20 18:07:12 thomasvs, would it be better for you to do it later/earlier? nov 20 18:07:55 juanjo: yeah, I'd prefer outside of work hours. like for example 20.00 GMT, or 19.00 spanish time nov 20 18:08:52 that time would be ok for me nov 20 18:09:08 I can talk to dape and check if it ok for him too nov 20 18:09:18 that would be ok for me too nov 20 18:11:01 ok guys, need to leave now, see you later! nov 20 18:11:05 <-- juanjo has quit (Leaving) nov 20 18:11:20 later!