21:00:33 #startmeeting Weekly Fedora ARM status meeting 21:00:33 Meeting started Wed Dec 12 21:00:33 2012 UTC. The chair is jonmasters. Information about MeetBot at http://wiki.debian.org/MeetBot. 21:00:33 Useful Commands: #action #agreed #halp #info #idea #link #topic. 21:00:41 .fas jonmasters 21:00:41 jonmasters: jcm 'Jon Masters' 21:00:42 .fas jcapik 21:00:44 jcapik: jcapik 'Jaromír Cápík' 21:01:00 .fas parasense 21:01:01 masta: parasense 'Jon Disnard' 21:01:12 * masta waves 21:01:18 we'll give it a couple of minutes 21:01:50 .fas chris@tylers.info 21:01:50 ctyler: ctyler 'Chris Tyler' 21:02:30 .fas dmarlin 21:02:30 dmarlin: dmarlin 'David A. Marlin' 21:02:45 .fas blc@ 21:02:46 bconoboy: blc '' 21:02:48 ok everyone let's get going then 21:03:08 #topic 0. Additional agenda items 21:03:28 Is everyone ok with the current agenda for today? Anything you want to add? (see arm@lists.fedoraproject.org mail) 21:03:47 jonmasters: Let's see where we get to in the time we have 21:03:53 ok 21:04:03 #topic 1. Current Problem packages 21:04:23 So, on texlive, we have an import now (thanks pbrobinson, dgilmore, ctyler) 21:04:33 however, repo generation is broken and I am looking at it 21:04:40 I am also going to poke at mongodb and condor 21:04:45 what else is on the hotlist? 21:05:28 #chair jonmasters bconoboy ctyler dgilmore dmarlin Frojoe masta pwhalen 21:05:28 Current chairs: Frojoe bconoboy ctyler dgilmore dmarlin jonmasters masta pwhalen 21:05:47 Those are the painpoints. 21:05:49 eclipse 21:05:53 it's waiting on createrepo though. 21:05:57 * nirik is happy to provide any info about repogen from primary if it helps any 21:06:37 nirik: we may need that. Right now, it appears as if there are a couple of genpkgmetadata.py processes stuck 21:06:54 nirik: the box in question does have a RAID problem, but I don't think it's that as ctyler was saying this happens on other hosts 21:07:10 * jonmasters has the offending processes open under gdb and has started gathering data 21:07:11 yeah, all I can think of is the libxml issue, but that was a crash on primary 21:07:27 jonmasters, I think the raid problem has to do with a bad sata cable/port 21:07:40 ctyler: can you confirm that you have this also blocking in the same way on another host that does not have hardware problems? 21:07:40 It's a red herring. 21:07:51 Same issue on all the hosts. 21:08:14 ok, good. Like I said, it didn't seem to be that. Just to be sure, I had a process open every package and read it succesfully on the offending host 21:08:51 so I'll look at that. I will be heading out this evening though, so is it ok if those processes are left running until I get back later in the night? 21:09:17 jonmasters: as long as no one cancels or frees task, they'll be there 21:09:24 ok 21:09:36 I'll poke it some more before tomorrow 21:10:16 ok, so other issues with problem packages? 21:10:34 We're basically at a standstill until we get createrepo sorted. 21:10:47 this I know 21:10:57 jonmasters: I think that's it for F18 beta in any case. 21:10:58 it's why I've dropped everything else :) 21:11:07 however, do we know about any other problem packages? 21:12:02 ok then, moving on 21:12:20 #topic 2). F18 Beta status 21:12:41 The plan is to just put something out and poke at it 21:12:45 what's preventing that? 21:13:06 it's a beta, it's not going to be perfect, but surely there's nothing really stopping us now? 21:13:10 I have made a new set of images, and bconoboy is going to post them to scotland 21:13:31 jonmasters: depends on the definition of 'beta', I suppose! 21:13:33 anyone can then test those images and report issue 21:13:33 ok, are we going to call those almost-beta and do a quick test VFAD while we wait for Dennis to get back? 21:13:49 or just post these as beta? I don't mind either way 21:13:54 f18 beta test candidates 21:13:55 I think we've been making good progress with the vfads 21:13:59 dmarlin: +1 21:14:09 ok, "test candidates" it is 21:14:15 beta-tcX 21:14:16 so we have a VFAD tomorrow or Friday? 21:14:18 I think these address the issues we saw with TC1 21:14:44 are packages getting signed+pushed while dgilmore is out? 21:14:49 I'm going to suggest Friday on the assumption that Paul will be back with us tomorrow or thereabouts and that Friday would therefore give enough time for him to get caught up first 21:15:16 agreed... the main issue last time was firstboot missing, selinux messign with NetworkManager, and ??? 21:15:18 ctyler: I've seen some unsigned packages, but nothing has prevented image creation (yet) 21:15:29 anyone got suggestions on Friday? I'm suggesting just a come-whenever all day thing same as before 21:15:57 * masta is on vacation until next year, has plenty of time to poke test candidates 21:16:09 I'm wondering if the vfad and 1.7.1 upgrades can happen in parallel on Friday 21:16:15 masta: poke early, poke often. :) 21:16:33 * pbrobinson is vaguely here 21:16:50 well hello vague pbrobinson 21:17:59 jonmasters: friday is fine 21:18:01 ok so VFAD on Friday for testing, anything else? 21:18:04 ctyler it's 21:17 and I've just got off a board call and dinner is ready.... I'm been smelling my chilli con carne cook for like 4 hours and I'm hank marvin! 21:18:23 alrighty then 21:18:39 #topic 3). Ownership of non-release blocking images (BeagleBoard-xM, etc.) 21:18:51 I think bconoboy has a good plan on this one already so no need really to discuss 21:19:03 do people know the plan? 21:19:03 jonmasters: why? 21:19:18 bconoboy: tell everyone the plan 21:19:28 jonmasters: non release blocking? 21:19:36 jonmasters: I still think we need people to 'own' images and be responsible for testing 21:19:58 jonmasters: just to make sure we have coverage 21:20:01 The plan: I'm going to start making nightly images using dmarlin's kickstarts 21:20:08 (people with the hardware and inclination) 21:20:14 jonmasters:I thought the agreement was that we drop v5 to community and platforms that have upstream issues (imx) 21:20:39 community will still need to test the images 21:20:46 dmarlin: we need public upstream kickstarts for us to be able to do that 21:21:01 Automated image creation is cheap- testing is not. I'll make the images but rely on community to take them for a test drive 21:21:07 This includes kirkwood and beagleboard. 21:21:15 pbrobinson: as soon as beta is out, I'm posting the kickstarts used for all to see/use 21:21:26 and my understanding is that spins-kickstart is the upstream and I don't believe that is usable 21:21:49 pbrobinson: I'll be posting updates for spins-kickstart as well 21:22:02 arm-kickstarts pkg? 21:22:05 dmarlin: so in the interim your the owner of all the builds then, congrats I look forward to the beagle image for me as a community member to test 21:22:14 pbrobinson: There are some differences but hopefully we'll be able to converge on spins-kickstart 21:22:24 pbrobinson: nice try 21:22:41 pbrobinson: I'll be making the images, dmarlin is in the clear here :-) 21:22:54 bconoboy: but in there interim there's nothing at all that's publicly usable 21:23:12 (btw, looks like the beaglebone 'Sitara' chip is not happy with our kernel-omap) 21:23:13 bconoboy: thanks, I look forward to testing the beagle then 21:23:25 pbrobinson: my last revisions are all publicly available, and have been since we started this 21:24:15 pbrobinson: thanks for testing 21:24:28 dmarlin: OK, I've not found them and given the number of people that ask nor have others, can you (re)send the details of that to the list? 21:24:38 #action bconoboy to make nightly images including kirkwood and beagleboard-xm. community to test. 21:24:44 pbrobinson: sure, I'll send it again 21:24:55 great 21:25:38 jonmasters: next 21:25:43 ok 21:25:56 #topic 4) F18 and kernel VFADs planning 21:26:12 so VFAD on Friday for beta candidate testing 21:26:25 then I'm suggesting we find a time next week to scrub kernel configs collectively 21:26:42 when would work, and who should we invite to join? I'm thinking of asking jwb and others from the kernel team? 21:27:10 jonmasters: better break those down separately. First f18 vfad? 21:27:11 [aside: I realize I skipped the ARMv6 topic - we'll come back to that in a few minutes] 21:27:25 well, I think we agreed Friday for F18 VFAD just no 21:27:26 now 21:27:32 who is going to run it? 21:27:38 I am, but hopefully Paul 21:27:45 if he's back, he has that 21:27:50 #action jonmasters or pwhalen to run Friday VFAD 21:27:54 jonmasters: next week I'm in Cairo so not sure if that will work for me 21:27:59 Okay, next explain kernel vfad 21:28:21 ok, so kernel VFAD is to collectively scrub 3.7+ options and look for things that need changing, especially for multiplatform 21:29:14 It would be good to have pbrobinson but also we can still make good progress in any case. I'm concerned about having it any later as last week is the last week some folks are around 21:29:26 and I want to make sure the beta is out next week, and some progress happens on this one too 21:29:30 Yes, next good option is 1st week Jan 21:30:09 Doing some work sooner is good 21:30:16 jonmasters: it means you just need to take that into account 21:30:32 jonmasters: will discuss OOB 21:31:47 * ctyler looks at calendar for next week, looks like bad graffiti 21:31:49 #action jonmasters to schedule kernel vfad later. date TBD 21:32:07 Should we talk date while we've got a few heads together? 21:32:32 next Wed? 21:32:44 +1 21:32:52 +0.9 21:33:01 ok, tentatively next Wed then 21:33:03 +0.1 21:33:22 fractions eh 21:33:36 #action jonmasters to send email about kernel VFAD on Wed 2012-12-19 21:33:38 depends on who needs to be there 21:33:45 Wed is ok 21:33:48 right 21:33:54 hence tentative 21:34:07 I suggest I email kernel@ and see who can make Wed, or other days 21:34:09 jonmasters: do we know when the kernel folks will be available? 21:34:16 :) 21:34:40 also, I'd like Mark L. from Calxeda and anyone else who's actively poking to turn up if they can 21:35:00 ok, I'll get that mail thread going 21:35:15 let's move on 21:35:27 #topic 2.1) ARMv6 initial build feedback 21:35:35 ctyler: go! 21:35:43 * pbrobinson doesn't see the point in non active fedora kernel contributors being there for the sake of doing +1 on things they don't contribute to 21:36:13 So we have a test image of v6hl for the Pi 21:36:21 seems appreciably faster in the GUI 21:36:36 pbrobinson: the purpose of having a VFAD is to get input on things that we might want to know. So "knowledgable kernel folk" I think are the target 21:36:50 enough so that we're continuing with F18 building for a 1Q13 release. 21:37:44 ctyler: is the kernel and other firmware stuff identical between the two images? 21:38:01 jonmasters: agreed, that was my point 21:38:23 The arm-side talk-to-the-firmware libraries are hf vs sf but otherwise the same. 21:38:47 pbrobinson: yea, I agree with you. If we didn't have the holidays very shortly I'd say let's hold until you're around, but let's at least get Josh and Mark types to help ponder stuff if possible 21:39:11 F17 build is basically a build-previous for the F18 build. 21:39:24 jonmasters: I might be, I just don't know 21:40:25 ctyler: per bconoboy 's question.... ? 21:40:50 A: [16:38:23] The arm-side talk-to-the-firmware libraries are hf vs sf but otherwise the same. 21:40:55 ah yea sorry 21:41:18 ctyler: So we're definitely seeing the difference of the ABI and the v6 instructions at this point- no out-of-band differences in kernel versions, codecs, etc? 21:41:24 some of the chatter I saw suggested actually there wasn't much in it at all. Do you have numbers that convincingly show a benefit? 21:41:28 ya i did compile a v6hl kernel but it was older than the v5 one we had so i just used v5 21:41:34 bconoboy: yes 21:41:51 ! ! 21:41:57 a person ran gtkperf and it was like 45 seconds faster 21:42:02 i forget who on the mailing list 21:42:18 is there a particular optimization with from the glibc for armv6hl ? 21:43:15 there are a number of v6 optimizations that can be done, I'm not opposed to the idea of the numbers being faster, or people having subjectively higher experiences, but what I want is not a Phoronix test but a real set of benchmarks 21:43:33 Phoronix are the Ubuntu of benchmarks: "hey, this looks pretty, it must be better" 21:43:53 Jonmasters, you just went up a few ranks in my good books 21:43:53 I basically just copied v7hl's flags but changed them for v6hl 21:43:55 side by side, same system, same kernel if possible, just different userspace 21:44:15 jonmasters: which benchmarks? 21:44:18 jonmasters: it's appreciably faster, and this is basically what we need to do if we're going to be serious on the Pi. 21:44:20 well quite 21:44:21 i didnt add any specific v6 tunings that the raspian folks did 21:44:26 there's also run time v6 detection that can improve code times based on underlying feature detection.... it's not perfect but improving... just like what's seen on x86 21:44:45 yea pbrobinson is referring to HWCAP and similar 21:44:48 it's early days, benchmarks are in the works, but the results are encouraging 21:45:05 jonmasters: I refer to the aforementioned site as moronix 21:45:11 Gtk uses cairo and cairo use floating point for drawing 21:45:42 they were responsible for the corruption non issue of ext4 on recent kernels 21:45:49 ok, one thing I would like to know is what are the measurement criteria, what are the benchmarks that will be done (subjective GUI isn't good enough, too much room for unintentional bias), and what will be the threshold to e.g. move to a v6 minimum eventually 21:45:55 but others mentioned it was possible to have memset memcopy specificly rewritten for armv6hl, 21:45:59 so can we propose some benchmarks? I'm sure somebody here can script it 21:46:07 would be nice to have for other boards too 21:46:07 and cairo uses pixman which has lots of v6 run time opt 21:46:27 pbrobinson: yea - to me Phoronix are a net harm to the the Linux industry, but if I start making that list....it's going to be long 21:47:01 jonmasters: yea, we have enough of your rants about other things anyway.... 21:47:05 :) 21:47:11 ctyler: Do you have further benchmark plans? 21:47:20 yea, can we get the plan? 21:47:26 For the sake of the topic I'd just like to record what people are planning on doing. 21:47:38 masta: do we wanna Gtk benchmarks? 21:47:41 well im unfamiliar to this area of benchmarking 21:47:48 what are some reputable tools to do this? 21:48:00 Just a minute, folks. 21:48:03 queing ctyler 21:48:31 We can argue all day about what value of N% better we need to make armv6hl really essential. 21:48:40 my thoughts are motivated around support burden for armv6 so I'd want to know things like which benchmarks and what thresholds would indicate armv6 is the way of the future 21:48:43 jcapik: I'm indifferent, but sure... I always like to run openssl speed 21:48:54 But it doesn't matter. If we're going to be serious about the Pi, we need armv6hl. 21:49:13 masta: I can create a testing Gtk app 21:49:13 We can do armv6hl and be viable to those users, or we can forget the Pi. 21:50:34 It's a one-board platform, so it has a high cost/benefit ratio, but it's also a very high-mindshare platform. 21:51:36 So benchmarking is on the to-do list, but it's not the gating factor here. 21:51:36 yep, happy to get as many percentage points as possible 21:52:10 if the GUI response is visibly faster, then we should go that way .... RPi is quite popular 21:53:04 The Foundation is pumped about getting a strong Fedora on the Pi, and I think it's a win for Fedora. 21:53:06 and we want to be popular too, don't we? 21:53:08 I don't think we're voting on the future of rpi, I'd just like to have some concrete numbers to demonstrate the value of v6 21:54:21 We'll get some preliminary benchmarks next week (remember that the F17 build is build-previous quality) and some better ones in January. 21:54:41 ctyler: great! do you know what benchmarks you'll be running? 21:55:03 Also, are there any other v6 platforms that might benefit from the work? 21:55:41 We'll run whatever we can find in Fedora. 21:55:55 okay 21:56:12 In terms of other v6 platforms, the only one I'm aware of is the APC, and it's a pretty bad board. 21:56:16 * nirik notes phoronix-test-suite is in fedora. ;) 21:56:22 * nirik runs from jonmasters 21:56:22 #action ctyler to provide preliminiary benchmarks showing differences between v5tel and v6hl builds on rpi next week using valid benchmarks in the fedora package set 21:56:46 * ctyler notes valid vs. phoronix-test-suite :-) 21:57:29 even the stuff in phoronix can be useful, you just have to be clear on what you're actually demonstrating. 21:58:44 In terms of the burden for building v6, Seneca will carry it if fedora-arm does not. 21:58:50 nirik: :) 21:59:00 ctyler: Good topic for fudcon I think. 21:59:07 nirik: best thing I've heard all day. Hilarious! Thanks :) 21:59:11 jonmasters: next? 21:59:13 ok so next topic? 21:59:22 please 21:59:25 #topic 5) Your topic here 21:59:33 fudcon! Who's going? 21:59:34 just one minute shy of the hour...anything else? 21:59:39 * jonmasters is going 21:59:42 * bconoboy is going 21:59:47 * ctyler is going 21:59:47 * pbrobinson is gone 22:00:06 * masta is going 22:00:32 lol 22:01:02 #info FUDCON: jonmasters, bconoboy, ctyler, masta to attend 22:01:05 that's it for me 22:01:43 ok then...I got nada and I'm heading out for a holiday party in a while, so my evil plan is to poke at repo stuff for a bit first 22:02:24 #endmeeting