**** BEGIN LOGGING AT Sat Sep 22 03:00:00 2018 Sep 22 07:38:19 New news from stackoverflow: Can bitbake include/require a file based on a variable? Sep 22 11:31:43 JPEW: I've thought we should do that by default for a while Sep 22 11:32:18 JPEW: -dbg and -dbgsrc Sep 22 13:23:03 nrossi: Oh, I see you have added migen to your meta-hdl layer! :-P Sep 22 13:52:03 mithro: yep, python stuff is very easy to add :). I was also looking at vtr but hit some issues w.r.t. nativesdk building Sep 22 14:12:05 nrossi: Oh cool Sep 22 14:12:14 nrossi: Could we do litex and friends? Sep 22 14:13:01 nrossi: https://github.com/timvideos/litex-buildenv/blob/master/scripts/download-env.sh#L406-L427 Sep 22 14:13:14 mithro: sure, but i'm learning about the tools as i go :) so it might take me a bit Sep 22 15:13:56 mithro: HI! Just fixed a warning from COW.py, was thinking of you :) Sep 22 15:14:08 RP: Ha! Sep 22 15:14:27 RP: I'm trying to not reinvent bitbake / yocto / openembedded.... Sep 22 15:14:38 nrossi: Seems to be making progress in making that possible Sep 22 15:15:28 mithro: you're having difficulty using bitbake/OE for something? Sep 22 15:17:32 RP: https://docs.google.com/document/d/1V47Zc0eyjUUGBNsJuYORdb4TqH4C1WjzrX9odX0YdZE/edit and https://docs.google.com/document/d/1hBvcE83vDRn5EVzcv4Qv_dcGlDuXp5pNI-_yzrEOwxs/edit Sep 22 15:19:04 RP: I know what I /want/ and think that bitbake/OE can _probably_ do what I want - but trying to find the time to understand how to make bitbake/OE do what I want is hard to find Sep 22 15:24:54 mithro: Looking at the requirements, I think it can do many of the things you mention. Some of the things are hard to get right like reproducibility and we've a lot of work on making it work right Sep 22 15:25:32 mithro: whilst I only have finite time, I am happy to give pointers on specific issues if it would help Sep 22 15:26:13 RP: Yeah - one big question is requirement checking is a big thing Sep 22 15:26:26 RP: IE Checking you have Xilinx toolchain and stuff installed Sep 22 15:27:02 mithro: that is basically what sanity.bbclass does for things like git being present and of the right version Sep 22 15:27:35 mithro: bitbake generates events at different parts of the build and you can do pretty much anything in an event handler Sep 22 15:29:48 RP: This file? https://github.com/openembedded/openembedded/blob/master/classes/sanity.bbclass Sep 22 15:30:44 mithro: yes Sep 22 15:31:57 mithro: its complicated by caching and compatibility porting code to upgrade old builds to new formats, it'd be possible to do something much simpler Sep 22 15:32:35 RP: What would be an example to check for Xilinx? Sep 22 15:32:42 RP: Our shell script does this -> https://github.com/timvideos/litex-buildenv/blob/master/scripts/download-env.sh#L165-L266 Sep 22 15:35:37 I can't seem to find why bitbake wants to rebuild a given package, while it takes it from shared-state on another machine (with the same sstate, and that's a pyro-based build) - how can I dig into that ? Sep 22 15:39:24 yann: such cases, usually mean there is some unwanted host dependency that crept into variable references used to calculate sstate hash, you can dump the signatures and compare them bitbake-diffsigs is handy this page might be helpful too https://wiki.yoctoproject.org/wiki/TipsAndTricks/Understanding_what_changed_(diffsigs_etc) Sep 22 15:40:10 thx khem, digging that Sep 22 15:42:38 mithro: http://git.yoctoproject.org/cgit.cgi/poky-contrib/commit/?h=rpurdie/mithro&id=9edc54e29e6f8778ab34d5b90e37fac1894042ed Sep 22 15:43:24 mithro: INHERIT += "sanity2" in distro config to enable. You'd call your script check-xlinx-toolchain, or you could put a shell function in that class Sep 22 15:44:30 mithro: its made ugly by having the sanity events mechanism to report back problems, really should just mkae that the default and simplfy the code but never got around to that Sep 22 15:45:17 mithro: oh, that code probably needs to give the handler another name so it doesn't clash with sanity.bbclass Sep 22 15:49:49 khem: I'm still not sure that the particular package i'm looking at is not getting rebuilt because one of its deps has been - is there a quick way to check this ? Sep 22 15:52:00 yann: I would suggest to do build from exact metadata on the offending machine then ideally it should not rebuild anything and if it does runinto do_compile tasks then you take the first do_compile task appearing in logs and chase it Sep 22 15:52:29 khem: safe trip back? Sep 22 15:52:55 khem: still here and its still raining :) Sep 22 15:53:59 thx Sep 22 15:55:20 khem: clearing tmp/work should be enough for such a test, right ? Sep 22 15:58:25 yes Sep 22 16:00:24 RP: my flight got cancelled and they let me in because when I was doing security it was not yet cancelled, apparantly I was first one to checkin for that flight, and luckily another flight to SF was delayed and I ran 200m dash to reach there and beg them to put me in that flight luckily I got in and was sitting on last row but yeah reached safely, Sep 22 16:01:53 btw. air canada send me a notice of cancellation and rebooking me to same flight on saturday, and just this morning I got notified that even the saturday flight is cancalled haha Sep 22 16:02:18 I wonder if that flight is just a scam Sep 22 16:03:03 khem: sounds like a lucky set of circumstances! glad you made it :) Sep 22 16:07:39 RP: Did git://git.yoctoproject.org/openembedded-layer disappear or never exist in the first place? "ERROR: Layer 'hdl' depends on layer 'openembedded-layer', but this layer is not enabled in your configuration" Sep 22 16:07:43 yeah, Sep 22 16:08:12 mithro: this is layername which you should find in layer.conf of meta-oe Sep 22 16:08:30 oh, after moving tmp/work away bitbake thinks it can compile ncurses-native without even fetching it - a bit scary :) Sep 22 16:09:37 khem: But where do I clone the meta-oe layer from? Sep 22 16:10:09 Oh it's http://git.openembedded.org/meta-openembedded/ ? Sep 22 16:11:34 mithro: yes Sep 22 16:12:12 yann: ncurses did have some quirks in the past so you might want to check patches from master and see if something pops up to help Sep 22 16:14:39 RP: I havent built gcc based musl builds for a while and now I see http://errors.yoctoproject.org/Errors/Build/68730/ :(. I wonder how it knows I am not watching Sep 22 16:21:57 khem: ouch. All related or diferent issues? OE-Core should at least be clean, we test that. Thisis what autobuilders are good at... Sep 22 16:27:21 hm, strange - I could get past the ncurses-native issue when I realized I had recently added a do_install_append for ncurses and omitted to make it class-target. But now (with a clean tmp/work or with my original) it's trying to rebuild gcc-cross-initial-aarch64 without running fetch first Sep 22 16:29:53 yann: gcc recipes are special in that they have a gcc-source recipe to handle the sourcecode Sep 22 16:33:52 RP: right, but it strikes me it gets the same problem I had with ncurses Sep 22 16:35:37 even when creating a new build dir (same local.conf so same sstate), its first task ever is gcc-cross-initial_6.4.bb:do_compile, tmp/work isn't even created Sep 22 16:36:16 I probably messed up something, can't see another reason why it would suddently break so badly Sep 22 16:36:18 it wont rebuilt anything except populating gcc-sources Sep 22 16:48:17 hm, apparently my immediate problem was that when switching from one build to another just re-sourced oe-init-build-env and some old values are apparently not rewritten Sep 22 17:00:23 so now a brand new build dir decides quilt-native must be compiled, so I now have sigdata for this host - however, the well-behaved one taking artifacts from sstate only has rm_work.sigdata as non-empty - I'm a bit at a loss about what to compare Sep 22 17:21:37 RP: I figured that its because this builders python = python3 but some recipes are expecting python = python2 Sep 22 17:21:52 RP: new build is on going lets see what comes out Sep 22 17:26:06 RP: Sstate summary: Wanted 14111 Found 13949 Missed 162 Current 0 (98% match, 0% complete) Sep 22 17:26:23 is there a way to dump the ones which did not match may be with bitbake -v ? Sep 22 17:52:44 khem: I think -DDDv probably does say which ones didn't but its ixed in with a lot of other things Sep 22 18:34:52 RP: probably its useful enough to be reported at a less verbose level maybe Sep 22 19:02:11 khem: yes, log output needs work in general Sep 22 19:22:31 if I take the sigdata generated by a spurious glic rebuild today, and compare it to the siginfo stored in the sstate (generated on another machine)... bitbake-diffsigs does not see any difference (yet the files have different size) Sep 22 19:26:33 no new sstate was stored for this compile state, which seems to confirm the sstate was here and should have been used, no ? Sep 22 19:26:33 RP: I have been maintaining all gcc patches on upstream master as well here https://github.com/kraj/gcc/commits/oe/master Sep 22 19:26:51 but this assignment thing might not work at all or might take long Sep 22 19:35:52 I'm a bit puzzled by the contents of this sstate - the oldest entry for that glibc package for this arch is a "package" one. That's not supposed to be possible, right ? Sep 22 19:48:16 RP: I am preparing a small musl update to address few regressions seen in 1.20 release Sep 22 19:59:14 khem: I was also just looking at the patches, I think we should talk to upstream about a few of them. I'm hoping some of the one line fixes may not need the full assignment process Sep 22 20:00:05 yann: only some tasks are stored in sstate. do_package is, do_compile is not Sep 22 20:00:49 yann: if the hashes for an sstate task match they will be used, if not they will be built Sep 22 20:21:44 RP: yeah they said there is 15 line cumulative limit on one liners Sep 22 20:22:03 at MV we had blanket assignment **** ENDING LOGGING AT Sun Sep 23 02:59:59 2018