| *** xjuan has quit IRC | 00:32 | |
| *** tristan has joined #buildstream | 04:12 | |
| *** ChanServ sets mode: +o tristan | 04:12 | |
| *** tristan has quit IRC | 04:57 | |
| *** tristan has joined #buildstream | 05:14 | |
| *** ChanServ sets mode: +o tristan | 05:19 | |
| tristan | juergbi, interestingly I'm finding that, while the depth sort is still good for prefetching sources in a decent order, I need to change the scheduler so that elements are moved from one queue to the next differently | 09:10 |
|---|---|---|
| tristan | I have a situation where a clone of linux.git is blocking the scheduler, where there are buildable elements behind it | 09:11 |
| tristan | (which I could already be building) | 09:11 |
| juergbi | tristan: at least in my wip branch the scheduler checked all fetched elements whether they can be moved to the build queue when the build queue is shorter than the number of available builders | 09:14 |
| juergbi | i.e., there shouldn't be any blocking in that scenario, iirc | 09:14 |
| tristan | I might be wrong :-/ | 09:17 |
| tristan | now I see there are failures (installs which do setcap) | 09:17 |
| tristan | juergbi, I think you're right, everything that passes through the fetch queue ends up on the build queue | 09:18 |
| tristan | regardless of whether a fetch was needed (and the whole plan goes on to the fetch queue first at once) | 09:18 |
| juergbi | yes, that's how i remember it as well | 09:19 |
| tristan | I run `bst show` regularly during a long build, but the 'buildable' elements it shows are the ones which failed | 09:20 |
| juergbi | hm, at the point when an element build is triggered, it is (or used to be) removed from the build queue | 09:21 |
| juergbi | which would mean that at the point of a build failure, it can't still be in the build queue | 09:21 |
| tristan | juergbi, yeah but since it's a looong build, I run `bst show` on the same pipeline in a different terminal :) | 09:24 |
| tristan | So I can see what remains to be built and what we've built | 09:24 |
| tristan | when there are ~500 elements, it's nice to check sometimes :) | 09:24 |
| * tristan realizes this pipeline is breaking because of arch, it should be 32bit but I did not do it under linux32 | 09:26 | |
| * tristan will still wait for the linux.git download to complete | 09:26 | |
| tristan | would be nice to automate that part in the sandbox | 09:27 |
| tristan | for host compatible arches | 09:27 |
| *** ssam2 has joined #buildstream | 10:12 | |
| ironfoot | is the arch of the host being used in the cache key calculation? | 10:41 |
| tristan | ironfoot, Yes, unless you override it with `bst build --arch foo` | 10:49 |
| tristan | in which case, the argument will be used instead of detected uname arch | 10:49 |
| tristan | ironfoot, my plans for this are a bit vague at the moment, but what I'm hoping for is for the BuildStream core to not care too much about arch, and not have any kind of specific list of supported arches | 10:50 |
| tristan | or what they mean | 10:50 |
| tristan | ironfoot, for the baserock definitions conversion, it expects the arch names that baserock supports and is pretty much handled in the project.conf | 10:51 |
| tristan | for flatpaks, it will need to support x86_64, i386, arm and aarch64 | 10:52 |
| tristan | where i386 is actually i586 | 10:52 |
| tristan | and arm is actually an armv7a on steroids | 10:52 |
| ssam2 | yeah, if buildstream wants to be generic then it can't really hardcode a list of arches | 10:53 |
| ssam2 | since every community uses the same names to mean different hings | 10:53 |
| ssam2 | *things | 10:53 |
| tristan | right | 10:57 |
| tristan | so right now I'm "doing nothing" until something becomes necessary, and so far it's working :) | 10:58 |
| jjardon | FWIW. the list of arches is in DEFAULTS in definitions | 11:10 |
| ssam2 | the mapping to GNU triples is somewhere else though | 11:11 |
| tristan | ssam2, that was in ybd sandbox.py, but we have it with arch conditionals in the project.conf | 11:11 |
| tristan | And, while porting build-essential, I believe I have found that the lists are inaccurate | 11:12 |
| tristan | I think they are not consistently supported across chunks in build-essential | 11:12 |
| ssam2 | not sure what you mean. I'm sure there's only enough support to work on the platforms we've tested on | 11:23 |
| tristan | ssam2, I dont recall exactly either, and I might be thinking about a mix of what definitions says and what hard code in ybd says | 11:26 |
| *** tristan has quit IRC | 13:11 | |
| *** tiagogomes has quit IRC | 14:49 | |
| *** tiagogomes has joined #buildstream | 14:50 | |
| *** mint has joined #buildstream | 15:03 | |
| *** mint has quit IRC | 15:26 | |
| *** tristan has joined #buildstream | 16:18 | |
| *** ssam2 has quit IRC | 18:02 | |
| *** tristan has quit IRC | 19:49 | |
| *** hergertme has joined #buildstream | 23:03 | |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!