| *** CTtpollard has joined #baserock | 02:04 | |
| *** CTtpollard has quit IRC | 02:05 | |
| *** gtristan has joined #baserock | 06:43 | |
| *** locallycompact has joined #baserock | 07:01 | |
| *** fay_ has joined #baserock | 07:06 | |
| *** ctbruce has joined #baserock | 07:31 | |
| *** cosm has quit IRC | 07:34 | |
| *** cosm has joined #baserock | 07:35 | |
| tiagogomes | jjardon not yet | 09:08 |
|---|---|---|
| locallycompact | pedroalvarez, the problem was | 09:26 |
| locallycompact | bison depends on git-minimal but not on openssl | 09:27 |
| locallycompact | so when bison try to run git it fails with git: error while loading shared libraries: libcrypto.so.1.0.0: cannot open shared object file: No such file or directory | 09:27 |
| pedroalvarez | i wonder if we are thinking about adding runtime deps to definitions | 09:31 |
| tiagogomes | It has been a long time since I heard from that. | 09:31 |
| locallycompact | After all this code is dead we think about that for V11 | 09:32 |
| locallycompact | because it will need to be slightly more sophisticated for things when you think about runtime dependencies for things that don't get included in the system, but are non-bootstrap, but their dependencies are supposed to end up in the system | 09:33 |
| locallycompact | at the minute all we have is staging: in, bootstrap: out, but you also need to do that for whole assemblages now | 09:34 |
| tiagogomes | How that will work, I can see 5 cylinders in each side of the engine but not sure what the 11th will do | 09:34 |
| pedroalvarez | " broom broom" ? | 09:35 |
| tiagogomes | Baserock could benefit from some extra cc | 09:40 |
| *** radiofree has quit IRC | 10:15 | |
| locallycompact | https://gitlab.com/baserock/ybd/merge_requests/255 | 10:19 |
| pedroalvarez | hah | 10:20 |
| pedroalvarez | "Pep8 forces you to turn perfectly reasonable expressions into twisted horrors to satisfy a character constraint" | 10:20 |
| tiagogomes | 👎 | 10:29 |
| pedroalvarez | need to install emojis to follow tiagogomes nowadays | 10:30 |
| tiagogomes | mm, a terminal or irc client that understands utf8 is all u need | 10:32 |
| franred | -1 | 10:33 |
| jmacs | Could we see an example of the twisted horror? | 10:33 |
| locallycompact | jmacs, I would have to split up the line I linked | 10:33 |
| locallycompact | which is a perfectly fine line and I shouldn't have to do that | 10:34 |
| * tiagogomes also points out that pep8 can be configured to ignore some types of warnings. E.g. I'll have to do it for sandboxlib as the includes are in a the bottom of a python file, but it needs to be that way without breaking sandboxlib api | 10:34 | |
| locallycompact | it reads better as a single line | 10:34 |
| jmacs | Ah, I didn't realise "d1df1ac3" was a link to a different issue | 10:34 |
| * SotK probably wouldn't go so far as to call that a "twisted horror" | 10:35 | |
| * tiagogomes observers ybd saying there is a cached system artifact after adding build-essential-runtime to the minimal system | 10:53 | |
| tiagogomes | artifact splitting is wrong, and I don't have much faith on the cache key calculation | 10:54 |
| locallycompact | splitting is happening in the wrong place for start | 10:55 |
| tiagogomes | jjardon, anahuelamo my branch is at https://gitlab.com/tacgomes/ybd/commits/tacgomes/artifact-splitting . No PR yet | 10:56 |
| anahuelamo | thanks tiagogomes :) | 10:56 |
| tiagogomes | anahuelamo jjardon you may want to clean up the split rules regexes on DEFAULTS for the usr-merge, as there is no need to consider /lib, /bin and /sbin . | 11:00 |
| *** radiofree has joined #baserock | 11:13 | |
| jjardon | tiagogomes: ah! so there was a problem in ybd, not only in definitions | 11:16 |
| tiagogomes | The problem was with ybd only. The suggested changes on DEFAULTS are just cleanups. | 11:17 |
| tiagogomes | lulz, now I have added a couple of system integration commands to a chunk and ybd reports there is an existing cached artifact | 11:36 |
| pedroalvarez | that's the expected behaviour I believe | 12:23 |
| *** gtristan has quit IRC | 12:23 | |
| tiagogomes | How so? The system artifact is created after the integration commands are run | 12:26 |
| SotK | a cached chunk artifact or a cached system artifact? | 12:27 |
| pedroalvarez | same question ^ | 12:27 |
| tiagogomes | sorry, cached system artifact. | 12:28 |
| * SotK thinks that is probably not the expected behaviour then | 12:28 | |
| pedroalvarez | indeed, a bug | 12:29 |
| pedroalvarez | that regression happened already, and was fixed. Maybe there is a related fix in git history | 12:30 |
| tiagogomes | ybd runs the integration commands before artifact splitting. Is there any reason for this? | 13:01 |
| pedroalvarez | before artifact splitting? | 13:02 |
| pedroalvarez | I don't understand | 13:02 |
| pedroalvarez | I thought ybd runs these commands just before system artifact caching | 13:02 |
| rdale | it creates a system using the artifact splitting rules, and then runs the integration commands, unless it's been changed | 13:03 |
| tiagogomes | no, it runs the integration commands on the whole system, and then does the splitting | 13:03 |
| pedroalvarez | what is "does the splitting" | 13:04 |
| pedroalvarez | takes out what it doesn't need? | 13:04 |
| rdale | ybd caches complete artifacts with no splitting, but with the lists of files for each split artifact in /baserock, and then when it assembles a system it uses those .meta files to only copy the required split artifacts into the system area of a sandbox before running the integration commands | 13:06 |
| SotK | why does it cache without splitting ooi? | 13:08 |
| pedroalvarez | to have only one artifact per chunk, I guess | 13:09 |
| tiagogomes | rdale, 1) ybd copies all artifacts into $stagingarea, 2) runs the integration commands 3) copies the files in $stagingare into $stagingaera/some-system.inst that should be on the system accordingly to the .meta files. | 13:09 |
| pedroalvarez | (note, i've seen "some-system.inst" sometimes in the rootfs of the deployed system with ybd) | 13:10 |
| rdale | it's possible that ybd has changed then | 13:11 |
| tiagogomes | SotK who knows. To reduce the code complexity perhaps. The bad side is increased download speeds and disk space | 13:11 |
| pedroalvarez | tests! | 13:12 |
| tiagogomes | There is a small chance that I am misinterpreting the code | 13:13 |
| tiagogomes | My argument to run the integration commands *after* the splitting, is that way we would only have to copy the files a single time. Otherwise, we have to copy them twice. | 13:15 |
| rdale | that's certainly how it used to be - it looks like development is going round in circles | 13:24 |
| rdale | i think perhaps the issue was that a miniimal system after artifact splitting didn't have the code left in it to run the integration commands, and so be running the commands before the splitting it would fix that. but if that was the reason, i don't think it was a good fix | 13:28 |
| locallycompact | devel build | 13:30 |
| * locallycompact dies | 13:30 | |
| tiagogomes | Right. That would be a reason for post-splitting system integration. However, with the current approach, system integration commands may have as input files that won't be in the final system. A bad example is ldconfig - it would potentially find libraries that couldn't be installed in the final system. | 13:34 |
| rdale | there should be build depends or system integration depends for adding packages that are needed for running the integration commands, but they would not be in the final system and may come from chunks that aren't even in the unsplit system | 13:38 |
| rdale | imho | 13:39 |
| persia | I very strongly believe that the semantically correct order is to install only the desired objects, and then run the system integration commands. For things like updating the ld.so cache, doing it in any other order is going to mean the system integration commands do the incorrect thing. | 13:57 |
| persia | Whether the tools currently do this is independent of this opinion. | 13:57 |
| persia | (and I'd be very interested in a description of any tool that should not be int he final system but is required to run system integration commands) | 13:57 |
| *** CTtpollard has joined #baserock | 14:16 | |
| *** leeming has quit IRC | 14:27 | |
| *** leeming has joined #baserock | 14:42 | |
| *** gtristan has joined #baserock | 15:04 | |
| *** tiagogomes has quit IRC | 15:23 | |
| *** tiagogomes has joined #baserock | 15:31 | |
| *** CTtpollard has quit IRC | 15:40 | |
| *** locallycompact has quit IRC | 15:42 | |
| *** ctbruce has quit IRC | 16:18 | |
| *** ctbruce has joined #baserock | 17:22 | |
| *** ctbruce has quit IRC | 17:25 | |
| *** ctbruce has joined #baserock | 17:27 | |
| paulsherwood | rdale: calling my baby ugly, again :) | 17:42 |
| paulsherwood | there are systems where the splits make it implossible to run the si commands | 17:44 |
| paulsherwood | hence the splitting has to happen after the si commands have been run | 17:45 |
| * paulsherwood had to rework the splitting code, to have any chance of making default splits work | 17:46 | |
| paulsherwood | if others want to rework it again, by all means have at it. i don't think it's been 'going round in circles' though... | 18:07 |
| *** CTtpollard has joined #baserock | 19:58 | |
| *** CTtpollard has quit IRC | 20:25 | |
| *** CTtpollard has joined #baserock | 21:10 | |
| *** ctbruce has quit IRC | 21:33 | |
| *** CTtpollard has quit IRC | 22:00 | |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!