| *** tristan <tristan!tristan@223.62.216.189> has joined #buildstream | 02:45 | |
| *** ChanServ sets mode: +o tristan | 02:45 | |
| *** tristan <tristan!tristan@223.62.216.189> has quit IRC | 03:00 | |
| *** tristan <tristan!tristan@223.62.216.189> has joined #buildstream | 03:00 | |
| *** ChanServ sets mode: +o tristan | 03:00 | |
| *** tristan <tristan!tristan@223.62.216.189> has quit IRC | 03:47 | |
| *** tristan <tristan!tristan@223.62.216.189> has joined #buildstream | 03:47 | |
| *** ChanServ sets mode: +o tristan | 03:47 | |
| *** tristan <tristan!tristan@223.62.216.189> has quit IRC | 04:23 | |
| *** tristan <tristan!tristan@223.62.216.189> has joined #buildstream | 04:23 | |
| *** ChanServ sets mode: +o tristan | 04:23 | |
| *** tristan <tristan!tristan@223.62.216.189> has quit IRC | 04:39 | |
| *** tristan <tristan!tristan@223.62.216.189> has joined #buildstream | 04:39 | |
| *** ChanServ sets mode: +o tristan | 04:39 | |
| *** tristan <tristan!tristan@223.62.216.189> has quit IRC | 04:47 | |
| *** tristan <tristan!tristan@223.62.216.189> has joined #buildstream | 04:48 | |
| *** ChanServ sets mode: +o tristan | 04:48 | |
| juergbi | nanonyme: buildbox-fuse will deny writes (outside specified directories) for read-only root but it shouldn't affect the file/directory modes | 06:09 |
|---|---|---|
| *** tristan <tristan!tristan@223.62.216.189> has quit IRC | 06:47 | |
| nanonyme | juergbi: x86_64 build-export took about 42 minutes when I combined flatpak_repo commands manually rather than relying on nested batching | 07:45 |
| nanonyme | So 40 minutes faster | 07:46 |
| nanonyme | I will next see if results are reproducible | 07:46 |
| nanonyme | I don't think nested batching works as well as people think | 07:48 |
| nanonyme | juergbi: if I'm right, removing sandbox.batch() here should speed builds *a lot* https://github.com/apache/buildstream/blob/master/src/buildstream/buildelement.py#L306 | 07:49 |
| nanonyme | For integration commands | 07:50 |
| nanonyme | Sorry, wrong place. It was supposed to be removed from inner layer | 07:50 |
| nanonyme | So here https://github.com/apache/buildstream/blob/cf88f343c13a54f0f768b9c96f0a2e58b2d9f55a/src/buildstream/element.py#L706 | 07:51 |
| nanonyme | That inner batch is redundant. It doesn't have a title, no flags and it is always called from within outer code which has an ongoing batch anyway | 07:51 |
| juergbi | if it's always called within another batch, sure, we can remove it. however, we should also fix nested batching | 07:52 |
| juergbi | doesn't really explain the slowness of integration of e.g. sdl2-net in CI where only a single dependency has integration commands | 07:53 |
| nanonyme | Indeed. It should be fine in that case | 07:54 |
| nanonyme | I'm anyway running another rebuild here https://gitlab.com/freedesktop-sdk/freedesktop-sdk/-/jobs/3464396395 | 07:54 |
| nanonyme | I am testing if the 42 minutes is consistent build time or a fluke | 07:54 |
| nanonyme | The scenario is now basically worst possible where i686 and x86_64 are running in different Docker containers on same VM | 07:56 |
| nanonyme | juergbi: I'm not sure if nested batching isn't working as designed. You cannot possibly combine two batch groups when they have different labels without breaking logs | 07:58 |
| juergbi | the labels of the inner batches could still be shown, just not with separate timestamps | 07:59 |
| juergbi | not really different from different commands running in a single batch | 07:59 |
| juergbi | that's a downside of batching but that's expected | 08:00 |
| nanonyme | I'm also not altogether sure if it's intuitive calling add_commands twice with same label in scriptelement.py will actually replace rather than append | 08:02 |
| nanonyme | scriptelement.py is happily using label as a group identifier | 08:02 |
| *** tristan <tristan!tristan@223.62.173.73> has joined #buildstream | 09:34 | |
| *** ChanServ sets mode: +o tristan | 09:34 | |
| *** tristan <tristan!tristan@223.62.173.73> has quit IRC | 10:20 | |
| nanonyme | Update: I ran that exact same code another time. When two instances of flatpak repo build were running on same VM in different containers, duration was more than double of original | 12:26 |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!