| 16 Dec 2025 |
 helle (just a stray cat girl) | looks merged now? | 12:06:57 |
| helle (just a stray cat girl) | it is iirc via a queue | 12:07:07 |
 Rutile (Commentator2.0) feel free to ping | see zulip; cl bot was down or something | 12:13:09 |
| helle (just a stray cat girl) | makes sense | 12:13:28 |
|  0x4fbb09 it/its ⛯✇ΘΔ changed their display name from 522 it/its ⛯ΘΔ to 0x4fbb09 it/its ⛯✇ΘΔ. | 15:15:22 |
| 17 Dec 2025 |
 SomeoneSerge (back on matrix) | What do Lix people think of these recent scheduling/resource allocation proposals on NixOS Discourse?
https://discourse.nixos.org/t/introducing-nix-scheduler-hook-a-build-hook-for-sending-builds-to-job-schedulers/73038/
https://discourse.nixos.org/t/proposal-for-new-experimental-feature-resource-management/73099 | 02:35:49 |
 raitobezarius | We saw and looked at it, it's cool to outsource to slurm, I have more opinions on this topic but I am on holidays right now, the key points are that Lix is removing the build hook to replace it by a structured RPC protocol, the second is that we are very close to plug all the pieces to have the collection infrastructure to make decisions about scheduling but we didn't start designing what would a scheduler look like, finally, we had some discussions with nikstur on a remote builder proxy that can have scheduling data baked into it | 02:45:58 |
 delroth | semi-related, but one thing that's been on my thinking list for a while is having the lix daemon manage a proper healthchecked connection pool to builders - i.e. instead of always creating builder connections on demand, keep connections ready ahead of time, with some basic regular healthchecking to make sure the connection isn't dead when it gets needed by something. Is it something that's already on the TODO list, is there a design or anything in that area? | 03:34:05 |
| delroth | I suspect this is in the category of "not very hard to implement but huge improvement to remote builds by decreasing constant per-build time, and also addresses some annoying behavior like random SSH timeouts blocking builds" | 03:35:07 |
| raitobezarius | https://git.lix.systems/lix-project/lix/issues/348 you're looking for this issue I believe but probably an extended form | 09:27:04 |
| raitobezarius | There's no design that achieves the right thing but this is an area where Cap'n'Proto is "installed" so it should not be that hard to rework the build-remote process to accept a ready builder connection and move the active health management in the subdaemon | 09:28:36 |
 John Ericson | we also want to get rid of the build hook | 17:57:07 |
| John Ericson | it should be possible to have this pluggability without such a thing | 17:57:16 |
 just1602 | In gerrit, if I start a CL, someone take it over and push some more changes to it, how can I sync my current version of the CL with the one currently in gerrit to be able to continue the work? | 18:20:27 |
 kloenk | Via the instruction in the download section? (d as shortcut) | 18:21:40 |
| just1602 | Thanks | 18:22:05 |
| SomeoneSerge (back on matrix) | Thanks, and enjoy the holidays!
Do we care about what interface (s.a. stringly system-features) derivations use to communicate hints&constraints to their schedulers, for derivations to still "make sense" on systems without the right plugins?
| 21:16:55 |
 piegames | If there were, hypothetically, a blessed Rust Nix parser library, what kind of interface would people who would use it want it to have? | 22:34:01 |
| piegames | Custom generic AST? CST? Whatever Rowan does? Something with tree-sitter? | 22:35:14 |
| kloenk | At what stage? But rowan cst sounds very reasonable. Just reexport the AstNode trait :) | 22:35:50 |
| piegames | I'm leaning towards something CST-like because I imagine sooner or later someone will want to implement code manipulation on it | 22:36:31 |
| piegames | I haven't used Rowan yet, so no clue how well it is even fit for that | 22:39:56 |
| piegames | Also it's generally weird to recycle the tree data structures from a parser framework but then not use the actual parser it provides ^^ | 22:40:22 |
| kloenk | Which parser do you mean? Dont think rowan has a parser? At least did not find one | 22:42:04 |
| kloenk | Redacted or Malformed Event | 22:42:39 |
| piegames | eh, I meant if I reused the rnix-parser AST which uses rowan | 22:45:55 |
| piegames | which I might do at least for getting headstart | 22:46:05 |
| kloenk | Ah that way. Yeah good question then. Did not look into rnix-parser yet. Could still make sense to define all the types, not sure how well they are in there | 22:46:52 |
 rosssmyth | Another option from rowan is cstree | 22:48:39 |
| rosssmyth | or syntree | 22:49:39 |
