| 18 Apr 2026 |
 Vladimír Čunát | It's quite unfair, as different NARs compress very differently. | 13:53:41 |
| Vladimír Čunát | (it could be just a fraction of the size or unchanged, commonly) | 13:54:01 |
 Arian | Wut why did the Amazon image grow? | 15:43:43 |
| Arian | We didn't change anything .. | 15:43:46 |
| Arian | Oh wait I know why | 15:43:53 |
| Arian | I changed the default output from VHD to raw | 15:44:02 |
| Arian | https://github.com/NixOS/nixpkgs/pull/511177 | 15:45:16 |
| Arian | I guess I can work around this by converting this to raw in the upload script but I agree that measuring compressed size would be more convenient | 15:46:55 |
| Vladimír Čunát | This time we get blocked by sway:
https://hydra.nixos.org/build/326659501#tabs-buildsteps | 19:38:48 |
| Vladimír Čunát | It doesn't seem just flaky: https://hydra.nixos.org/build/326659503#tabs-buildsteps | 19:39:39 |
 mdaniels5757 | https://github.com/NixOS/nixpkgs/commit/f1bcb61731224bd8440510fc620d3c51f3e51c85 is the problem I think | 20:13:45 |
| mdaniels5757 | Fixing | 20:15:56 |
| mdaniels5757 | https://github.com/NixOS/nixpkgs/pull/511225 | 20:23:52 |
| mdaniels5757 | Reviewed and merged, if someone wants to kick off another eval of NixOS unstable. | 20:32:06 |
| 19 Apr 2026 |
 tfc | oh my god i think i force-pushed to staging-nixos, can someone please check and help me revert if i am right? | 17:11:31 |
| tfc | that was not my intention | 17:12:02 |
| tfc | oh it was also not my intention to post it here and wanted to psot in staging instead. sorry for the noise. | 17:12:19 |
 hexa | is that even possible? | 17:16:01 |
| tfc | i think it should not. continuing the discussion in staging channel | 17:16:36 |
| hexa | emily, Alyssa Ross? | 17:16:50 |
| 20 Apr 2026 |
|  picnoir changed their display name from Picnoir to picnoir. | 10:05:32 |
|  PhiliPdB joined the room. | 20:57:59 |
| 21 Apr 2026 |
|  gigacode joined the room. | 01:37:18 |
 dramforever | how far are we from flipping on ca-derivations for nixpkgs and {hydra,cache}.nixos.org? i'm looking for some estimate on the level of like "in 2026" "probably before 2028" "at least five years" | 02:32:15 |
 K900 | Possibly never | 04:44:25 |
| K900 | There are fundamental problems with ca-drvs that have yet to be solved | 04:45:02 |
| dramforever | "nowhere near" is fine, thanks for the info | 04:46:09 |
| K900 | The very short version is that ca-drvs relying on rewriting all references at build time basically means every single thing that embeds a reference in a non-trivial way needs to somehow be special cased AND rewritten without breaking things | 04:48:05 |
| K900 | With refscan it's at least fairly likely that even if you miss a spot, you'll find the same dependency in another location | 04:48:39 |
| K900 | With ca-drvs, the failure mode of missing a spot is "fire" | 04:48:52 |
