| 29 Jun 2026 |
 Myria | Would this be a gc problem? | 01:14:33 |
 vcunat | Yes, there's been lots of these. | 05:50:57 |
 K900 | https://github.com/scipy/scipy/issues/25488 | 09:10:26 |
| K900 | Scipy has more weird float precision issues | 09:10:32 |
| K900 | Going to skip that test once I confirm it builds on my 3588 | 09:10:45 |
| K900 | Skipped that test, found another test that's also flaky | 09:31:43 |
| K900 | smh my head | 09:31:45 |
| K900 | Well good news it also fails on hydra | 09:40:03 |
| K900 | https://hydra.nixos.org/build/333391133/nixlog/6/tail | 09:40:04 |
| K900 | With a DIFFERENT TEST | 09:40:07 |
| K900 | The test_convergence ones are flaky, test_nyquist seems to fail consistently | 09:40:55 |
 kuflierl | This might be related to the gui pop-up. Maybe the gui pop up has gamepad input support? Steamdeck? | 09:45:15 |
| K900 | It's in webkitgtk | 09:46:40 |
| K900 | Not openconnect | 09:46:43 |
| K900 | And webkitgtk supports the gamepad API because it's a browser engine | 09:46:55 |
| kuflierl | "gui-popup" | 09:48:17 |
| K900 | The point is, the webkitgtk is the same everywhere | 09:48:52 |
| K900 | Every application that uses webkitgtk uses the same webkitgtk | 09:49:00 |
| K900 | And that webkitgtk needs all the features | 09:49:15 |
| vcunat | Ah sure, gamepad on web games 😁 | 10:00:01 |
| vcunat | * Ah sure, gamepad for web games 😁 | 10:00:13 |
| vcunat | Rebuild progress after 2.5 days, counting successful jobs vs. all jobs:
- ~41 % on x86_64-linux
- ~29 % on aarch64-{darwin,linux)
With this it seems realistic that we can make the staging* iterations much faster than the current pace of one cycle per two weeks.
| 10:56:19 |
| vcunat | * Rebuild progress after 2.5 days, counting successful jobs vs. all jobs:
- ~41 % on x86_64-linux
- ~29 % on aarch64-{darwin,linux)
With this it seems realistic that we can make the staging* iterations much faster than the current pace of one cycle per two weeks.
| 10:56:24 |
| vcunat | * Rebuild progress after 2.5 days, counting successful jobs vs. all jobs:
- ~45 % on x86_64-linux
- ~29 % on aarch64-{darwin,linux)
With this it seems realistic that we can make the staging* iterations much faster than the current pace of one cycle per two weeks.
| 10:58:04 |
| vcunat | 2.5/0.29 = 8.6 | 11:00:35 |
| vcunat | (and spare x86_64-linux capacity for experimental jobsets, incl. haskell-updates, etc.) | 11:01:22 |
|  cSanthiago joined the room. | 12:24:40 |
| K900 | OK I think I'm skipping more tests | 12:27:00 |
| K900 | On scipy | 12:27:04 |
| K900 | Because this is awwful | 12:27:08 |
