| 16 Mar 2025 |
 John Ericson | hexa: We would like to no more about the bug that caused the builders to go to Lix | 18:59:34 |
| John Ericson | that earlier github thread went of the rails, but the original idea of having some more information written down still stands | 19:00:02 |
| John Ericson | Thanks to infinisil (!) I got a link that works on the element electron app to the prior conversation in infra #infra:nixos.org, the link is: https://matrix.to/#/!RROtHmAaQIkiJzJZZE:nixos.org/$G9MBOn9CfSLkLiPqMhoN7KMnBEl6R4uRkLbFBm_KWBs?via=nixos.org&via=matrix.org&via=nixos.dev | 19:37:51 |
| John Ericson | rereading that, it does look like that no one really knows what the bug is, and other coredumps that e.g. might come from tests suites doing SIGABORT mean we do not have a good paper trail to figure out retroactively | 19:39:02 |
| John Ericson | Do you agree with that? | 19:39:06 |
 hexa | I can roll back to nix in a few days | 20:25:01 |
| hexa | I'm currently running with a workaround for nix not cleaning up build dirs in /tmp | 20:25:26 |
| hexa | that got fixed in lix recently by pennae | 20:25:32 |
| hexa | crucial since we build in a tmpfs | 20:25:40 |
| 17 Mar 2025 |
 jade_ | yeah, and currently debugging protocol bugs is absolutely maddening. lix nightly currently has a known protocol bug due to concurrency and nar transfers and remote builders and gestures the whole protocol being extremely easy to screw up.
this is the idea we had for improving the experience of debugging protocol misbehaviour; implementation has not yet started: https://git.lix.systems/lix-project/lix/issues/734
| 00:12:20 |
| jade_ | * yeah, and currently debugging protocol bugs is absolutely maddening. lix nightly currently has a known protocol bug due to concurrency and nar transfers and remote builders and gestures the whole protocol being extremely easy to screw up and it is sorta reproducible (unusual for a protocol bug!).
this is the idea we had for improving the experience of debugging protocol misbehaviour; implementation has not yet started: https://git.lix.systems/lix-project/lix/issues/734
| 00:12:58 |
| jade_ | (we do not know of other protocol bugs in lix, fwiw) | 00:14:58 |
| jade_ | https://gerrit.lix.systems/c/lix/+/2639 here's the CL that fixed it IIRC, but i think there might be a second somewhere | 00:16:46 |
| jade_ | https://gerrit.lix.systems/c/lix/+/2666 ah it would be this | 00:17:08 |
| John Ericson | hexa: (or jade_) can you point me to the tmp cleaning commit? | 00:17:28 |
| jade_ | see above | 00:17:33 |
| John Ericson | OK | 00:17:37 |
| John Ericson | I am ripping out building from scheduling finally right now | 00:17:50 |
| jade_ | https://git.lix.systems/lix-project/lix/issues/678 issue here with discussion | 00:17:52 |
| John Ericson | it is very satisifying | 00:17:56 |
| hexa | and here https://github.com/NixOS/nix/issues/5207 | 00:18:08 |
| John Ericson | I made a OOP inheritence rats nest to wind up the balista, and then twang | 00:18:27 |
| jade_ | we have been doing different types of things wrt such fixing: soon getting rid of fork() from the builder | 00:18:32 |
| hexa | not sure if the "not empty" part is accurate at this point | 00:18:40 |
| hexa | * not sure if the "not empty" part is still accurate at this point | 00:19:26 |
| John Ericson | it would be very nice to have some reference C++ that was sufficiently decoupled from the rest that we both could use it | 00:19:47 |
| jade_ | another couple changes i would be fairly happy if they got yoinked directly into cppnix:
- https://gerrit.lix.systems/c/lix/+/2801
- https://gerrit.lix.systems/c/lix/+/2800
| 00:19:48 |
| jade_ | that's what deleting the fork() is about | 00:20:04 |
| jade_ | makes it much easier to riir the sandbox setup | 00:20:12 |
| jade_ | since it no longer knows what a "store" is | 00:20:21 |
