| 15 Aug 2021 |
 andi- | Apparently we aren't the only ones with systemd v249 issues.. There is a similar issue with Gentoo and I just added more details to the issue: https://github.com/systemd/systemd/issues/20003#issuecomment-899095461 | 18:58:10 |
 Arian | Interesting | 19:00:19 |
| Arian | Could you see if cherry picking the final 2 commits of this PR helps? https://github.com/NixOS/nixpkgs/pull/121371/commits | 19:02:49 |
| Arian | It shouldn't make a difference per se. But the commits align our boot graph more with upstream | 19:03:32 |
 ElvishJerricco | In reply to @andi:kack.it
ElvishJerricco: do you have a writeup how you envision this to go into nixos? I've had some discussion around the topic in the past and would like to coordinate efforts a bit better. There have already been three attempts to do that inside and outside of nixpkgs. I don't have any kind of write up outside of the PR I've opened | 19:04:59 |
| andi- | Oh, there is a PR.. I haven't noticed (or forgot :D) | 19:06:01 |
| 16 Aug 2021 |
| ElvishJerricco | Question: For initrd systemd, should I have compatibility units for all the boot.initrd.*Commands things? E.g. I could have a unit with After=systemd-udev-settle.service and Before=local-fs-pre.target that runs postDeviceCommands | 02:37:58 |
| ElvishJerricco | The reason I don't want to is just that this is meant to be the way out of that pattern. Plus who knows what kind of harm is going to be done by the code already in nixos that assumes its in control, not systemd. | 02:44:59 |
 ajs124 | While I do make great use of those options, I think we should just migrate away from them, if we're finally transitioning to systemd in the initrd.
We can just have the usual migration plan: add optional support for systemd, deprecate non-systemd and default to systemd, remove non-systemd. | 09:19:27 |
| Arian | I'd put the initrd in a new "namespace" and make the old options errors if the new initrd is opted in to to be honest | 09:23:55 |
| Arian | wouldnt get stuck too much in backwards compat | 09:24:10 |
|  RqndomHax removed their profile picture. | 09:37:24 |
| RqndomHax removed their display name RqndomHax. | 09:37:25 |
| RqndomHax left the room. | 09:37:25 |
| andi- | Yeah, put it in a new namespace perhaps with a bit of a common option set but otherwise leave the legacy behind. Eventually deprecate it if ever possible. There are too many hacks people have been using over the years and it will take a few releases to get rid of those users (if at all). | 09:41:47 |
| andi- | In reply to @andi:kack.it
Apparently we aren't the only ones with systemd v249 issues.. There is a similar issue with Gentoo and I just added more details to the issue: https://github.com/systemd/systemd/issues/20003#issuecomment-899095461 I got a reply and I am not sure if this is going into a good direction. Are they saying that they only care about the name of a symlink target and not about the actual file anymore? That means we can't be super precise by linking to the actual unit anymore but should rather just use the names?
https://github.com/systemd/systemd/issues/20003#issuecomment-899312479 | 09:43:47 |
| Arian | I think they're saying /etc/systemd/system/default.target -> /etc/systemd/system/multi-user.target instead of $out/systemd/system/multi-user.target | 10:42:45 |
| Arian | and then multi-user.target -> $out/systemd/system/multi-user.target | 10:43:13 |
| andi- | yeah | 13:25:32 |
| andi- | I hope that is it | 13:25:37 |
| andi- | ln -s /etc/systemd/system/${cfg.defaultUnit} $out/default.target isn't that exactly what I have here? | 13:31:48 |
| andi- | we have to create the symlink in $out as that is the folder that is symlinked to /etc/systemd/system. | 13:32:15 |
| andi- | And that is already a modified version. In nixpkgs we just do ln -s ${defaultUnit} $out/default.target. | 13:34:01 |
| andi- | The problems appears to be that /etc/systemd/system is already a symlink if I understand their new logic.. | 13:37:33 |
| Arian | If that's not allowed that would be a big complication :/ | 15:37:21 |
| andi- | yeah | 15:39:57 |
| andi- | So far they are very receptive for feedback. Once this stuff is done I'll pickup my work on continously testing against systemd master. | 15:40:48 |
|  Finn Behrens (he/him) changed their display name from Finn Behrens to Finn Behrens (he/him). | 20:41:15 |
| 17 Aug 2021 |
|  hexagonk set a profile picture. | 05:57:12 |
| Arian | Alright let me know | 07:41:39 |
