| 21 Dec 2021 |
 andi- | Would be nice to make use of some of the systemd faciilities for "switching ostrees" | 20:41:11 |
 Mic92 | What does switching ostrees do? | 20:42:16 |
| andi- | well they already support a similar to NixOS model | 20:42:30 |
| andi- | where you also have atomic system upgrade IIRC. I think fedora silverblue is using that for updates. | 20:42:50 |
| andi- | I have never fully looked into it but that is very close to our model. | 20:43:00 |
| andi- | IIRC arian checked it out but that had the issue of always restarting all the services? | 20:43:21 |
| andi- | They wouldn't (yet) detect that some unit sare identical. | 20:43:33 |
| andi- | Thinking about it that could be fixed with my current systemd PR that we carry since about a release or so... | 20:43:51 |
| andi- | Looking forward for a few days next week to look into these kinds of things. | 20:44:01 |
|  jbedo joined the room. | 20:49:59 |
| Mic92 | In reply to @andi:kack.it
Still. I try to help out in a distro for every day people not research student with access to compute farms ;-) I just build cntr on an RPI3 with 1GB ram and LTO enabled. I think we are fine. | 20:50:36 |
| andi- | how long did that take? :-) | 20:50:49 |
|  Artturin joined the room. | 20:51:16 |
| Mic92 | mhm. I did not measure it. Not too long. It took longer to download the build environment with nix develop | 20:51:34 |
| andi- | ok | 20:55:27 |
|  lukegb (he/him) joined the room. | 21:58:38 |
 Qubasa | Mic92: Kennst du dwarffs? https://github.com/edolstra/dwarffs | 22:35:51 |
| 22 Dec 2021 |
 fzakaria (Old) | Mic92: thanks for your suggestion -- actually started a post on the mailing list about the subject | 00:49:54 |
| fzakaria (Old) | some interesting choices between linkers that's not coded in the ELF specification for dynamic objects. | 00:50:11 |
| Mic92 | In reply to @qubasa:gchq.icu
Mic92: Kennst du dwarffs? https://github.com/edolstra/dwarffs I have seen but not used it. | 04:43:48 |
 tomberek | Mic92: i've ported the guix "stat storm" approach to Nix to experiment (this is where there are per-package ld.so.cache's and glibc knows to look there. Is this something you are interested in/worked on before? | 04:46:45 |
| Mic92 | In reply to @tomberek:matrix.org
Mic92: i've ported the guix "stat storm" approach to Nix to experiment (this is where there are per-package ld.so.cache's and glibc knows to look there. Is this something you are interested in/worked on before? It sounds interesting but I am not so sure how the nix community would feel about the extra packages. Especially now that fzakaria has a promising alternative: https://github.com/NixOS/patchelf/pull/357 | 04:48:42 |
| Mic92 | In reply to @tomberek:matrix.org
Mic92: i've ported the guix "stat storm" approach to Nix to experiment (this is where there are per-package ld.so.cache's and glibc knows to look there. Is this something you are interested in/worked on before? * It sounds interesting but I am not so sure how the nix community would feel about the extra files. Especially now that fzakaria has a promising alternative: https://github.com/NixOS/patchelf/pull/357 | 04:48:51 |
| tomberek | (yeah, i saw that, hence my question) | 04:49:04 |
| Mic92 | I think we would need to test shrinkwrap in a large scale experiement. | 04:49:30 |
| Mic92 | To see if there are any major downsides | 04:49:41 |
| tomberek | i was under the impression that there was a reason absolute paths would not be desirable. something like it prevents other overloading behavior from working | 04:51:00 |
| tomberek | (i don't recall, but if it works out, great!) | 04:51:19 |
| Mic92 | In reply to @tomberek:matrix.org
i was under the impression that there was a reason absolute paths would not be desirable. something like it prevents other overloading behavior from working I guess LD_PRELOAD still works for hacks? Would ld.so cache still allow this? | 04:52:52 |
| tomberek | I believe so. | 04:53:04 |
