Nix Hackers | 895 Members | |
| For people hacking on the Nix package manager itself | 191 Servers |
| Sender | Message | Time |
|---|---|---|
| 7 Oct 2021 | ||
| Yeah, if they're not regressions | 12:54:38 | |
| isn't the musl issue a regression? | 14:07:06 | |
| Don't think so, since it was not part of the previous release iirc | 14:50:21 | |
| But it would be great if somebody can look into it | 14:50:32 | |
While you're here sometimes I'm not getting any output even with -L on nix (Nix) 2.4pre20211001_4f49615 (and also I'm not getting the list of all derivation that failed), and it's a bit painful to reproduce. Is this a known bug that was fixed in the meantime? | 14:51:33 | |
| I would update and try to check on master, but I can't quite figure out when it happens | 14:51:46 | |
| And also I would think it's a regression | 14:52:00 | |
niksnut: nixStatic 2.3.x is exposed top level in nixpkgs which does use musl | 14:53:08 | |
In reply to @balsoft:balsoft.ruSounds like https://github.com/NixOS/nix/pull/5339 | 14:53:22 | |
| No, I think mine is different | 14:53:57 | |
| It's just in the regular terminal | 14:54:03 | |
| And not consistently so | 14:54:19 | |
| E.g. two consequtive runs of
And second one: | 14:56:18 | |
Also reproducible with passing -L to new nix commands | 14:56:32 | |
| (and by "reproducible" I mean it happens... sometimes) | 14:57:40 | |
Before diving into doing the change I had a thought yesterday while looking at related code: Any reason we don't implement Expr{Sub,Add,...} like we do for ExprNotOp etc? In the case of subtraction it could get rid of about 30k Symbol table looks, a few thousand new objects and execution wouldn't probably do a few less function calls. That would probably break overriding substraction by providing a __sub (or whatever) function in the current scope (if that even works right now). | 14:58:58 | |
In reply to @balsoft:balsoft.ruInterestingly, I think it never happens once logs for the derivation were shown at least once | 14:59:15 | |
* Before diving into doing the change I had a thought yesterday while looking at related code: Any reason we don't implement Expr{Sub,Add,...} like we do for ExprNotOp etc? In the case of subtraction it could get rid of about 30k Symbol table looks, a few thousand new objects (while building a NixOS test). And execution wouldn't probably do a few less function calls. That would probably break overriding substraction by providing a __sub (or whatever) function in the current scope (if that even works right now). | 14:59:20 | |
| I don't mind at all if this ends up in the release, but just wanted to note that it happens sometimes | 15:01:23 | |
| niksnut: any chance we could get a 2.3 release or merge this? https://github.com/NixOS/nixpkgs/pull/137197 I think it should fix the errors we're seeing on https://r13y.com | 15:03:53 | |
| Sure, I can do a release | 15:04:16 | |
In reply to @andi:kack.itNo, I think the reason it's implemented that way is because builtins.sub etc. already existed | 15:14:19 | |
In reply to @niksnut:matrix.org Would you see any downside to adding those implementations? I could have a go on the in one of the coming evenings. I don't think that it will improve performance by a lot but I also haven't measured the time spent in those functions. I was just surprised that IIRC the sub primop was the most used one (by a few 10k) when I did eval a NixOS VM test. | 15:16:49 | |
| The only downside is that it's a bit more code | 15:17:36 | |
| musl issue is related to: 4993174 | 15:19:43 | |
In reply to @niksnut:matrix.org2.3.16 is up | 15:36:21 | |
In reply to @tomberek:matrix.orgOkay, we could revert that | 15:37:09 | |
| niksnut: thank you so much! | 15:54:59 | |
| I've created a 2.4-maintenance branch and a hydra jobset | 15:57:10 | |
| I'll do a rc1 release for testing | 15:57:19 | |