Nixpkgs Architecture Team | 233 Members | |
| https://github.com/nixpkgs-architecture, weekly public meetings on Wednesday 15:00-16:00 UTC at https://meet.jit.si/nixpkgs-architecture | 52 Servers |
| Sender | Message | Time |
|---|---|---|
| 2 Aug 2022 | ||
| and while I'm personally very familiar with sexprs, I think they represent "weird new syntax" for most folks | 14:58:17 | |
| Currently listening to the Summer of Nix live talk "The Architecture and History of Nixpkgs" https://www.youtube.com/watch?v=TKgHazs3AMw | 15:21:37 | |
| Not that much about nixpkgs architecture as of yet thouggh | 15:22:12 | |
| 15:53:13 | ||
| 15:54:15 | ||
| 16:12:16 | ||
In reply to @infinisil:matrix.orgthe question about localization somebody just asked is weird to me | 16:25:25 | |
| isn't the software packaged by nixpkgs localized already? | 16:25:46 | |
I'm guessing it's localization of e.g. meta.description, NixOS option descriptions, eval error messages, etc. | 16:26:25 | |
| That's about the most nixpkgs could do | 16:26:41 | |
| and maybe the output of some scripts that are original in nixpkgs? | 16:27:19 | |
Yes, it was mostly about package and module descriptions (e.g., output from nix search nixpkgs, etc.) | 16:27:28 | |
| 16:27:30 | |
| * and the manual | 16:27:33 | |
| Oh yeah manual is a big one | 16:28:04 | |
| Here's a related issue: https://github.com/NixOS/nix/issues/6842 | 16:28:30 | |
| I was presented with the following SoN issue: "i18n and l10n are a big part of NGI0. Nixpkgs currently doesn't support this." The following is an excerpt from my response after an initial assessment: I think this could do wonders for Nix ecosystem adoption in general. However, the scope of this particular issue is not quite clear to me, making it difficult to curate. It seems that all actionable tasks fall under the category of generated documentation, targeting the following components:
While I would consider it out of scope for this issue, the effort could be rounded out by implementing i18n for:
A note regarding NixOS itself (as opposed to its documentation): I don’t see anything to be done. With the existing impl, it seems that most use cases are accommodated:
Prior art:
| 16:31:51 | |
| So, the question was to get Jon’s feeling regarding l10n infrastructure in Nixpkgs, but in retrospect, I agree it probably felt pretty random w/o context😅 | 16:42:30 | |
| ctem: I recently learned that DeepL supports a glossary, so even if you wanted to translate the manual, you could give it a list of nix terms and their technical translation and let it translate the “stuff in between” automatically | 16:50:26 | |
In reply to @profpatsch:augsburg.oneThanks! That’s not something I’d considered. | 16:52:48 | |
| Would this be expected of the user? From the docs: "With the DeepL API Free plan, you can translate up to 500,000 characters per month for free." There’s also the issue that it isn’t open-source/self-hostable. I’m struggling to understand how this would integrate with Nix documentation CI/CD. | 16:59:08 | |
| Or perhaps we could set up the conventional infrastructure for community-contributed translations (e.g., with Weblate), and use DeepL in the beginning to generate starting content in bulk... | 17:15:22 | |
| Surely people can machine translate things themselves if they want a machine translation | 17:22:58 | |
| the hard part with translation imo is keeping it up to date | 17:24:13 | |
| it's one thing to get a one-off translation done, but it's quite another to have people keeping it up to date with the authoratative source | 17:24:46 | |
| and if it's not up to date, it ends up being potentially worse than no translation, because people will learn outdated practices etc | 17:25:18 | |
| I don’t have much experience in this area, but for example with gettext, it’s my understanding that translated strings are identified by the authoritative originals, falling back to those originals where a match isn’t found for the specified locale. It seems that (implemented correctly) this would prevent the issue of outdated translations. (Whether this holds up in practice, however, I can’t say.) | 17:30:12 | |
| But then it would result in patchy, half-translated docs for all but the most popular languages. | 17:31:34 | |
| (I think this is the norm for many for whom English is a second language, though.) | 17:32:34 | |
| gettext is very different to full length prose manuals | 17:33:25 | |