Show HN: a18n – Automated I18n JavaScript Solution(github.com)
github.com
Show HN: a18n – Automated I18n JavaScript Solution
https://github.com/FallenMax/a18n
26 comments
I naturally assumed, and by naturally I mean egrep '^a.{18}n$' /usr/share/dict/words, that a18n abbreviated anthropomorphization, and so in context that perhaps we'd be helping a recently translated program feel more settled in its new repository.
Not an English speaker but this is still awkward: I always think "i18n" expands to 18 characters until I see your comment and checked `"internationalization".length` in console to be 20
yeah 18 is replacing the middle 18 characters in the word, not including that i and n at the ends. I personally don't like it because it's just an English speaking programmer in-group shibboleth that has bled over to k8s for kubernetes and a16z for Andreessen-Horowitz
Thanks! Yes, Automation is what actually I thought of the 'a' in a18n :)
But it's named as `a18n` for some complicated reasons:
1. It aligns well with translation function name and CLI command
I want the translation function to be something:
- looks like `i18n` (so user can tell its purpose at a glance)
- but NOT `i18n` (so it's unique enough that we can safely assume its role when we parse/manipulate user's code)
- also NOT too short like '_', or 'i' (for the same reason above)
By these rules, I think `${someChar}18n` is a great fit as the function name, and to simplify things, I just use it as the CLI name and the package name.
2. (the real reason) It's really hard to name a JavaScript i18n lib today
If you search for "i18n" in npm, more that 7000 libs pops up. Here is a list of lib names I searched and found taken:
- simple-i18n
- i18n-simple
- s18n
- v18n
- next-i18n
- i18n-next
- react-i18n
- i18n-react
- ...(maybe 20 more?)
So when I finally found that `a18n` is not taken? -- I know instantly it's a digital treasure that I have to grab :)
1. It aligns well with translation function name and CLI command
I want the translation function to be something:
- looks like `i18n` (so user can tell its purpose at a glance)
- but NOT `i18n` (so it's unique enough that we can safely assume its role when we parse/manipulate user's code)
- also NOT too short like '_', or 'i' (for the same reason above)
By these rules, I think `${someChar}18n` is a great fit as the function name, and to simplify things, I just use it as the CLI name and the package name.
2. (the real reason) It's really hard to name a JavaScript i18n lib today
If you search for "i18n" in npm, more that 7000 libs pops up. Here is a list of lib names I searched and found taken:
- simple-i18n
- i18n-simple
- s18n
- v18n
- next-i18n
- i18n-next
- react-i18n
- i18n-react
- ...(maybe 20 more?)
So when I finally found that `a18n` is not taken? -- I know instantly it's a digital treasure that I have to grab :)
This is like a11y - "accessibility" a[eleven letters]y - https://www.a11yproject.com/
As abbreviations go, a11y is a quiet masterpiece, being simultaneously a numeronym, a homograph, a domain-relevant noun, and an Internet in-joke.
I'd hope that whosoever dreamed it up, went home for lunch and never returned.
With the best will in the world, I'd struggle to rate this a18n in the same ballpark of poetic genius as a11y.
I'd hope that whosoever dreamed it up, went home for lunch and never returned.
With the best will in the world, I'd struggle to rate this a18n in the same ballpark of poetic genius as a11y.
Shameless self-promotion hijack: https://github.com/whyboris/JSON-i18n-Editor
A web interface for letting your team translate JSON i18n files - in case you store your i18n in JSON. See the live demo: link on the GitHub page.
A web interface for letting your team translate JSON i18n files - in case you store your i18n in JSON. See the live demo: link on the GitHub page.
Nice. Being a happy customer of https://poeditor.com I'd like to point out that there's also a service for this out there.
I tried POEditor -- it has nice features that my tool does not. I'd like to think my tool is a good-enough free alternative for many cases.
Great job! A multi-user translate platform is essential to the entire workflow!
Our CMS for JSON files goes into a similar direction: https://frontaid.io/
From what I can tell, this scans code for strings that can be translated, and then you supply translations from the initial language resulting in a mapping of Chinese to other languages (using Chinese as an example).
What happens if you update that initial translation that every other language hooks off? Do you end up with orphans? At the same time, you may not speak all your languages so you may end up with no -equivalent messages too.
What happens if you update that initial translation that every other language hooks off? Do you end up with orphans? At the same time, you may not speak all your languages so you may end up with no -equivalent messages too.
Thank you! This concern is valid. If we change original (let's say) Chinese texts in code, and re-run the CLI (wrap+extract). These happens (in my company's workflow):
1. CLI removes (oldKey -> oldValue) entry from en.json (a18n don't keep stale keys)
2. CLI adds (newKey -> null) entry to en.json
3. Developer commits updated en.json (yeah it's versioned)
4. A script (not in a18n yet) scans en.json, found untranslated text and warns
5. It's up to the developer to decide what to do with missing translation
Not too perfect, but I think keeping original text in code instead of "some.text.id" also has its benefit:
1. Devs understand the texts, this brings more context and makes code easier to understand
2. Some huge projects in my company exists long before they need to support i18n, and it is just too expensive to manually migrate from `text` to `code`
1. CLI removes (oldKey -> oldValue) entry from en.json (a18n don't keep stale keys)
2. CLI adds (newKey -> null) entry to en.json
3. Developer commits updated en.json (yeah it's versioned)
4. A script (not in a18n yet) scans en.json, found untranslated text and warns
5. It's up to the developer to decide what to do with missing translation
Not too perfect, but I think keeping original text in code instead of "some.text.id" also has its benefit:
1. Devs understand the texts, this brings more context and makes code easier to understand
2. Some huge projects in my company exists long before they need to support i18n, and it is just too expensive to manually migrate from `text` to `code`
This is great. As someone who spent a great deal of time working on i18n with JS for my own software the current state of the art is not amazing. A lot of half finished and abandoned projects, many of which don't entirely work or conform to pre-existing standards. Most don't play well with newer web frameworks and those that do tend to be very limited.
Finding a good solution for VueJS, which has it's own .vue files was particularly hard. I ended up building on top an existing i18n extract tool and integrating it with the VueJS compiler to extract both HTML and JS translations.
What I'd like to see, that I have not seen anywhere (other than a broken and not currently maintained Webpack module) is a plugin for Babel or Typescript compilers that allows the compiler to compile translations into the final output packages and produce one for each language. I think this is preferable to loading translations at runtime and swapping them in.
Finding a good solution for VueJS, which has it's own .vue files was particularly hard. I ended up building on top an existing i18n extract tool and integrating it with the VueJS compiler to extract both HTML and JS translations.
What I'd like to see, that I have not seen anywhere (other than a broken and not currently maintained Webpack module) is a plugin for Babel or Typescript compilers that allows the compiler to compile translations into the final output packages and produce one for each language. I think this is preferable to loading translations at runtime and swapping them in.
Will definitely look into .vue support!
But for compile translations (per-language), I'm not so sure if added performance can justify introduced complexity (for users' build/serve workflow).
But for compile translations (per-language), I'm not so sure if added performance can justify introduced complexity (for users' build/serve workflow).
Nice work!
What's the difference between this and .po/.mo files and the ugettext standard?
Update:
As far as I know, this should be a subset of the PO format (http://pology.nedohodnik.net/doc/user/en_US/ch-poformat.html....).
The file format used by a18n is JSON, where its keys are extracted from code (using `a18n extract`) into one of these formats:
- "static text"
- "dynamic text with one variable: %s"
- "dynamic text with multiple variables: %1, %2, %3"
- "#some.text.id" or "xxx#some.text.id" -- user manually assign text an ID with: a18n("xxx", {_: "some.text.id"})
(It worth noting that my company further converts this JSON to Excel before sending to translators)
As far as I know, this should be a subset of the PO format (http://pology.nedohodnik.net/doc/user/en_US/ch-poformat.html....).
The file format used by a18n is JSON, where its keys are extracted from code (using `a18n extract`) into one of these formats:
- "static text"
- "dynamic text with one variable: %s"
- "dynamic text with multiple variables: %1, %2, %3"
- "#some.text.id" or "xxx#some.text.id" -- user manually assign text an ID with: a18n("xxx", {_: "some.text.id"})
(It worth noting that my company further converts this JSON to Excel before sending to translators)
Thank you for asking this question - it lead me to find this library (and I think the quickstart highlights some differences to how the gnu gettext standard could be leveraged in js/ts vs a "propiatary" solution based on json) :
https://ttag.js.org/
https://ttag.js.org/docs/quickstart.html
https://ttag.js.org/
https://ttag.js.org/docs/quickstart.html
Thanks for feedback! It looks like a decent lib and I would recommend anyone who need richer i18n support (like plural forms) take a look.
As for a18n, it's a deliberate choice to use current format (JSON with extracted key/value pair) instead of standard like ICU messageformat. Here are reasons:
1. This lib is built to enable automated i18n workflow for existing projects (~200kloc), with minimal human interfere and impose minimal complexity for build configuration, in this regard current format is working great.
2. This lib is used by both projects, plugins and components in our company, so to keep its footprint down (in case it's not deduped by bundler), it means fewer features (like plural forms, date, list) are supported, and...
3. ...we originally planned to add support for ICU messageformat if these features are desired. But after two years in production, we found the demand is not too strong to justify the cost.
As for a18n, it's a deliberate choice to use current format (JSON with extracted key/value pair) instead of standard like ICU messageformat. Here are reasons:
1. This lib is built to enable automated i18n workflow for existing projects (~200kloc), with minimal human interfere and impose minimal complexity for build configuration, in this regard current format is working great.
2. This lib is used by both projects, plugins and components in our company, so to keep its footprint down (in case it's not deduped by bundler), it means fewer features (like plural forms, date, list) are supported, and...
3. ...we originally planned to add support for ICU messageformat if these features are desired. But after two years in production, we found the demand is not too strong to justify the cost.
Thank you! I'm not familiar with .po/.mo/ugettext, will take a look and answer later.
Is it doing anything besides of localization? If not, why did you name i18n (internationalization is a superset of localization domain).
I get it's a18n like Automated18n like AutomatedI18n.