search

hexenq / kuroshiro

Japanese language library for converting Japanese sentence to Hiragana, Katakana or Romaji with furigana and okurigana modes supported.
https://kuroshiro.org
MIT License
804 stars 92 forks source link

async conversion should be sync #50

Open iliakan opened 5 years ago

iliakan commented 5 years ago

Right now, there's totally no point to make convert call async.

More than that, all modules that do similar stuff (limax, slug, slugify) do the conversion synchronously.

Initialization stuff can be async - all right, but the conversion? That makes kuroshiru unusable in many situations where sync method is required.

I looked into the code of https://github.com/hexenq/kuroshiro-analyzer-kuromoji, the conversion may be sync for sure.

Right now I can't use kuroshiro, because another module that needs the slug is 3rd-party and it expects a sync call. Not await. And that's perfectly normal, because all similar modules are sync.

hexenq commented 5 years ago

Hello iliakan,

Admittedly like what you said, kuroshiro-analyzer-kuromoji might be able to be sync. However, when it comes to kuroshiro-analyzer-mecab and kuroshiro-analyzer-yahoo-webapi , web or device I/O make them hard to be sync. To reach a compromise and offer a unified explicitly api, Kuroshiro uses async/await now.

See if we can find some workarounds.

iliakan commented 5 years ago

Right now all similar modules for other languages (including Chinese) use sync API. That's kind of a standard.

So people who add Japanese to their website, like me, get surprised and upset, because they can't use async API (all enclosing functions are sync, hard to rewrite, especially 3rd-party stuff).

If you'd want to make your tool more powerful and it requires async API - fine, but please make it optional.

The main API that most people will use should be sync, as expected.

hexenq commented 5 years ago

Thanks for the explanation. I understand the pain point. Could you name some of the modules you mentioned? I'd like to look a little deeper first.

iliakan commented 5 years ago

For instance, https://www.npmjs.com/package/slugify, a module with 149k weekly downloads, or this one with 198k https://www.npmjs.com/package/slug. These seem pretty popular.

And other modules labelled slug https://www.npmjs.com/search?q=keywords:slug.

iliakan commented 5 years ago

That's for Chinese (includes dictionaries I suppose) https://github.com/hotoo/pinyin

micky2be commented 4 years ago

Same issue here. Using kuroshiro-analyzer-kuromoji should have an option for a synchronous option. Very difficult to have a reactive search result ATM

lvl99 commented 4 years ago

I had a look into turning kuroshiro sync, and it seems that the callback on kuromoji (via using kuroshiro-analyzer-kuromojii) has a slight delay when loading the dictionary files during building which meant sometimes between the time I initialise the analyser and then do the first parse, it threw an error due to the analyser not being loaded.

At the moment I have kuroshiro.convert() and kuroshiroAnalyzer.parse(), except leaving kuroshiro.init() and kuroshiroAnalyzer.init() as sync, which is probably necessary since @hexenq mentioned the other analysers need to be async as well:

For the moment I've done the work purely for feasibility -- I wouldn't advocate for what I've done to be merged into the existing library, but I'd hope for maybe a convertSync and parseSync option to allow for the existing async convert/parse to not break existing systems.

sudofox commented 3 years ago

If this was sync, I could use it properly in Vue.