search

cfinke / Typo.js

A client-side JavaScript spellchecker that uses Hunspell-style dictionaries.
Other
501 stars 110 forks source link

Two issues I deal with a lot on tablets + one requested addition #39

Open egervari opened 8 years ago

egervari commented 8 years ago

  1. Typo crashes the application on an iPad whenever you use Typo.js to get suggestions. It sometimes works the first time, but iOS will close the application if you spike memory usage - such as loading the dictionary and doing suggestion searches. I think this really needs to be optimized.

    As a temporary work-around, I went through the entire .dic file and removed all of the words that my application doesn't care about, getting the .dic file down to 37k.

    For example, if you know that a word has no rules, then dictionaryTable does not need to contain so many empty arrays, or arrays that contain empty arrays. These objects take up memory. Just null everything out. The biggest culprits seem to be the generated words based on the rules in the dictionaryTable. This one modification alone would dramatically reduce memory footprint.

  2. Also, the API needs to be changed so that it's asynchronous and uses promises by default. For the longest time, I had no idea it could be asynchronous and so it was blocking the UI. Upgrading to promises though is just going to be more and more mandatory for libraries. Modern apps in Angular or whatever are all promise-based now, as is most browser databases. So it's actually a bit awkward to see a defaulted synchronous library. I think at this point in time, the people who want to be synchronous should have to jump through hoops, instead of the other way around.
  3. A nice function to have would be stems() that simply accepts a word and returns all of the replacement words for the rules associated with it. For example, if you pass in 'vibrate', it returns something like ["vibrations", "vibration", "vibrating", "vibrates", "vibrated"]. This is very useful for building full text searches. Since you already have the rules and the dictionary, why not make this function public? I see most of the code buried in _parseDIC, but it could be refactored out as a prototype method.
cfinke commented 8 years ago

All good ideas. I'll work on implementation as soon as I have time.

kofifus commented 8 years ago

see https://github.com/cfinke/Typo.js/pull/45 for async version

groe commented 6 years ago

I would be interested in a way to access the stem(s) as well. This would be very useful for client-side search, e.g. with lunr