Wrong Korean behavior

[tenextractor]

The Korean keyboard has wrong behavior, for example:
Right behavior: ㅅㅏㄹㅏㅁ -> 사람
Unexpected Keyboard: ㅅㅏ[space]ㄹㅏㅁ -> 사람

We need an interface based on composing text to compose characters like this.

[chlwnsgud7]

Hi @Julow,

I've been reading through #558 and #815(this), and also looking through KeyModifier.java / KeyValue.java to understand how Hangul composition currently works.

From what I can tell, Hangul is handled as a special case in the codebase: dedicated Hangul_initial/Hangul_medial Kinds with hand-written combination logic in KeyModifier, separate from how every other script is handled (the generic Compose_pending + ComposeKeyData state-table system used for accents etc). That generic system is a single-step deterministic lookup, so it can't express the "wait for the next keystroke before deciding what the previous one was" behavior that standard Dubeolsik typing needs (e.g. a trailing consonant being reassigned to the next syllable's initial when a vowel follows).

I'm interested in working on this, but before attempting anything I wanted to check: does this match your understanding of where the difficulty lies, and is this something you'd want fixed at all, or is the current swipe/space-based approach intentional?

Happy to dig in further once I know if this is a direction you're open to.

[Julow]

Sorry for the slow reply!
There's #1275 that I couldn't review yet due to a lack of time. I expect that it implements the behavior you are interested in ?

That generic system is a single-step deterministic lookup, so it can't express the "wait for the next keystroke before deciding what the previous one was" behavior that standard Dubeolsik typing needs

That's right, we need a system that can rewrite the character that was already typed. (or a system that can hold inputs for longer but I don't think that's a good UX as well)

[chlwnsgud7]

Thanks for pointing me towards #1275! I've tested it, and the basic Dubeolsik behaviour works well: resyllabification, backspace and compound vowels all work as expected. I'll leave some review comments on the issues I noticed there.

[tenextractor]

Suggestion: please implement an interface similar to the AOSP keyboard's Combiner (example) so that existing combiners already written for this problem may be reused.