@subins2000 https://gitlab.gnome.org/guidog/phosh-osk-stub / https://packages.debian.org/experimental/phosh-osk-stub does presage based completion and I've a hack to allow to play with uim for Japanese: https://gitlab.gnome.org/guidog/phosh-osk-stub/-/merge_requests/34 - maybe give varnam a go too ?
@subins2000 I don't understand. We do word suggestions in phosh-osk-stub (see video below). (I had filed that issue you link to 😃 ). What do you think is missing? I was able to integrate with UIM just fine too: See https://social.librem.one/system/media_attachments/files/011/424/006/original/cf3f50be82c07bea.mp4?1668181672
@agx Hi, I made this video to explain more, hopefully it's clear :)
@subins2000 thanks a lot for the video! I think we can do what you want if we plug a completion engine that uses varnam instead of the presarge, uim or fzf based ones. This is presarge based completer. With varnam it would show the results from varnam_transliterate instead.
@subins2000 I had trouble getting govarnam to work but did a quick hack to plug libvarnam (I have no idea if it predicts anything useful though 😃):
@agx This is awesome! Thank you! I can do the plugging of govarnam into it once I figure out how to do the development locally. Can you please share the patch you made (git branch or just a diff) so that I can get started ?
@subins2000 See https://gitlab.gnome.org/guidog/phosh-osk-stub/-/commit/01828f188ec6e647d0ccc294d903b4ada8466e73 - it's really just a quick hack reusing one of the current completers. Would be great if you'd look into plugging govarnam into it as proper shared library.
Currently the different completion engines are compile time (not run time). Ill post an MR to change that soon.
@agx I will take a look into it and get back to you. Thanks for the pointer.
@subins2000 Great! I have split out the varnam code into a separate completer in the meantime and made it use libvarnam (instead of the varnamc script): https://gitlab.gnome.org/guidog/phosh-osk-stub/-/commits/wip/libvarnam . (I won't do futher work here, I just wanted you to have a nicer example).
This should be simpler as a base for you to start.
@agx Varnam would be different in this aspect, it requires word suggestions. Here's how Varnam looks like in Indic Keyboard on Android.
It's more easier to use English -> malayalam/hindi/tamil etc. transliteration to type.
A basic word suggestions display + on-space-click-commit-primary-suggestion etc. is needed, similarly to how ibus works. Varnam has an ibus engine.
So this issue would be the relevant one https://source.puri.sm/Librem5/squeekboard/-/issues/54
Thanks!