I think it requires a lot of efforts to achieve this, including extensions of existing text-input protocol.

At least, the text-input-v1 protocol (the IME protocol I brought to components/exo) doesn't support providing bounds for the virtual keyboard, which needs to be added by someone else.

Does it have to be text input protocol? or can it be separate vk management protocol?

Of course we could provide our own extension for virtual keyboards. I'm just saying that the existing protocol (used for Crostini) wouldn't be enough for ARC++.

I've chatted shortly with Scott about slightly related topic and it is also unclear for me whether we should move more to wayland for ARC++ or we should rely more on Mojo.

I'm inclined to think we should use the mojom for this. Is there a reason not to do that? If we need versioning, mojom does support it.

We currently have two ime protocols, mojo for arc++ and wayland for crostini. It'd be nice if we have to maintain just one, not two.

We may even have three. Is there a reason we couldn't use the mojo one everywhere?

At least we need to keep wayland for Crostini, as long as wayland is the only way to communicate with Crostini VM (can Sommelier support mojo eventually? Is it even possible?)