Incorrect caretPosition on a first keyboard activation after showing an element about rmlui

[mikke89]

I see, so if I understand correctly, you are unhiding the element, and then manually focusing on it right after. I believe the ElementDocument::Show does the right thing: Updating the layout before the autofocus takes place. However, when you do this manually on some element, this doesn't take place automatically.

I would say the proper approach for users in this case is to manually call ElementDocument::UpdateDocument() right after showing the element. This should take care of the layout before you call focus. You can read more about this here: https://mikke89.github.io/RmlUiDoc/pages/cpp_manual/documents.html#manually-updating-the-document

I've generally delegated the responsibility of updating the layout in special cases like these to the user, mainly due to the combination of: (1) figuring out exactly what needs to be updated based on which changes can be very complicated in the CSS/HTML model, (2) a correct but widely applied dirtying/re-layouting system can have extreme performance implications, (3) a conservative dirtying/re-layouting system can be faster but will likely have many false negatives (unacceptable) unless serious effort is spent on this.

[mikke89]

Glad to hear you got it working. I agree it's not ideal, but I believe it's the right trade-off for us right now. In the long-term it would be nice to look at handling such cases automatically, but that is still far off.

[ShawnCZek]

Thank you for the quick and detailed response. I thought the UpdateDocument method was private for some reason. Anyway, replicating the behavior of the ElementDocument::Show you brought up fixed the issue for me.

The only problem is that a data variable managed the state. Now, I have to manage the state programmatically as data models are processed by the context and not by the document, of course.

Regarding the library, I understand and quite agree with how things are right now. But then the obvious problem is when some events are queued for the context update, and other events are processed immediately. There is no silver bullet for this, sadly.