Positioning cursors using awareness (`StickyIndex` support)

[krassowski]

While the awareness is exposed by pycrdt, it does not expose the StickyIndex often needed to place collaborators cursors at a relative position.

[krassowski]

It looks like it would be sufficient to expose python binding to sticky_index on Text and Array, do I see it right?

https://github.com/y-crdt/y-crdt/blob/0968e1dd3889df369d3adff60e771632bd754af7/yrs/src/moving.rs#L892-L906

[davidbrochart]

cc @brichet who contributed to the awareness feature in pycrdt.

[krassowski]

Just to share my research into how we could possibly use it downstream (in jupyterlab):

ycell = document.ycells[0]
text: Text = ycell["source"]
document.awareness.set_local_state_field(
    "user",
    {
        "username": "krassowski",
        "initials": "MK",
        "display_name": "M. Krassowski",
        "name": "Krassowski",
        "color": "#ffb61e",
    },
)
# set the cursor in the first cell
cursor = text.sticky_index(index=0, assoc=0)
document.awareness.set_local_state_field(
    "cursors",
    [
        {
            "anchor": cursor,
            "head": cursor,
            "primary": True,
            "empty": True,
        }
    ],
)

Some questions here are:

• should the stick_index object be natively json-serializable when passed to set_local_state_field or have to_py() method?
• do we need to pass doc.transact_mut() as in y-crdt or can it be accessed only at the bindings level (as implied by my pseudocode above)

[brichet]

I wonder what would be the use case for cursors position from the server side document awareness. Is it intended for LLM use?

[krassowski]

Is it intended for LLM use?

Yes, spot on. Having a collaborator agent on the backend being able to work and continue working even if you refresh the browser, and their cursor position being visible when they type.