LoroText's Delta updates not correct when pushing null attributes.

[anlumo]

I'm trying to use LoroText together with the Fleather editor. In theory, both support Delta, so this should work fine.

However, certain operations (like changing a line from bullet point to normal line) require removing existing attributes. This is done in Delta by having an attribute with null as the value.

This works fine with LoroText when applying the Delta and then getting a full Delta for the whole text, but when subscribing to changes to the text the change Deltas are incorrect. See my comment below for a unit test that shows this behavior.

Connected Fleather ticket: fleather-editor/fleather#469

[zxch3n]

This seems to work fine:

#[test]
fn test_apply_delta_to_remove_mark() {
    let doc = LoroDoc::new();
    let text = doc.get_text("text");
    text.insert(0, "Hello").unwrap();
    text.mark(0..5, "bold", true).unwrap();
    let delta = text.to_delta();
    assert_eq!(
        delta,
        vec![TextDelta::Insert {
            insert: "Hello".to_string(),
            attributes: Some(fx_map! { "bold".into() => LoroValue::Bool(true) }),
        }]
    );
    text.apply_delta(&[TextDelta::Retain {
        retain: 5,
        attributes: Some(fx_map! { "bold".into() => LoroValue::Null }),
    }])
    .unwrap();
    let delta = text.to_delta();
    assert_eq!(
        delta,
        vec![TextDelta::Insert {
            insert: "Hello".to_string(),
            attributes: None,
        }]
    );
}

Can you provide the code that can reproduce this issue?

[anlumo]

Ok, so I am kinda confused by the behavior, and I'm not sure if Fleather or Loro is at fault, but this test reproduces what I'm seeing:

#[test]
fn test_apply_delta_to_remove_mark_ticket_690() {
    let doc = LoroDoc::new();
    let text = doc.get_text("text");
    let deltas = Arc::new(Mutex::new(Vec::new()));
    let inner_deltas = deltas.clone();
    let _subscription = doc.subscribe(
        &text.id(),
        Arc::new(move |diff_event| {
            let mut deltas = inner_deltas.lock().unwrap();
            for event in diff_event.events {
                deltas.push(event.diff.into_text().unwrap());
            }
        }),
    );
    text.apply_delta(&[TextDelta::Insert {
        insert: "H".to_owned(),
        attributes: None,
    }])
    .unwrap();
    doc.commit();
    text.apply_delta(&[
        TextDelta::Retain {
            retain: 1,
            attributes: None,
        },
        TextDelta::Retain {
            retain: 1,
            attributes: Some(HashMap::from_iter([
                ("bold".to_owned(), LoroValue::Null),
                ("italic".to_owned(), LoroValue::Null),
            ])),
        },
    ])
    .unwrap();
    doc.commit();

    let deltas = &*deltas.lock().unwrap();
    assert_eq!(
        deltas,
        &vec![
            vec![TextDelta::Insert {
                insert: "H".to_string(),
                attributes: None,
            }],
            vec![
                TextDelta::Retain {
                    retain: 1,
                    attributes: None,
                },
                TextDelta::Retain {
                    retain: 1,
                    attributes: Some(HashMap::from_iter([
                        ("bold".to_owned(), LoroValue::Null),
                        ("italic".to_owned(), LoroValue::Null),
                    ])),
                },
            ]
        ]
    );
}

Output:

thread 'api::documents::test_apply_delta_to_remove_mark_ticket_690' panicked at src/api/documents.rs:1900:5:
assertion `left == right` failed
  left: [[Insert { insert: "H", attributes: None }], [Retain { retain: 1, attributes: None }, Insert { insert: "\n", attributes: Some({"italic": Null}) }, Insert { insert: "\n", attributes: Some({"bold": Null}) }]]
 right: [[Insert { insert: "H", attributes: None }], [Retain { retain: 1, attributes: None }, Retain { retain: 1, attributes: Some({"bold": Null, "italic": Null}) }]]

The first weird thing is that Fleather retains 1 character it hasn't inserted yet. I'm not seeing anything about that in the Quill Delta documentation. Apparently LoroText assumes this character to be '\n', which is probably an ok thing to do.

However, the diff event outputs two separate retains for the two attributes. I'm a bit puzzled by this, because this isn't a direct forward of the Delta I'm inserting, but since LoroText knows that removing these non-existing attributes doesn't do anything (due to it having the full data), it could also just not include them if it does some processing.

[zxch3n]

I see. This looks like a different issue, though.

[anlumo]

Yeah, I might have misidentified the problem, because I only ever looked at the subscription data.

[anlumo]

I've updated title and description to match the new findings.