Bug: TextNode in Token mode gets split when format is changed on portion of its text

[9still]

Lexical version: 0.9.0

Steps To Reproduce

1. Open a modified MentionsPlugin example that simply uses RichTextEditor instead of PlainTextEditor
2. Create a mention node by entering @Ham and selecting @Hammerhead.
3. Select a portion of the newly created node (e.g. mer), press Cmd/Ctrl+B to bold the selected text
4. Observe that the node that's supposed to be a token gets incorrectly split into multiple nodes

Link to code example:

https://codesandbox.io/s/purple-river-pmc11v

The current behavior

Applying format to a range selection does not respect boundaries of text nodes in token mode.

The expected behavior

RangeSelection's formatText method (and likely other methods) should not invoke splitText on "token" text nodes.

[acywatson]

Good one. I guess the expected behavior here would be for the formatting operation to fail? What do you think?

[9still]

So since the node is supposed to be treated as a token, it feels like anything that happens to a portion of the node should happen to the entire node.

I feel like ideally that would translate to the whole node getting selected when any portion of the node is selected. Basically I think of a token node like I would think of a ligature - it's an atomic thing, so it's weird to be able to select a portion of it.

If extending the selection is not practical for some reason, I think the next best thing would be to format the entire node when a portion of the node is formatted. Seems reasonable to assume that's what the user would intend...

[HarrySIV]

I think it should fail, if you select a row of text that has a hashtag or mention node, you probably want the node to fit a specific style to indicate it's not normal text and not be bolded/italicized/etc. Maybe I'm wrong, but I don't think it's expected to be able to change such a node besides maybe its font size.

[9still]

To me a token text node is still a text node & it should accept any formatting a text node should. I think it's totally fair game to have a bold hashtag or any other token. In-fact that's why I am using a text node in the first place - otherwise I'd just use a decorator node & be done with it.

[AkhrHysd]

I've noticed that TextNodes in Token mode also get split with the insertion of newlines. Ideally, I believe the whole node should be replaced by a newline instead. Could this be considered as part of the fix for this issue?

[cschwebk]

This is still an issue with the latest versions of Lexical. Here is an updated sandbox link. https://codesandbox.io/s/affectionate-microservice-r5r38c

[Maxim1saev]

@9still Hi! I faced a similar problem, did you find a solution then? If not, how was the situation resolved

[9still]

@Maxim1saev unfortunately couldn't figure out how to make it work using text nodes. We ended up using decorator nodes instead & writing custom (de)serializers.

[Maxim1saev]

@9still and how did you get a text node to calculate its format? Deserializers for json ? That is, have you taught the decorator node to include formatting and apply it to the entire text with partial selection?