Typing is incompatible between patch versions

[krassowski]

If one package (say jupyterlab main branch) has version 3.0.0 and another (say jupyter-collaboration) has version 3.0.2 we see:

    ✖  @jupyter/collaborative-drive:build
       src/tokens.ts(35,18): error TS2430: Interface 'ICollaborativeDrive' incorrectly extends interface 'IDrive'.
         The types of 'sharedModelFactory.registerDocumentFactory' are incompatible between these types.
           Types of parameters 'factory' and 'factory' are incompatible.
             Call signature return types 'YDocument<DocumentChange>' and 'YDocument<DocumentChange>' are incompatible.
               The types of 'ydoc' are incompatible between these types.
                 Type 'Doc' is missing the following properties from type 'Doc': isDestroyed, getXmlElement
       src/tokens.ts(47,18): error TS2430: Interface 'ISharedModelFactory' incorrectly extends interface 'ISharedFactory'.

Maybe we should avoid exposing any objects from ydocument.js? Not sure yet.

Edit: To be fair, we also have this issue with react, which is why we are aggressively pinning.

[krassowski]

Just in case if someone else sees this, the workaround is to add "@jupyter/ydoc": "3.0.0" to resolutions in package.json

[davidbrochart]

Isn't it because the two @jupyter/ydoc versions are using different versions of yjs?

[davidbrochart]

In particular:

• jupyter-ydoc v3.0.2 has yjs v13.5.44.
• jupyter-collaboration v3.1.0 has yjs v13.6.20.

[krassowski]

Yup. The question is how (if) we can make the developer experience better. If it was in our own code usually the answer is "use an interface which has to remain compatible across minor versions". Since this is in a third-party package, I am not sure if we can do much other than synchronizing versions across our packages more consistently.

[davidbrochart]

Actually I'm not familiar with how package resolution works in JupyterLab. Isn't there a way to say that yjs must not be bundled with @jupyter/ydoc nor jupyter-collaboration, since it should only be bundled with JupyterLab, so that only one version exists?