Document models have two sources of truth (widget context & widget content)

[dlqqq]

Description

While researching #76 and #93, I found the source code on what sets the dirty state.

Most "top-level" widgets (that get a dedicated tab in JupyterLab) inherit from the DocumentWidget class, which implements this method:

/**
 * Handle the dirty state of the context model.
 */
private _handleDirtyState(): void {
  if (
    this.context.model.dirty &&
    !this.title.className.includes(DIRTY_CLASS)
  ) {
    this.title.className += ` ${DIRTY_CLASS}`;
  } else {
    this.title.className = this.title.className.replace(DIRTY_CLASS, '');
  }
}

The issue is that the document model is stored in two places: this.context.model and this.content.model. We only update this.content.model, but the dirty indicator reads from this.context.model. This is because this.context: DocumentRegistry.IContext does not provide a way to update the document model.

We ought to understand:

• What is document context, and why does it have a separate reference to the model?
• How can we stick to a single source of truth for the document model?

[dlqqq]

Inside of Notebook, you can use this.parent to access the NotebookPanel, so it can set this.parent.context._model manually to synchronize the context & content models.

However, this may have side-effects, so it seems reasonable to keep this issue open until we determine whether we need to update how doc context is handled.

[dlqqq]

#93 was reworked to not reset the document model. This is much safer as generally other extensions do not expect the model to change across its lifetime. This issue is no longer relevant.