really good work and the issues raised are valid. Ideally the logic can implemented all in coreActions but currently we can't due to following limitations in the core

• Modifying cell content: right now we will create a text buffer when doing search for every cell and attach TextModel to the cell once it's visible. We should probably expose an ITextBuffer like interface to support rich text editing other than simple set
• Layering: putting the split/join actions in either the NotebookEditor or NotebookViewModel are both fine. Using the latter is better as we then only touch the viewModel/textModel code and the UI will be updated properly automatically. One missing feature is merging undo points. Deleting a cell and then inserting one are two separate undo points now.
  • We need to add pushUndoStop API like what Monaco Editor does, to allow contribs control how operations should be merged
  • Join Cells is a bit difficult as cell operation and cell content change mixed together.

If we are going to implement this feature in an extension, the missing APIs are:

• Shortcut to get TextEditor for a notebook cell if it exist. Currently extensions have to go through all editors and check if their resource URI matches. (but it's not necessarily a bad idea)
• Reveal/Focus cell API. After a cell is split or joined, extensions may need to set selections correctly.
  • For joining two cells, the text editor for the first cell should not change and there should only be a modelContentChange event emitted.

@rebornix Based on our discussion, I reworked the PR so that we now work directly with lines, i.e. string[]. I added a todo in the code where we'd need to make sure we don't have to join the line contents again. This is a bit more complicated but avoids copying content unnecessarily. It also allows to work with positions directly.

Pls merge if you feel that's good enough for now before you rework the cell text buffers.

@kieferrm thanks, it looks great. I'll fix the conflicts and merge.