Reload the current notebook when it is modified on disk or in the text editor

[mwouts]

Hi @mahendrapaipuri , I see that the CSV viewer in Jupyter appears to be connected to the text editor - I mean, when I edit a CSV file the changes appear shortly afterwards in the table view:

Do you know how that works? Do you think we might be able to achieve something like this to connect the editor of a .py file to the notebook view on the same file? Or on the paired .ipynb notebook?

If we could achieve a kind of auto-reload that would be more user-friendly than the current need to "reload notebook from disk" manipulation. See also #406

[mahendrapaipuri]

Hello @mwouts

No, I am aware of this actually! I agree that this would be an ideal behavior for Jupytext as well. I ll look into it and see if we can implement a similar feature!

[mwouts]

As discussed on linkedin with @jtpio , I believe that these two-editor synchronisation would make it way easier to use paired notebooks.

Until we get this I think the simplest way to use paired notebooks is to close the files that we are not editing, but that's not ideal. It is also possible to deactivate the auto-save and to reload the view that is outdated but that requires more effort. Having an automated reload or update would be way more user-friendly.

[mwouts]

Hi @mahendrapaipuri , hope you're well ! I was wondering if you could help me locate in the Jupyter lab code where this two editor synchronization is happening (for the CSV file) ? Thanks !

[mahendrapaipuri]

Hello @mwouts , I am doing alright. Thanks. How about you?

I did have a quick look at the source code but I could not find the logic that they are using to achieve it for CSV file. I will have some time this week and I will look again and see if I can find any clues.

[mwouts]

Thank you @mahendrapaipuri ! Here traveling and starting a busy semester but all fine. For Jupytext I would really love to improve the editing experience and at least I expect to improve the autoreload in VS code with Jupytext v1.18 (just starting)

Jeremy, if you see someone who could help us with identifying what makes this happen for the CSV editor, could you put us in touch ? Thanks !

[mahendrapaipuri]

Hello @mwouts

So, I took a look into the CSV file editing. If you actually look into it, the only way to edit a CSV file is by opening as text file and doing regular edits. The left side view that you have in the screencast is the "render" of the text file in "CSV format". Moreover it is the same file in "edit" mode and "render" mode. As soon as you save the file in edit mode, file gets saved on the disk and eventually the render changes. As far as I understand, there is no special logic to make it happen on CSV extension.

In the case of Jupytext, we are talking about two different "editable" files: a notebook file and a Python file. So, bi-directional changes possible which is not the case in CSV viewer (It is uni-directional always from edit -> render). That is why we get the dialogue "File has been changed on the disk since last time it was opened". This is generic dialogue that can happen, for instance, when a file is opened in JupyterLab and the same file has been modified directly on the disk. So, I guess we cannot turn off this dialogue just for Jupytext and make the reload automatic.

@jtpio Do you know if there is any way to achieve what @mwouts is looking for? Cheers!

[mwouts]

Thank you @mahendrapaipuri for looking into it! That's very useful! The part that I want to learn more about is the following: how is the CSV renderer informed that the file has changed?

Yes I see what you mean about the current reload message about notebooks. I would say that the fact that a notebook is made of two files is not that much of a problem (for instance, the current manager reports the most recent modification file of these two to Jupyter).

Something else I notice about the current reload message for notebook is that currently it happens only on saves (explicit or auto-saves). I would like Jupyter to check more often and reload the document if it has changed on disk only (i.e. no edit was made in the editor).

[mahendrapaipuri]

how is the CSV renderer informed that the file has changed?

A signal must be emitted every time the CSV file has been modified and using this signal the frontend renderer makes an API request to get latest contents of the file and updates the CSV table. This is entirely happening in the browser without any modifications on the backend

I would like Jupyter to check more often and reload the document if it has changed on disk only (i.e. no edit was made in the editor).

AFAIK this is not possible with current Jupyter server architecture. Jupyter server is API based which means when user "opens" a file, an API request is fired and based on the response the frontend will "draw" the notebook file contents. What you are suggesting is that the Jupyter server continuously "redraw" notebook file contents on the frontend based on changes made to backend file. This needs a protocol like websockets and a reimplementation of contents manager frontend component which is a huge work.

Maybe core JupyterLab maintainers might have more ideas on how to achieve it!!

[mwouts]

Thank you @mahendrapaipuri ! Yes good point, let's see with the core JupyterLab maintainer what they think. I have opened jupyterlab/jupyterlab#17888, let's see how it goes!