Settings for controlling if outputs should be saved to disk

[DonJayamanne]

Originally filed here by @rebornix microsoft/vscode-jupyter#4670

I think we should consider adding this feature into VSCode. I've implemented this same feature in Kusto notebooks.
Feels like something generic for VS Code.

Similar to having a cell toolbar position per viewtype, we can have Save Notebook Outputs per viewtype. This can be achieved easily from VS Code by passing an empty array into the serializer, when saving notebooks.

I personally think this would be useful for rest notebooks as well, one might want the output from the REST & GithubIssues Notebook to be saved in the notebook, just like kusto notebooks (some might want the charts/data & some might not).

Thus all notebooks can be considered serializable (including output for sharing) & user choses whether the output is to be persisted on disc or not.

E.g. with .NET team building their notebook format (*.dib) this would have to be impelemented yet again in their extension.

@rebornix /cc

[vscodebot]

(Experimental duplicate detection)
Thanks for submitting this issue. Please also check if it is already covered by an existing one, like:

• Notebooks: pasting notebook JSON into notebook should automatically add cells (#96943)
• New cells are created in the wrong Notebook document (#120641)

[rebornix]

E.g. with .NET team building their notebook format (*.dib) this would have to be impelemented yet again in their extension.

I doubt so, dib is a concatenation of cell inputs so it doesn't save outputs.

---

if outputs should be saved to disk

if we are talking about serializing the content and saving to disk, then it's the responsibility of the content serializer. VS Code has no idea how the outputs will be used to generate the final file format. If we ever implement something in the core, there are many open questions:

• should the content serializer completely drop outputs info in the file, or save them as outputs: []? how would it know it's empty because users disable it or it's simply empty?
• should the content serializer save execution count? Same question as above, how would it know?
• is there any other metadata that can be linked with outputs and should have the same persistence story as outputs?

In Kusto extension, IIRC if users set not to save outputs, then we also make the executionCount transient. Maybe we do the same in Jupyter, but we don't know yet if it's generic enough.

cc @jrieken @tanhakabir @roblourens

[jrieken]

Yeah, I don't believe that it can be done. The closest request we have is support for transient per output-instance, not per notebook type: #120600

[tanhakabir]

I don't have this case where I'm not saving outputs yet (no request yet) but this case might be possible from core by not giving output cell items in NotebookData when the serialize function is called?

So far though I don't have such a feature planned

[andreaschiappacasse]

+1
This would be also super useful in managing version control of .ipynb, in order to avoid committing large outputs/sesntive output which may be generated by the notebook.

[MrNickArcher]

I came looking for this. Sad to see it hasn't got much enthusiasm. I also want it

• Preparing notebooks for teaching without accidentally giving students any pre-run outputs
• Hiding sensitive output and metadata when committing to git repositories

[PiotrCzapla]

It would be best to expose an ability to filter what get's saved, NBDEV that facilitate building python libraries in notebooks in literate programming, has special comment markup to remove only some outputs. I would love to be able to run the same hook that is already implemented for Jupyter in vscode.

[JotaRata]

Does anybody know if this can be made possible by using the new Notebook: Code Actions On Save setting?

[NickCrews]

Workaround: add a git filter so that in the middle of committing, the notebook is intercepted and the outputs are removed before they are saved by git. The outputs are still saved to disk locally, so you can still see them, but when you git push the outputs won't be included, or if you git checkout a different commit, you won't see the outputs there.

See this gist (PS, run git add --renormalize . as a 4th step before your first commit, so that all of your existing notebooks get scrubbed, or you will get heinous confusing merge conflicts later.)

[ndgayan]

This is a big issue when you working on remote SSH with Jupyter Notebook files. During the VSCode's "save" action, a notebook that contains variables/data frames with a large amount of data in the output is written to the remote file. Sometimes this can cause VSCode to crash or wait a few minutes to finish the save action. It would be helpful If you could add an option to disable saving the notebook output to the file and keep it only in the memory during the session.

[DonJayamanne]

This issue has been open for over 2 years and we have received very few votes on this.
Given that there are other appraoch such as clearing outputs upon saving notebooks or the like, I think that is sufficient (instead of providing a built-in solution).