Use IDrive instead of file browser

[juntyr]

Maybe this helps with #28?

[github-actions]

👈 Launch a Binder on branch juntyr/litegitpuller/idrive

[juntyr]

@jtpio What do you think (this is the first step towards #31)?

[jtpio]

Thanks @juntyr for looking into this.

cc @brichet who commented on the original issue and may want to have a look.

Looks like some of the CI failures are not related.

[brichet]

Thanks @juntyr.
#32 should fix the CI failure not related to this PR.

[juntyr]

@brichet How do I update the lockfile?

[brichet]

Running jlpm install should update the lock file locally, it just need to be committed.

[brichet]

Also, for consistency (and , if may worth bumping all the @jupyterlab dependencies all together (@jupyterlab/application and @jupyterlab/coreutils).

And running jlpm dlx yarn-berry-deduplicate -s fewerHighest may help avoiding duplicate dependencies in different version (it's often helpful for @jupyterlab and @lumino dependencies).

[juntyr]

I don't know how the upload handles e.g. mimetypes so I tried to leave as much of the guessing up to the newUntitled method by giving it the file extension. However, this approach is likely less efficient, so I'd be happy for insights from @jtpio and @brichet

[brichet]

I wonder if we need to use a newUntilted, can't we directly use the save method ?

And maybe we can guess if it is a text or bytes mimetype using something like the following (suggestion from chatGPT, not sure if it misses some case):

function isTextMimeType(mime: string): boolean {
  const textTypes = [
    "application/json",
    "application/xml",
    "application/javascript",
    "application/x-www-form-urlencoded",
    "image/svg+xml"
  ];

  return (
    mime.startsWith("text/") ||
    textTypes.includes(mime) ||
    mime.endsWith("+json") || // e.g. application/ld+json
    mime.endsWith("+xml")
  );
}

I tested successfully the following implementation, with the example from RTD (which contains only notebook, image and text files).

    const isText = isTextMimeType(type);
    const content = isText ? await blob.text() : await blobToBase64(blob);
    await this._drive.save(filePath, {
      content: content,
      type: 'file',
      mimetype: type,
      format: isText ? 'text' : 'base64'
    });

[juntyr]

This is the logic that the JupyterLite browser storage drive uses: https://github.com/jupyterlite/jupyterlite/blob/b0f84a44894e3edc554bb0e03f193c44d05f20ff/packages/services/src/contents/drive.ts#L398-L431

I fear that we'd have to duplicate this logic exactly, and the current approach circumvents this code duplication

[juntyr]

The RTD build now fails to load the extension, so something is broken :(

[juntyr]

The RTD build fails to load the extension with

[Error] Plugin '@jupyterlite/litegitpuller:plugin' failed to activate.
	(anonymous function) (jlab_core.cb9a852.js:1:1404219)
[Error] TypeError: undefined is not an object (evaluating 'e.name') — index.es6.js:1933
	(anonymous function) (jlab_core.cb9a852.js:1:1404270)

Do you have any idea about what might be wrong?