Binder and yarn tweaks

[ferdnyc]

(Separate) commits in this PR:

1. Upgrade jupyterlab in Binder from 3.x to 4.0.2

2. Run 'yarn set version stable' in Binder, to use yarn 3

3. Remove yarn-deduplicate, which is only compatible with yarn 1. (It's unnecessary with yarn 2+, anyway.)

4. Adds packageExtensions: configs to the .yarnrc.yml, to silence warnings about undeclared dependencies in imported packages. (Except for the fundamental incompatibility of istanbul-instrumenter-loader with webpack 5.) See Setting up repo with yarn 3 fails #3797 for more info.

   (Note: There's also a remaining undeclared react dependency in jupyterlab/settingregistry, but adding that to the packageExtensions makes react a dependency of the project as well, which was a road I decided not to go down.)

5. Create a .prettierignore file in the repo root, and add .yarnrc.yml to it.
   Add a comment,

   # prettier-ignore

   to the section of .yarnrc.yml that Yarn will edit on its own. (Using a .prettierignore file didn't work, because the scripting is already passing --ignore-file .gitignore on the command line, and Prettier doesn't support multiple ignore files.)

   Yarn edits the .yarnrc.yml file automatically, and its own style rules conflict with Prettier's. The two are in constant tension if Prettier is also making changes. It's best to consider the file's contents as managed by Yarn.

   Additional # prettier-ignore comments may be needed as the file grows. The comment only disables Prettier for the single block of code immediately following it.

[github-actions]

👈 Launch a binder notebook on branch ferdnyc/ipywidgets/patch-1

[ferdnyc]

Well, that did nothing. 3.6.4 is already preinstalled, so it's kept as-is.

[martinRenou]

which is now a bad thing as ipywidgets wants 4.x.

ipywidgets works with JupyterLab 3 and 4

[ferdnyc]

@martinRenou

Ah! Well, the explicitly not-matching versions of both jupyterlab and yarn seem bad, here...

Checking JupyterLab (assuming JupyterLab >=4)... 3.6.4
Installing and building all yarn packagesyarn install v1.21.1
[1/5] Validating package.json...
error @: The engine "yarn" is incompatible with this module. Expected version ">=3". Got "1.21.1"
error Found incompatible module.
info Visit https://yarnpkg.com/en/docs/cli/install for documentation about this command.

[martinRenou]

Oh yes. It is indeed more subtle and binder may need JupyterLab 4 if it uses a dev version of ipywidgets.

ipywidgets now requires to be built against JupyterLab 4. Once built and published on PyPi/conda-forge, it will work with both JupyterLab 3 and JupyterLab 4.

Before #3752, ipywidgets required to be built against JupyterLab 3, but that does not mean the built and published package did not work with JupyterLab 4 (apart from a couple small issues it is working).

So you're right, we should update the binder setup to use JupyterLab 4 if binder builds its own version from dev.

[martinRenou]

I would be in favor of not using a dev version of ipywidgets on Binder though, and instead use the latest published version on either PyPi or conda-forge.

Because it may be surprising to users to not get the same features in their local setup as in the binder.

[ferdnyc]

@martinRenou

Binder from pull requests should probably build ipywidgets from the PR's source branch, though, no? Otherwise, that automatic comment with a link to a Binder notebook... running ipywidgets without any of the code from the PR... seems not entirely helpful/useful.

[ferdnyc]

Speaking of yarn invocations, apparently with Yarn 2+ it's... well, debatably a good idea, but at the very least "not wrong" (according to them, anyway) to check into the repo either:

1. The Yarn version in use, so that everyone is running the same tools, or
2. The entire dependency cache, iff it's in the form of compressed package archives rather than a giant, exploded node_modules hierarchy. That requires switching to using their pnp linker, instead of the node-modules one that's the default for projects coming from yarn 1.

Not necessities by any stretch, but apparently options possibly worth considering.

[ferdnyc]

The test failure is just a change in font selection, here's the expected:

And the actual:

....Aand, somehow that magically cleared itself up. Weird.

[martinRenou]

Thanks! Sorry for the late review