MANIFEST: Graft labextension and exclude node_modules

[martinRenou]

This PR also commits yarn.lock which should be commited

[mwouts]

Thank you again @martinRenou ! I am so sorry to bother you with this...

Just one thing, I am not sure the exclusion works, at least when I build locally I get again a very large .tar.gz file.
Did you get a different outcome?

I'd prefer not to take you more time on this - in the next days I'll try to understand how the MANIFEST.in works, and also I'll had a few tests on the CI... to check the size of the package, and the presence of the two extensions as we do for the conda package.

[martinRenou]

No worries really :)

Just one thing, I am not sure the exclusion works, at least when I build locally I get again a very large .tar.gz file.
Did you get a different outcome?

I also had this issue. And looking at the logs, I understood that there was cached information under the jupytext.egg-info which was still containing false instructions. I think it was coming from SOURCES.txt.

So if you remove this egg-info directory or clean your local copy with git clean -fdx (be careful this will remove all the file untracked by git, so if you have Notebooks/files that you need there that are not tracked don't run this command) it should work.

[mwouts]

Hi @martinRenou , I've been trying to add a few tests on the CI in this branch: https://github.com/mwouts/jupytext/tree/martinRenou-manifest_graft_labext_exlude_node_modules

I think that the first test does detect node modules in the package.

Maybe we should do what you suggested above, i.e. remove the .egg-info on the CI? Can we build the package in two steps?

Also I'd be interested in having your input for testing presence of the extensions when Jupytext is installed, do you know which folder I could test?

Thanks Martin! 👋

[martinRenou]

I think that the first test does detect node modules in the package.

Indeed. I must have forgotten BUILD_JUPYTERLAB_EXTENSION=1, which meant I had no generated node_modules which might be the reason I didn't see it locally.

Maybe we should do what you suggested above, i.e. remove the .egg-info on the CI?

This will not work. My PR is wrong. I'll do it differently.

[martinRenou]

Also I'd be interested in having your input for testing presence of the extensions when Jupytext is installed, do you know which folder I could test?

We do this kind of thing on some packages. For example in ipycanvas: https://github.com/conda-forge/ipycanvas-feedstock/blob/master/recipe/meta.yaml#L33-L38

[martinRenou]

@mwouts I've updated the PR. It is less elegant, but it seems that MANIFEST didn't like contradictory information.

[mwouts]

Thank you @martinRenou for the update! I'll give it a try very soon, tonight if I can.