Integration of all parts

[fcollonval]

This is a major refractoring of the code for integrating the extensions and the navigator as well as publishing them.

• [x] Setup multipackages repo
• [x] Fixes #80
• [x] Fixes #85
• [x] Migrate tests
• [x] Package the nbextension
• [x] Update setup scripts
• [x] Update README
• [x] Update binder
• [ ] Update publication workflow

[fcollonval]

Binder link: https://mybinder.org/v2/gh/mamba-org/jupyter_conda/integration?urlpath=lab

[fcollonval]

@jtpio This is ready for review.

The new structure is as follow:

• All backend code in jupyter_conda
• Frontend code split in 4 parts:
  • packages/common : Common code for the lab extension and the navigator
  • packages/labextension: Lab extension
    Published as npm package in jupyter_conda/labextension
  • packages/navigator: Navigator frontend
    Pubished as static files in jupyter_conda/navigator/static
  • jupyter_conda/nbextension: Classical notebook extension

The navigator can be launch from an entry point navigator or as the python package main python -m jupyter_conda

Questions:

• What naming for the different packages? For now
  • @mamba-org/common
  • @mamba-org/conda-lab
  • @mamba-org/navigator (no public publication)
  • jupyter_conda for Python code
• What logic for versioning and publication?

I think we can discuss that with Sylvain in a call.

[coveralls]

Pull Request Test Coverage Report for Build 316029837

• 0 of 0 changed or added relevant lines in 0 files are covered.
• 13 unchanged lines in 2 files lost coverage.
• Overall coverage increased (+11.6%) to 69.444%

---

Files with Coverage Reduction	New Missed Lines	%
src/index.ts	4	0%
src/CondaEnvWidget.tsx	9	0%
<!--	Total:	13		-->

Totals
Change from base Build 278617331:	11.6%
Covered Lines:	737
Relevant Lines:	971

---

💛 - Coveralls

[fcollonval]

I did a quick analysis of the navigator bundle:

We got the blueprintjs of course... But more surprising is codemirror. I realized that @jupyterlab/application implies codemirror due to @jupyterlab/rendermime. This is used only to render code in markdown (https://github.com/jupyterlab/jupyterlab/blob/2.2.x/packages/rendermime/src/renderers.ts#L809). This makes me wondering if there is a space for modularity improvement in jlab core.

Back to this code, switching from react-virtualized to react-window should really be considered #77.

[jtpio]

Wow thanks a lot for tackling this!

What logic for versioning and publication?

Lab does it by keeping versions in sync in all packages, even if there were no real change. Maybe it could make things simpler that way.

This is used only to render code in markdown (https://github.com/jupyterlab/jupyterlab/blob/2.2.x/packages/rendermime/src/renderers.ts#L809). This makes me wondering if there is a space for modularity improvement in jlab core.

Looks like this could indeed be open as an issue on the core repo (if there isn't one already).

[jtpio]

Looks like the builds fails on Binder with:

[jtpio]

Just tried locally with this change https://github.com/mamba-org/jupyter_conda/pull/91#discussion_r507752583:

diff --git a/packages/navigator/package.json b/packages/navigator/package.json
index 95b519e..28fca2a 100644
--- a/packages/navigator/package.json
+++ b/packages/navigator/package.json
@@ -68,7 +68,7 @@
     "typescript": "~3.9.0",
     "url-loader": "~3.0.0",
     "watch": "~1.0.2",
-    "webpack": "^5.0.0",
+    "webpack": "^4.0.0",
     "webpack-bundle-analyzer": "^3.9.0",
     "webpack-cli": "^3.3.10",
     "whatwg-fetch": "^3.0.0"

And it seems to be working nicely. Thanks again!

[fcollonval]

Looks like the builds fails on Binder with:

Strange I just succeeded to launch it (OVH instance).