Why are official plugins installed differently than entrypoint plugins?

[kdmccormick]

Context

There are three types of plugins:

1. DictPlugins: simple YAML files placed at plugin root.
2. EntrypointPlugins: Python projects; when installed, they expose plugin(s) via the tutor.plugin.v0 setup.py entrypoint.
3. OfficialPlugins: Python projects; Tutor loads a hard-coded list of them in main.py.

Why is the OfficialPlugin path necessary? Could Tutor instead just treat the official plugins like any other EntrypointPlugin, using a hard-coded list of plugins to enable by default?

Potential benefits:

• tutor repo users (eg nightly users) get the same plugin experience as pip install tutor users
• plugin loader implementation simplicity

Acceptance

See @regisb's comment bellow explaining why OfficialPlugin is used. See whether EntrypointPlugin could be used at the build step reliably instead.

• If so: Replace OfficialPlugin usage at the build step with EntrypointPlugin usage.
• If not: Add a note to the OfficialPlugin and/ortutor/main.py docstrings explaining their necessity.

[regisb]

A couple assumptions are made about OfficialPlugin instances:

1. The plugin import path is tutor{plugin_name}.plugin.
2. Their version number is defined as tutor{plugin_name}.__about.__version__.

Not all Python plugins generated by the plugin cookiecutter would follow this pattern, so we should not make these assumptions for all plugins.

These assumptions allow us to load these plugins explicitly with OfficialPlugin.load when building the binary bundle with pyinstaller. I forget why I didn't go with EntrypointPlugin, but I believe it has to do with the fact that we cannot reliably iterate on entrypoints during the build step.

[kdmccormick]

This doesn't seem important.