Pass-through configuration directly to Thebe

[choldgraf]

Description

For any configuration that exists in Thebe, we should simply pass it through directly to thebe rather than having our own configuration key name for it and translating it.

Benefit

This way we standardize on a single set of configuration keys, and can simply refer to the thebe docs for the full reference.

Implementation

Here is where thebe documents its configuration: https://thebe.readthedocs.io/en/latest/configure.html#

and we translate and configure thebe in this package here:

sphinx-thebe/sphinx_thebe/__init__.py

Lines 124 to 142 in c036ac4

	thebe_html_config = f"""
	<script type="text/x-thebe-config">
	{{
	requestKernel: true,
	binderOptions: {{
	repo: "{org}/{repo}",
	ref: "{branch}",
	}},
	codeMirrorConfig: {{
	theme: "{codemirror_theme}",
	mode: "{cm_language}"
	}},
	kernelOptions: {{
	kernelName: "{kernel_name}",
	path: "{path_to_docs}{str(Path(docname).parent)}"
	}},
	predefinedOutput: true
	}}
	</script>

Note now we are "manually" translating many things into thebe's configuration. Instead of this, we should just accept something like thebe_config, which would be a dictionary that would directly map onto Thebe configuration.

[akhmerov]

Coming here from #19, this seems like the preferred approach; is there anything blocking it?

[choldgraf]

Nope, just hours in the day, I am happy to review a pr 🙂

[akhmerov]

I can see what I can do. What's your preferred approach to backwards compatibility? A separate config variable + deprecation cycle? Moving fast and breaking things? Another option?

[choldgraf]

Good question. Maybe we could add a key for "thebe_config_raw = True" or something, and deprecate over a cycle or two?

[choldgraf]

TBH I'd also be fine just changing it now, as long as there wasn't a strong reason to keep the current behavior

[akhmerov]

My preferred approach would be indeed to change the interface directly. The arguments why it wouldn't be too harmful are:

• v0.0.x should be enough of a warning
• the package users are likely capable of either pinning to the older version or updating to the new one

The only argument against I can see is the requirement of sphinx-thebe>=0.0.6 in jupyter-book.

[choldgraf]

I good point, we should pin jupyter book to ~= 0..0.* first