fix: Handle anyOf in JSON Schema property

[eecavanna]

Fixes #378

[eecavanna]

I've drafted this function, but haven't tested it yet. I'm trying to get unit tests or doctests running locally.

[dwinston]

Grep the codebase for additional uses of “$ref” in .py files. I think there are a few other instances that we want to convert to more robust access.

[eecavanna]

I will do that and add fixes to this PR.

---

Update: That's turning out to be a bit of a rabbit hole (since some things, and the things that use those things, were written under the assumption that a collection always maps to at most one class).

[eecavanna]

@dwinston, I don't know how I can adapt this statement to work in the new "a collection can be associated with multiple classes" situation. It's within a 175-line function in the changesheet code.

nmdc-runtime/nmdc_runtime/api/core/metadata.py

Lines 171 to 172 in 1eacc43

	else:
	class_name = class_name_dict[collection_name_to_class_name[collection_name]]

In this PR branch, I have arbitrarily chosen to use the first class in the list of classes. 🤷

        else:
            # FIXME: As of nmdc-schema v9.1.0 (maybe earlier), a collection name can map to multiple class names.
            #        In the case of multiple class names, which one do you want to assign to `class_name`?
            #        Here, I have arbitrarily chosen to use the first class name in the list.
            class_names = collection_name_to_class_names[collection_name]
            class_name = class_name_dict[class_names[0]]

Source: https://github.com/microbiomedata/nmdc-runtime/pull/379/files#diff-a981e2387918aa46e3575eddf8d3b3efd43dc78ca8be2a03f3b974e3887d6368R171-R176

[eecavanna]

@dwinston, I adapted this function and the associated search.html page, to work in the new "a collection can be associated with multiple classes" situation; with the caveat that the web page will ignore any classes after the first one.

nmdc-runtime/nmdc_runtime/api/endpoints/find.py

Lines 190 to 222 in 1eacc43

	def documentation_links(jsonschema_dict, collection_names):
	rv = {"Activity": []}
	for cn in collection_names:
	last_part = jsonschema_dict["$defs"]["Database"]["properties"][cn]["items"][
	"$ref"
	].split("/")[-1]
	entity_attrs = list(jsonschema_dict["$defs"][last_part]["properties"])
	if last_part in ("Biosample", "Study", "DataObject"):
	assoc_path = [cn]
	else:
	assoc_path = ["activity_set", cn]
	rv = assoc_in(
	rv,
	assoc_path,
	{
	"collection_name": cn,
	"entity_url": "https://microbiomedata.github.io/nmdc-schema/"
	+ last_part,
	"entity_name": last_part,
	"entity_attrs": sorted(
	[
	{
	"url": f"https://microbiomedata.github.io/nmdc-schema/{a}",
	"attr_name": a,
	}
	for a in entity_attrs
	],
	key=itemgetter("attr_name"),
	),
	},
	)
	return rv

[eecavanna]

Hi @dwinston, I finished making all the changes to this branch that I expect to make. I want to know what you think about what I said in these two comments:

• fix: Handle anyOf in JSON Schema property #379 (comment)
• fix: Handle anyOf in JSON Schema property #379 (comment)

I'm OK with you making additional commits to this branch.

[dwinston]

thanks for tackling this, @eecavanna. I don't love the added complexity. I'd like to try and identify a more elegant solution than this band-aid, realizing that this bug is blocking folks, so if I can't identify a better route soon, I'll go with this (after careful review). Again, thank you.

[dwinston]

I can't think of a more straightforward way to handle the appearance of one-to-many collection-to-class relationships, and I think we should shoot for straightforward for now.