refactor(gatsby): Convert inference metadata to TypeScript

[chooban]

Description

I took a more sensible approach than with #23093; I switched off the linting errors about use before define so that an easier side-by-side comparison was possible. I also did the conflict reporter class as migrating the test highlighted the need for it on the IMetadata interface.

I was lax on the types in the test, as some of the input had very little crossover with what I think is the correct Node type. That meant there was a lot of as Node and I had to pass it through as as unknown as well. So there's a minimum of change in the contents of the tests, but I'm not going to say "Wow! Isn't this a wonderful new paradigm!".

References

ref #21995

[vladar]

Thank you for this PR! I've left several comments inline.

Also I think it makes sense to remove those comments (as they are now actually represented as types):

gatsby/packages/gatsby/src/schema/infer/inference-metadata.js

Lines 49 to 112 in 4043129

	### Metadata structure
	```javascript
	type TypeMetadata = {
	total?: number,
	ignored?: boolean,
	ignoredFields?: Set<string>,
	fieldMap?: { [string]: ValueDescriptor },
	typeName?: string,
	dirty?: boolean, // tracks structural changes only
	disabled?: boolean,
	}
	type Count = number
	type NodeId = string
	type ValueDescriptor = {
	int?: TypeInfoNumber,
	float?: TypeInfoNumber,
	date?: TypeInfoDate,
	string?: TypeInfoString,
	boolean?: TypeInfoBoolean,
	array?: TypeInfoArray,
	relatedNode?: TypeInfoRelatedNode,
	relatedNodeList?: TypeInfoRelatedNode,
	object?: TypeInfoObject,
	}
	abstract type TypeInfo = {
	first?: NodeId | void, // Set to undefined if "del"
	total: Count,
	}
	type TypeInfoString = TypeInfo & {
	empty: Count,
	example: string,
	}
	type TypeInfoDate = TypeInfo & {
	example: string,
	}
	type TypeInfoNumber = {
	example: number,
	}
	type TypeInfoBoolean = {
	example: boolean,
	}
	// "dprops" is "descriptor props", which makes it easier to search for than "props"
	type TypeInfoObject = TypeInfo & {
	dprops?: {[name: "number" | "string" | "boolean" | "null" | "date" | "string" | "array" | "object"]: Descriptor},
	}
	type TypeInfoArray = TypeInfo & {
	item: ValueDescriptor,
	}
	type TypeInfoRelatedNode = TypeInfo & {
	nodes: { [NodeId]: Count }
	}
	```

[chooban]

Thanks, @vladar. It's an interesting challenge migrating a codebase you don't know, and trying to not break it! 😆 I'll get on those comments now, and try to keep it to logical commits.

I'll probably edit the code comments at the top of the file last, if that's okay, so I can take a pass at rewriting if needs be but referring to new interface names.

[vladar]

@chooban Sure, you are doing a great job here! Special kudos for making it easy for review 🏆

[chooban]

There's too much teling the compiler what to do here, but otherwise it complains about the lack of a value for empty in the case it's a string field. I could add a small factory function that returns an ITypeInfo matching object based on the typeName. Is that extra code to maintain worth it compared to the quick compiler hint that's covered by the tests?

[vladar]

Yeah, I think it's OK

[chooban]

OK to add the factory function, or to leave it alone?

I've no idea why I've got failing tests now. I can't seem to get master to pass either. :/

[vladar]

Leave it as is for now. As for tests - there were some unrelated issues recently. Should be fixed now. Let's wait a bit when they finish.

[vladar]

OK, I think we are good to go. Ignore failing www tests as those should be fixed in master now. Thanks for the PR! 💜

[chooban]

On to the next one!