Optimize conditional uniqueness rules to partial unique indexes

[thgreasi]

Change-type: minor
See: https://balena.fibery.io/Work/Project/961

[thgreasi]

Removed es-toolkit

[Page-]

Suggested change

	booleanNode[1][0] === 'ReferencedField' &&
	booleanNode[2][0] === 'ReferencedField';
	isReferencedFieldNode(booleanNode[1]) &&
	isReferencedFieldNode(booleanNode[2]);

[Page-]

Suggested change

	if (fromNode[1][1][0] !== 'Table') {
	if (!isTableNode(fromNode[1][1])) {

[Page-]

Suggested change

	if (fromNode[1][0] !== 'Alias') {
	if (!isAliasNode(fromNode[1])) {

[Page-]

Suggested change

	const queryChecksAbsenseOfRows =
	const queryChecksAbsenceOfRows =

[Page-]

I think at this point it's probably worth creating a separate file to host individual optimizations (eg schema-optimizations/partial-unique-index.ts) with a shared code folder as well, since reviewing this PR has been very tricky and that's whilst being able to see the things related to this optimization in isolation, looking at it later when it's mixed in with other optimizations would make it even even harder which I don't want to imagine trying

[thgreasi]

Moved as is from src/abstract-sql-schema-optimizer.ts

[thgreasi]

Extracted mostly as is from src/abstract-sql-schema-optimizer.ts

[Page-]

Seeing this now makes me realize it should probably accept a table name to strip explicit references for, as a safety measure in case the ReferencedField is referencing something we don't expect

[thgreasi]

#322

[Page-]

Can this/other cases like this become

Suggested change

	const [targetTableInfos, parentTableInfos] = partition(
	const { true: targetTableInfos, false: parentTableInfos } = groupBy(

?

[thgreasi]

That's how I had it initially, but at the end a didn't like how it looked in the already complex code and wanted to automatically do the ?? [], so that we can have simpler checks from that point onward.
But that's pretty much how I implemented partition() it in utils.ts.
If you insist I can inline it, but I prefer not to, to keep the sources simpler.

[thgreasi]

Oh, more importantly!!!
Having a dedicated partition allowed me to use any return type assertions that the provided fn does, to automatically narrow down the typing of the true group.

export function partition<T, U extends T>(
	entries: T[],
	iteratee: (item: T) => item is U,
): [U[], T[]];

[Page-]

Maybe I'm missing something but I think this can be:

Suggested change

	const result: BooleanTypeNodes[] = [];
	for (const n of nodes) {
	if (Array.isArray(n) && n[0] === 'Exists' && Array.isArray(n[1])) {
	const maybeFieldTypeNode = n[1];
	const fieldName =
	isReferencedFieldNode(maybeFieldTypeNode) &&
	maybeFieldTypeNode[1] === resourceAlias
	? maybeFieldTypeNode[2]
	: undefined;
	if (
	typeof fieldName === 'string' &&
	fieldsByFieldName[fieldName]?.required === true
	) {
	continue;
	}
	}
	result.push(n);
	}
	return result;
	return nodes.map((n) => {
	if (!Array.isArray(n) && n[0] === 'Exists' && Array.isArray(n[1])) {
	const maybeFieldTypeNode = n[1];
	const fieldName =
	isReferencedFieldNode(maybeFieldTypeNode) &&
	maybeFieldTypeNode[1] === resourceAlias
	? maybeFieldTypeNode[2]
	: undefined;
	if (
	typeof fieldName === 'string' &&
	fieldsByFieldName[fieldName]?.required === true
	) {
	return false;
	}
	}
	return true;
	});

(with further simplification via changing where the returns are, but I left it as similar to current as possible for the example)

[thgreasi]

I guess you mean .filter() instead of .map() and you also accidentally added a ! in from of the Array.isArray(n), right?
Yes it seems so 👍