Sanitizing misidentifies diamond-shaped references as cycles

[SkelatorIndy]

This is a very similar issue to #348, but occurs when the same record is referenced from two different paths in a create call.

For example, this code results in the same type of validation error as #348:

import { Collection } from '@msw/data';
import * as v from 'valibot';

const profileSchema = v.object({
 id: v.string(),
 bio: v.string(),
});

const userSchema = v.object({
 id: v.string(),
 name: v.string(),
 get profile() { return profileSchema; },
});

const postSchema = v.object({
 id: v.string(),
 title: v.string(),
 get author() { return userSchema; },
 get editor() { return userSchema; },
});

const profiles = new Collection({ schema: profileSchema });
const users = new Collection({ schema: userSchema });
const posts = new Collection({ schema: postSchema });

users.defineRelations({ one }) => ({ profile: one(profiles) }));
posts.defineRelations({ one }) => ({
 author: one(users),
 editor: one(users),
}));

const profile = await profiles.create({ id: 'p1', bio: 'hi' });
const user = await users.create({ id: 'u1', name: 'Alice', profile });

// Diamond: same user referenced from two sibling fields
await posts.create({
 id: 'post1',
 title: 'Hello',
 author: user,
 editor: user,
});

In this setup, either author.profile or editor.profile will be undefined after sanitization.

From my testing, I've been able to fix the error by adding the following to the inner sanitize method within the #sanitizeInitialValues method:

const result = Object.fromEntries(...);

if (record && !isRevisit) {
 visited.delete(record[kPrimaryKey]);
}

return result;

[hiSandog]

Removing a record from visited when its traversal returns matches path-local cycle detection, but the key may need collection identity as well as the primary-key value. Otherwise, unrelated records in different collections that share an ID could still be mistaken for the same node. Tests for a true cycle, a diamond reference, and equal IDs across two collections would define the intended traversal semantics.

[kettanaito]

Thanks for reporting this @SkelatorIndy. Honestly, I'd very much like to remove sanitization entirely (see #342). I hope I'll find a way around it eventually. But yeah, if this fails on the latest release, I say we add a test for it and fix it, especially if it seems like a trivial fix.