fix: type single relation/select/file fields as single, not arrays

[Karnak19]

Closes #28.

Problem

PocketBase stores single-value relation, select, and file fields with maxSelect: 0 — not 1. The generators only treated maxSelect === 1 as single, so every single relation/select/file field was emitted as an array:

// before — owner is a single relation in PocketBase
export type AssessmentBlocksRecord = BaseRecord & {
  owner?: string[] | null   // ❌ should be string
}

Thanks to @-the reporter for the precise diagnosis in the issue.

Fix

Follow PocketBase's own semantics — a field is multiple only when maxSelect > 1 (matching core's IsMultiple()), so 0 and 1 are both single. Centralized in a shared isMultipleField helper to stop the rule from drifting across call sites:

• type-generator/generate.ts — select, relation, file
• schema-parser/normalize.ts — extractRelations multiple flag (also affected expand-path arity)
• pbkit-zod — select, relation, file schemas

// after
export type AssessmentBlocksRecord = BaseRecord & {
  owner?: string | null
}

Tests

• Added regression tests for maxSelect: 0 (relation, select, file) in the type generator
• Full chain green: bun run ci (lint, typecheck, test, build) — 13/13 turbo tasks pass

Release

Patch changeset added for @karnak19/pbkit and @karnak19/pbkit-zod. The TanStack plugin is unaffected (it derives from generated types, not maxSelect).

🤖 Generated with Claude Code

[github-actions]

This PR correctly fixes the maxSelect: 0 → single-value typing bug across the type generator, relation extraction, and Zod plugin, and the shared isMultipleField helper is a clean abstraction. One test-coverage gap noted below.

• Missing Zod regression test: the type generator has explicit maxSelect: 0 tests, but the Zod plugin does not — the fixture only uses maxSelect: 1 for single fields, so a future regression in fieldToZod would go undetected.

[github-actions]

Missing regression test for maxSelect: 0 in the Zod plugin.

The type generator gains three explicit maxSelect: 0 tests in type-generator.test.ts, but the Zod plugin's zod.test.ts has no equivalent. The test fixture (full-schema.json) uses maxSelect: 1 for every single-value relation/select/file field, so the snapshot and assertion tests never exercise the maxSelect: 0 path that this PR fixes.

Impact: if someone later replaces isMultipleField(field) with a direct maxSelect === 1 check (re-introducing the original bug), CI will not catch it in the Zod output.

Fix: either add a maxSelect: 0 field to packages/pbkit-zod/src/test/fixtures/full-schema.json (and update the snapshot), or add a targeted unit test similar to the type generator's:

test("relation maxSelect=0 → z.string() (not array)", () => {
  const schema = parseJson({ collections: [{ name: "x", type: "base", fields: [{ name: "ref", type: "relation", system: false, maxSelect: 0, collectionId: "abc" }] }] });
  const output = generateZod(schema, { ir: schema, typesImport: "./t", sdkImport: "./s" });
  expect(output).toContain("ref: z.string()");
});

[Karnak19]

Good catch — added in c050a63. New regression test in zod.test.ts builds a collection with maxSelect: 0 relation/select/file fields and asserts they generate z.string() / z.enum([...]) (not z.array(...)), so a future regression in fieldToZod would now fail CI.

[github-actions]

No new major issues found. The isMultipleField helper correctly centralizes the maxSelect > 1 rule, the fall-through in relation/file cases is clean, and both packages now have maxSelect: 0 regression tests.