Zod schema validation fails on union type with object schemas

[salmanrf]

Hi, i've been struggling to figure out why my zod schema doesn't work,

I'm using langchain's StrudturedOutputParser which uses zod internally,

The following is my schema:

import { z } from 'zod';

const PROPOSAL_CONTENT_TYPE = ['header', 'list', 'table', 'text'];

const TEXT_CONTENT_SCHEMA = z.object({
  type: z.string(),
  text: z.string().describe('Regular text content'),
});

const HEADER_CONTENT_SCHEMA = z.object({
  type: z.string(),
  size: z.number().describe('Header size: 1-3 (largest to smallest)'),
  text: z.string().describe('Header text'),
});

const LIST_CONTENT_SCHEMA = z.object({
  type: z.string(),
  ordered: z
    .boolean()
    .optional()
    .describe('Whether the list is numbered or not'),
  items: TEXT_CONTENT_SCHEMA.array().describe('Items of the list'),
});

const TABLE_CONTENT_SCHEMA = z.object({
  type: z.string(),
  headers: z
    .string()
    .describe(
      'Table headers must be unique and cannot be empty string, Must be set to appropriate value',
    )
    .array()
    .describe('Common headers shared by all items'),
  items: z.array(
    z.record(
      z.union([
        TEXT_CONTENT_SCHEMA,
        LIST_CONTENT_SCHEMA,
        z.lazy(() => TABLE_CONTENT_SCHEMA),
      ]),
    ),
  ).describe(`
Array of objects representing rows of the table, field names must match the table headers if the items share common fields,
Each field could contain a text, a list, or another table
`),
});

export const PROPOSAL_SCHEMA_STRUCT = z.object({
  code: z.string().describe('Leave empty'),
  version: z.string().optional().describe('Leave empty'),
  title: z.string().describe('Title of the document'),
  date: z.string().describe('Date of the document'),
  company_name: z
    .string()
    .describe('The target company for which the proposal is written'),
  content: z.array(
    z.union([
      HEADER_CONTENT_SCHEMA,
      LIST_CONTENT_SCHEMA,
      TABLE_CONTENT_SCHEMA,
      TEXT_CONTENT_SCHEMA,
    ]),
  )
    .describe(`Array of contents, each content can be one of: ${PROPOSAL_CONTENT_TYPE.toString()}
The type of content determines the schema for that content`),
});

And here is a sample data:

{
  "code": "",
  "version": "",
  "title": "Scope of Supply",
  "date": "",
  "company_name": "",
  "content": [
    {
      "type": "table",
      "headers": ["", "Description", "Unit", "Quantity"],
      "items": [
        {
          "": "1",
          "Description": {
            "type": "table",
            "headers": ["", "Description"],
            "items": [
              {
                "": "",
                "Description": "Integrated Gearbox Pump Set"
              },
              {
                "": "",
                "Description": "Gearbox\nManufacturer: Domestic\nType: Built-in Hydraulic Clutch\nCombined Speed: Main Engine Idle (TBD)\nWorking Speed: 1000rpm\nBV Certificate"
              },
              {
                "": "",
                "Description": "Fire Pump 1500m³/hr, 150m Head, 1870rpm\nPower: 800kw\nMaterial: Pump Body: Ductile Iron\nImpeller: Bronze\nPump Shaft: Stainless Steel\nSeal: Mechanical Seal\nBV Certificate"
              },
              {
                "": "",
                "Description": "High Elastic Coupling (BV Certificate)"
              },
              {
                "": "",
                "Description": "Base"
              }
            ]
          },
          "Unit": "Set",
          "Quantity": "2"
        }
      ]
    }
  ]
}

Validating the above data against my schema fails with the following error:

[
  {
    "code": "invalid_union",
    "unionErrors": [
      {
        "issues": [
          {
            "code": "invalid_type",
            "expected": "number",
            "received": "undefined",
            "path": [
              "content",
              0,
              "size"
            ],
            "message": "Required"
          },
          {
            "code": "invalid_type",
            "expected": "string",
            "received": "undefined",
            "path": [
              "content",
              0,
              "text"
            ],
            "message": "Required"
          }
        ],
        "name": "ZodError"
      },
    ]
  }
]

The actual errors and data are much larger than these, i want to start by understanding the very first error.
As you can see the path specifies content[0].size, which is missing, but the "table" schema doesn't actually require size to be present.

[PaulCordonnier]

@salmanrf why is it closed? Have you resolved this issue?