codegen: Support for allOf intersections

[yogyam]

Description

Hey! I've been testing the new codegen feature (great work on this, btw!) and ran into an internal error when using allOf with multiple schemas to create intersection types.

Environment

• CLI Version: 14.1.0
• OS: macOS 15.5
• Build: From source using make
• Target: TypeScript

Minimal Reproduction Schema

{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "type": "object",
  "allOf": [
    {
      "properties": {
        "name": { "type": "string" }
      }
    },
    {
      "properties": {
        "age": { "type": "integer" }
      }
    }
  ]
}

Command

jsonschema codegen schema.json --target typescript --name Person

Expected Behavior

Should generate a TypeScript interface combining both schemas:

{
  "error": "Rule condition holds after application: unnecessary_allof_wrapper"
}

The unnecessary_allof_wrapper optimization rule appears to be triggering incorrectly when allOf contains multiple schemas. The rule is designed to unwrap single-element allOf wrappers (e.g., { "allOf": [X] } → X), but the post-condition check seems to fail when processing actual intersection types.

Offer to Help

I'd be happy to work on a PR for this if you can point me in the right direction! I've already built the project locally and have test cases ready. Let me know if there's a specific approach you'd prefer for handling allOf intersections in the TypeScript codegen.

[jviotti]

Hey @yogyam ,

Thanks for reporting. As explained on Slack, the "core" of the Codegen project is the canonicaliser. This Codegen project just adds a thin layer around it. The bug you raised is very relevant, and it belongs to the canonicaliser.

If you want to help:

• The canonicaliser lives in https://github.com/sourcemeta/core/tree/main/src/extension/alterschema
• The rule the error points out is https://github.com/sourcemeta/core/blob/main/src/extension/alterschema/common/unnecessary_allof_wrapper.h
• The canonicaliser tests are split by dialect here: https://github.com/sourcemeta/core/tree/main/test/alterschema

In the case you report, I would have expected the canonicaliser to convert:

{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "type": "object",
  "allOf": [
    {
      "properties": {
        "name": { "type": "string" }
      }
    },
    {
      "properties": {
        "age": { "type": "integer" }
      }
    }
  ]
}

To:

{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "type": "object",
  "properties": {
    "name": { "type": "string" },
    "age": { "type": "integer" }
  }
}

Which the Codegen project would have been able to handle out of the box.

Overall, you probably see the idea: The codegen project remains simple. The canonicaliser handles most complexity of JSON Schema by converting into a simpler subset. As such, like 90% of the work is actually on the canonicaliser :)

[RitoG09]

Hi @jviotti , My understanding is that this behavior should be handled entirely by the canonicaliser and that codegen is just surfacing the issue.
I’d be happy to work on this in sourcemeta/core. My plan is to:

• Add a failing canonicaliser test for allOf intersections of object schemas under the 2020-12 dialect.
• Investigate the unnecessary_allof_wrapper rule and its pre/post-conditions, which seem to be triggering incorrectly when allOf contains multiple compatible subschemas.
• Update the canonicaliser so that compatible object subschemas in allOf are merged into a single object schema (as in your expected transformation) instead of hitting the failure message.

Should i open a different issue on the core repo referencing this issue and start working on that?

[jviotti]

No need to open new issues on Core. The rest is correct! The intention is for the canonicaliser to be able to collapse EVERY case of allOf. The only case where allOf has a use and it's harder to collapse is if it has 2 or more branches with $ref (unless you inline the references). Otherwise, the canonicaliser should be able to always get rid of allOf