Generating arktype definitions from an external source

[VanTanev]

Hello!

I am looking at generating arktype definitions from a Solr schema. Right now the best path I could figure out has been to generate a JSON Schema from the Solr schema, and then use @sinclair/typebox-codegen to generate arktype.

Is there a better approach I might be missing?

[ssalbdivad]

We're publishing @ark/json-schema today which should simplify the process quite a bit!

[VanTanev]

Would that allow to actually do codegen? I ultimately want to output a .ts file

[ssalbdivad]

What is the advantage of codegen if you already have an external source of truth?

It will allow you to do something similar to:

export const TypeA = parseJsonSchema(typeASchema)
export const TypeB = parseJsonSchema(typeBSchema)

And then you can use TypeA is if you had defined it directly.

@ark/json-schema was thoroughly reviewed and tested so I have a much higher degree of confidence in the output.

[VanTanev]

The external source of truth is an XML file. I could generate and write an intermediate JSON Schema file, but if I'm doing file generation either way, would rather generate arktype code directly.

[ssalbdivad]

I don't know if there are projects out there besides @sinclair/typebox-codegen that generate ArkType types.

Regardless, I would still recommend an approach like I described because...

• @ark/json-schema is guaranteed to be exactly 1-1 by arktype. If you've tested @sinclair/typebox-codegen and it works well for your needs, I'm very glad, but I can't make any guarantees.

• If you are already running codegen, you should generate TS types directly from your JSON schema as well and then use those to cast the corresponding Types:

import { TypeA } from "./generatedTypes.ts"

export const TypeA = parseJsonSchema(typeASchema).as<TypeA>()

This saves a bunch of unnecessary tsserver overhead validating and inferring autogenerated definitions.

[VanTanev]

Thanks for your directions!

For right now, it seems @sinclair/typebox-codegen works well for my needs. And yeah, even with it, I could still generate the TS types and do the casting for even better tssserver performance, although I think just using the arktype inferred types is likely fast enough in my case (flat objects string/number/date fields)

[kwonoj]

@ssalbdivad Sorry for maybe derailing thread (should I create a new one?)

What is the advantage of codegen if you already have an external source of truth?

Regarding this I'm trying to understand performance impact of using json schema as external sources. We have large surface of schema definitions, generated json schema is larger than several MB in a raw - and given it have multiple external $ref that @ark/json-schema can't resolve duplicated types need to be inlined. In results, total JSON payload size easily exceeds 10MB or more. Given we construct schemas synchronously for the frontend apps this is not a small amount of size & runtime costs.

Before this, we handcrafted arktype schema in ts manually, which seems much smaller than embedding json schema as a source to construct. My naïve assumption is if there's a codegen / preprocess to generate arktype schema would remove need of embedding json schema and reduce the payload size, curious if this is something considerable feature request. typebox-codegen unfortunately not able to fully support arktype fully.