Accept init shapes in Standard Schema validation

[Adrastopoulos]

Fixes #95.

createStandardSchema() only accepts message instances. A plain object fails before any rule is evaluated, even when the data is valid:

const schema = createStandardSchema(UserSchema);
await schema["~standard"].validate({ email: "test@example.com" });
// { issues: [{ message: "Cannot validate message undefined with schema example.User" }] }

Standard Schema consumers usually hold plain values, not message instances: form state, or deserialized RPC payloads (e.g. TanStack Start's inputValidator).

Changes:

• validate() now converts plain objects to messages with create() before validating. Message instances are validated directly, as before. Errors thrown by create() become issues, since Standard Schema validators must not throw.
• createStandardSchema() keeps its declared types. The Standard Schema input parameter is covariant, so widening it to MessageInitShape would break existing consumers, e.g. TanStack Form with defaultValues: create(Schema).
• New createStandardSchemaInit() shares the runtime but declares MessageInitShape as its input, for consumers where the schema's input type dictates what callers pass. MessageInitShape is not assignable to MessageShape, so one input type cannot serve both.

[CLAassistant]

All committers have signed the CLA.

[timostamm]

Thanks for the PR!

I need some help understanding the changes.

The Standard Schema input parameter is covariant, so widening it to MessageInitShape would break existing consumers, e.g. TanStack Form with defaultValues: create(Schema).

MessageInitShape is either the fully materialized message, or a partial message where all fields are optional, see https://github.com/bufbuild/protobuf-es/blob/v2.14.1/packages/protobuf/src/types.ts#L195.

I think that means it should be possible to simplify the changes?

validate() now converts plain objects to messages with create() before validating. Message instances are validated directly, as before. Errors thrown by create() become issues, since Standard Schema validators must not throw.

Makes sense. When create throws, there is nothing to validate further. So I believe that the new createStandardSchemaInit could simply wrap the existing function, no? That would probably make the code paths easier to follow, and be optimal for bundle sizes.

[Adrastopoulos]

Yes, you're right on both counts. We can just return MessageShape since it's assignable to MessageInitShape. Indeed, createStandardSchemaInit can just be a thin wrapper.

[Adrastopoulos]

I also realized that create() doesn't throw. It already checks if the input is a Message instance, hence our try/catch duplicated that check. Now it's even simpler.