Improve type inference from Zod schemas

[jspahrsummers]

Adds some type utilities that attempt to flatten and simplify inferred Zod types, for better readability in hoverover, typechecking, and IDE features.

Since TypeScript is structurally typed, not nominally typed, this shouldn't be a breaking change (as long as the resultant object types are indeed the same structurally!).

For example, a couple of types before:

type ReadResourceRequest = {
    params: {
        uri: string;
        _meta?: z.objectOutputType<{
            progressToken: z.ZodOptional<z.ZodUnion<[z.ZodString, z.ZodNumber]>>;
        }, z.ZodTypeAny, "passthrough"> | undefined;
    } & {
        ...;
    };
    method: "resources/read";
}

type ListPromptsRequest = {
    method: "prompts/list";
    params?: z.objectOutputType<z.objectUtil.extendShape<{
        _meta: z.ZodOptional<z.ZodObject<{
            progressToken: z.ZodOptional<z.ZodUnion<[z.ZodString, z.ZodNumber]>>;
        }, "passthrough", z.ZodTypeAny, z.objectOutputType<{
            progressToken: z.ZodOptional<z.ZodUnion<[z.ZodString, z.ZodNumber]>>;
        }, z.ZodTypeAny, "passthrough">, z.objectInputType<...>>>;
    }, {
        ...;
    }>, z.ZodTypeAny, "passthrough"> | undefined;
}

And after:

type ReadResourceRequest = {
    params: {
        [x: string]: unknown;
        uri: string;
        _meta?: {
            [x: string]: unknown;
            progressToken?: string | number | undefined;
        } | undefined;
    };
    method: "resources/read";
}

type ListPromptsRequest = {
    method: "prompts/list";
    params?: {
        [x: string]: unknown;
        _meta?: {
            [x: string]: unknown;
            progressToken?: string | number | undefined;
        } | undefined;
        cursor?: string | undefined;
    } | undefined;
}

[jerome3o-anthropic]

awesome - tested in my ide and the hover tips are soo much better.

I don't really understand why there isn't a nested z.infer already. But this works!

[jerome3o-anthropic]

q: why doesn't z.infer just do this already?

[jerome3o-anthropic]

Looks like the weirdness only happens on zod types with .passthrough().

[jspahrsummers]

Honestly, no idea. The only relevant thing I could find is https://github.com/colinhacks/zod/blob/f7ad26147ba291cb3fb257545972a8e00e767470/src/helpers/partialUtil.ts, which doesn't apply to general z.infer usage—only benefitted a feature that is now deprecated in Zod.

I'm surprised not to see even an issue in the backlog that looks related here. 🤷