Choosing a Serialization Library

Do you need extended type serialization?

If your data only contains strings, numbers, booleans, arrays, and plain objects, use JSON.stringify. It's built-in, fast, and produces the smallest output.

// If this is all you have, just use JSON
const data = { name: 'Alice', count: 42, tags: ['admin'] }
JSON.stringify(data)

If you have Date, Set, Map, BigInt, RegExp, URL, or custom classes—keep reading.

The landscape

Three libraries solve extended type serialization:

fatline — Schema-aware serialization. Smaller payloads, faster encoding, works with Zod/Valibot/ArkType. Best for typed RPC frameworks.

SuperJSON — The established choice. Path-based metadata format. Larger payloads but battle-tested.

devalue — Minimal and fast. No schema support or custom types. Used internally by SvelteKit.

Performance

Serializing 100 users with dates, sets, and nested objects:

fatline is 2x faster than SuperJSON with 30% smaller payloads.

Payload size

The difference comes from how type information is stored.

SuperJSON uses path-based metadata—a separate object that maps paths to types:

{
  "json": {
    "user": { "createdAt": "2024-01-01T00:00:00.000Z", "roles": ["admin"] }
  },
  "meta": {
    "values": {
      "user.createdAt": ["Date"],
      "user.roles": ["set"]
    }
  }
}

fatline uses inline markers—type info lives with the data:

{
  "user": {
    "createdAt": ["~", 1, 1704067200000],
    "roles": ["~", 4, ["admin"]]
  }
}

For arrays of objects, this difference compounds. 100 users with dates means 100 path entries in SuperJSON's metadata. fatline's inline format scales linearly.

Bundle size

Features

Quick recommendation

Use fatline if:

• You're building with tRPC, oRPC, Hono, or TanStack Query
• Payload size matters (mobile, edge, high-traffic)
• You want schema validation with serialization

Use SuperJSON if:

• You're already using it and migration isn't worth the effort
• You need a library with years of production history

Use devalue if:

• Bundle size is your top priority
• You don't need custom types or schemas

Migrating from SuperJSON

- import superjson from 'superjson'
+ import { transformer } from 'fatline/trpc'

const t = initTRPC.create({
-   transformer: superjson,
+   transformer,
})

Your procedures work unchanged. Update both client and server.