sveltejs/devalue

devalue

Like JSON.stringify, but handles

• cyclical references (obj.self = obj)
• repeated references ([value, value])
• undefined, Infinity, NaN, -0
• regular expressions
• dates
• Map and Set
• BigInt
• ArrayBuffer and Typed Arrays
• URL and URLSearchParams
• Temporal
• custom types via replacers, reducers and revivers
• promises (via stringifyAsync)

Try it out here.

Goals:

• Performance
• Security (see XSS mitigation)
• Compact output

Non-goals:

• Human-readable output
• Stringifying functions
• Stability of serialization mechanisms between versions (i.e. if you devalue.stringify with one version and devalue.parse with another, things may break)

Usage

There are two ways to use devalue:

uneval

This function takes a JavaScript value and returns the JavaScript code to create an equivalent value — sort of like eval in reverse:

import * as devalue from 'devalue';

let obj = { message: 'hello' };
devalue.uneval(obj); // '{message:"hello"}'

obj.self = obj;
devalue.uneval(obj); // '(function(){let a={};a.message="hello";a.self=a;return a}())'

Use uneval when you want the most compact possible output and don't want to include any code for parsing the serialized value. The result is suitable for embedding in a <script> element or a generated JavaScript module, for example.

stringify and parse

These two functions are analogous to JSON.stringify and JSON.parse:

import * as devalue from 'devalue';

let obj = { message: 'hello' };

let stringified = devalue.stringify(obj); // '[{"message":1},"hello"]'
devalue.parse(stringified); // { message: 'hello' }

obj.self = obj;

stringified = devalue.stringify(obj); // '[{"message":1,"self":0},"hello"]'
devalue.parse(stringified); // { message: 'hello', self: [Circular] }

Use stringify and parse when evaluating JavaScript isn't an option (such as when sending untrusted data from client to server).

stringifyAsync

stringifyAsync is an async version of stringify that can handle promises:

import * as devalue from 'devalue';

let obj = {
  quick: 'data',
  slow: fetch('/api/slow').then((r) => r.json())
};

let stringified = await devalue.stringifyAsync(obj);
devalue.parse(stringified); // { quick: 'data', slow: { ... } }

Promises are awaited and their resolved values are serialized. The output format is identical to stringify, so parse and unflatten work unchanged.

unflatten

In the case where devalued data is one part of a larger JSON string, unflatten allows you to revive just the bit you need:

import * as devalue from 'devalue';

const input = { message: 'hello' };
const json = `{
  "type": "data",
  "data": ${devalue.stringify(input)}
}`;

const data = devalue.unflatten(JSON.parse(json).data); // { message: 'hello' }

Binary data

Node.js Buffer values are serialized as Uint8Arrays containing only the Buffer's visible bytes. Their backing stores are copied, since small Buffers can share an allocation pool containing unrelated, sensitive data. Repeated references to the same Buffer are preserved, but distinct Buffers are given separate backing stores.

For ordinary typed arrays and DataViews, devalue serializes the entire backing ArrayBuffer, preserving shared views and byte offsets. This includes bytes outside a subarray or subview. Only serialize these views if their entire backing store is safe to disclose, or copy the intended bytes first:

const copy = new Uint8Array(new Uint8Array(view.buffer, view.byteOffset, view.byteLength));
const serialized = devalue.stringify(copy);

This also applies if you explicitly pass buf.buffer, or create an ordinary typed array or DataView over a Node Buffer's backing store — these values cannot be identified as Node Buffers. To copy a Buffer's contents into an independent typed array, use new Uint8Array(buf).

Custom types

You can serialize and deserialize custom types by passing a second argument to stringify containing an object of types and their reducers, and a second argument to parse or unflatten containing an object of types and their revivers:

class Vector {
  constructor(x, y) {
    this.x = x;
    this.y = y;
  }

  magnitude() {
    return Math.sqrt(this.x * this.x + this.y * this.y);
  }
}

const stringified = devalue.stringify(new Vector(30, 40), {
  Vector: (value) => value instanceof Vector && [value.x, value.y]
});

console.log(stringified); // [["Vector",1],[2,3],30,40]

const vector = devalue.parse(stringified, {
  Vector: ([x, y]) => new Vector(x, y)
});

console.log(vector.magnitude()); // 50

If a function passed to stringify returns a truthy value, it's treated as a match.

You can also use custom types with uneval by specifying a custom replacer:

devalue.uneval(vector, (value, js) => {
  if (value instanceof Vector) {
    return js`new Vector(${value.x},${value.y})`;
  }
}); // `new Vector(30,40)`

The replacer must return a source created with the supplied js tag, or undefined, null or false to serialize the value normally. Each value "hole" is recursively serialized. Identifier-like words in the literal template strings (including nested templates) are reserved, so generated variables cannot shadow your constructors, helpers or local bindings.

Note that any variables referenced in the resulting JavaScript (like Vector in the example above) must be in scope when it runs.

Custom operations

Every introspection stringify performs on the value being serialized — property reads, prototype method calls, iteration, type classification — goes through an operations interface that you can override via the operations option. Omitted members fall back to the defaults (exported as defaultStringifyOperations), which behave exactly as devalue always has.

This is useful in two situations:

Side-effect-free serialization. By default, serializing a value can execute user code: getters and proxy traps fire during property reads, Object.prototype.toString consults (potentially getter-defined) Symbol.toStringTag, and patched prototype methods like Date.prototype.toISOString or Map.prototype[Symbol.iterator] are invoked. Deterministic or sandboxed runtimes can replace these operations with implementations based on captured intrinsics and property descriptors:

const originalToISOString = Date.prototype.toISOString;

const stringified = devalue.stringify(value, undefined, {
  operations: {
    // use a captured intrinsic instead of a (possibly patched) prototype method
    toISOString: (date) => originalToISOString.call(date),

    // read through descriptors so getters are never invoked
    get: (object, key) => {
      const descriptor = Object.getOwnPropertyDescriptor(object, key);
      if (descriptor?.get) throw new Error(`refusing to invoke getter for "${key}"`);
      return descriptor?.value;
    }
  }
});

Foreign-runtime serialization. The stringify algorithm never touches the value directly, so "value" can be an opaque handle to something living in another JavaScript runtime — a node:vm context, a WASM-hosted engine, a remote process — as long as the operations know how to inspect it. Implement typeOf/tagOf for classification, toPrimitive/get/entriesOf/etc. for extraction, and identify to key deduplication and cycle detection on the underlying value's identity rather than the handle's:

const stringified = devalue.stringify(rootHandle, undefined, {
  operations: {
    identify: (handle) => handle.pointer,
    typeOf: (handle) => handle.typeOf(),
    get: (handle, key) => handle.getProperty(key)
    // ... see StringifyOperations for the full interface
  }
});

Some operations have a non-obvious contract that is easy to get subtly wrong. Where the work is not specific to your values, devalue exports the pieces so you don't have to reimplement them — filterArrayIndices does the array-index filtering that indicesOf needs, given keys you already have:

indicesOf: (handle) => devalue.filterArrayIndices(handle.ownEnumerableStringKeys());

Reducers compose with custom operations: they receive the raw value/handle, and whatever they return is serialized through the same operations.

Customizing parse

The mirror image: parse and unflatten build every value through construction operations (ParseOperations, defaults exported as defaultParseOperations), so you can control what gets created. The members mirror StringifyOperations with the host/value-space boundary running the other way: each fromXxx inverts the corresponding toXxx, fromXxxInfo inverts xxxInfo, and the bare-verb mutators invert the bare-verb accessors (set/get, addValue/valuesOf, addEntry/entriesOf, box/unbox).

Cross-realm revival. By default the revived value is built from the intrinsics of whichever realm devalue is running in, so instanceof checks fail elsewhere. Constructing from a target realm's intrinsics fixes that:

const revived = devalue.parse(serialized, undefined, {
  operations: {
    fromISOString: (iso) => new sandbox.Date(iso),
    createMap: () => new sandbox.Map(),
    createObject: () => sandbox.makeObject()
  }
});

Foreign-runtime revival. parse never inspects the values it creates — it only passes them back into other operations — so the operations can build values inside another runtime and return opaque handles:

const rootHandle = devalue.parse(serialized, undefined, {
  operations: {
    fromPrimitive: (primitive) => vm.toHandle(primitive),
    createObject: () => vm.newObject(),
    set: (handle, key, value) => handle.setProp(key, value)
    // ... see ParseOperations for the full interface
  }
});

Containers are created empty and populated afterwards (createMap then addEntry, createObject then set, and so on) — that ordering is what allows cyclic values to be revived, since the empty container is cached before its contents are built.

Revivers compose the same way reducers do: they receive whatever the operations built, and their return value is used as-is.

Error handling

If uneval or stringify encounters a function or a non-POJO that isn't handled by a custom replacer/reducer, it will throw an error. You can find where in the input data the offending value lives by inspecting error.path:

try {
  const map = new Map();
  map.set('key', function invalid() {});

  uneval({
    object: {
      array: [map]
    }
  });
} catch (e) {
  console.log(e.path); // '.object.array[0].get("key")'
}

XSS mitigation

Say you're server-rendering a page and want to serialize some state, which could include user input. JSON.stringify doesn't protect against XSS attacks:

const state = {
  userinput: `</script><script src='https://evil.com/mwahaha.js'>`
};

const template = `
<script>
  // NEVER DO THIS
  var preloaded = ${JSON.stringify(state)};
</script>`;

Which would result in this:

<script>
  // NEVER DO THIS
  var preloaded = {"userinput":"
</script>
<script src="https://evil.com/mwahaha.js">
  "};
</script>

Using uneval or stringify, we're protected against that attack:

const template = `
<script>
  var preloaded = ${uneval(state)};
</script>`;

<script>
  var preloaded = {
    userinput:
      "\\u003C\\u002Fscript\\u003E\\u003Cscript src='https:\\u002F\\u002Fevil.com\\u002Fmwahaha.js'\\u003E"
  };
</script>

This, along with the fact that uneval and stringify bail on functions and non-POJOs, stops attackers from executing arbitrary code. Strings generated by uneval can be safely deserialized with eval or new Function:

const value = (0, eval)('(' + str + ')');

Other security considerations

While uneval prevents the XSS vulnerability shown above, meaning you can use it to send data from server to client, you should not send user data from client to server using the same method. Since it has to be evaluated, an attacker that successfully submitted data that bypassed uneval would have access to your system.

When using eval, ensure that you call it indirectly so that the evaluated code doesn't have access to the surrounding scope:

{
  const sensitiveData = 'Setec Astronomy';
  eval('sendToEvilServer(sensitiveData)'); // pwned :(
  (0, eval)('sendToEvilServer(sensitiveData)'); // nice try, evildoer!
}

Using new Function(code) is akin to using indirect eval.

See also

• lave by Jed Schmidt
• arson by Ben Newman. The stringify/parse approach in devalue was inspired by arson
• oson by Steffen Trog
• tosource by Marcello Bastéa-Forte
• serialize-javascript by Eric Ferraiuolo
• jsesc by Mathias Bynens
• superjson by Blitz
• next-json by Daniele Ricci

License

MIT