Features

Serializers

Whether we are sending data to the blockchain or reading from it, serialization is a big part of the process. The serialization logic may vary from one program to another and, whilst Borsh serialization is the most popular choice for Solana programs, it is not the only one.

Umi helps with this by providing a flexible and extensible serialization framework that allows you to build your own serializers. Namely, it includes:

  • A generic Serializer<From, To = From> type that represents an object that can serialize From into a Uint8Array and deserialize a Uint8Array into a To which defaults to From.
  • A bunch of serializer helpers that map and transform serializers into new serializers.
  • Last but not least, A set of baked-in serializers that can be used to serialize common types, including string encoders, number serializers, data structures, and more. These primitives can be used to build more complex serializers.

Let's see how all of this works.

Serializer definition

The Serializer type is the central piece of Umi's serialization framework. With a Serializer instance on a type T, you should have all you need to serialize and deserialize instances of T. For instance, a Serializer<{ name: string, age: number }> instance can be used to serialize and deserialize instances of { name: string, age: number }.

In some cases, the data we want to serialize might be slightly looser than the data we get when deserializing. For that reason, the Serializer<From, To> type allows a second type parameter To that extends From and defaults to From. Using our previous example, imagine the age attribute is optional and will default to 42 when not provided. In that case, we can define a Serializer<{ name: string, age?: number }, { name: string, age: number }> instance that serializes { name: string, age?: number } into a Uint8Array but deserializes a Uint8Array into { name: string, age: number }.

Here's how the Serializer type is defined.

type Serializer<From, To extends From = From> = {
  /** A description for the serializer. */
  description: string;
  /** The fixed size of the serialized value in bytes, or `null` if it is variable. */
  fixedSize: number | null;
  /** The maximum size a serialized value can be in bytes, or `null` if it is variable. */
  maxSize: number | null;
  /** The function that serializes a value into bytes. */
  serialize: (value: From) => Uint8Array;
  /**
   * The function that deserializes a value from bytes.
   * It returns the deserialized value and the number of bytes read.
   */
  deserialize: (buffer: Uint8Array, offset?: number) => [To, number];
};

On top of the non-surprising serialize and deserialize functions, the Serializer type also includes a description, a fixedSize and a maxSize.

  • The description is a quick human-readable string that describes the serializer.
  • The fixedSize attribute gives us the size of the serialized value in bytes if and only if we are dealing with a fixed-size serializer. For instance, an u32 serializer will always have a fixedSize of 4 bytes.
  • The maxSize attribute can be helpful when we are dealing with variable-size serializers that have a bound on the maximum size they can take. For instance a borsh Option<PublicKey> serializer can either be of size 1 or 33 and therefore will have a maxSize of 33 bytes.

Using serializers

You can import the Serializer type and everything serializer-related from the @metaplex-foundation/umi/serializers submodule that is bundled with the Umi framework. You may also import it as the standalone @metaplex-foundation/umi-serializers library if you want to use it without the rest of the framework.

// Bundled with Umi.
import { Serializer } from '@metaplex-foundation/umi/serializers';

// As a standalone library.
import { Serializer } from '@metaplex-foundation/umi-serializers';

Once imported, you may use all of the baked-in serializers and helpers that Umi provides. We will dig through each of them in the following sections but, for now, let's have a look at a quick example to see how they work. Say we had a MyObject type containing various attributes, including a name attribute of type string, a publicKey attribute of type PublicKey and a numbers attribute of type number[] such that each number is a u32 integer. Here's how we could create a serializer for it.

import { PublicKey } from '@metaplex-foundation/umi';
import { Serializer, struct, string, publicKey, array, u32 } from '@metaplex-foundation/umi/serializers';

type MyObject = {
  name: string;
  publicKey: PublicKey;
  numbers: number[];
};

const mySerializer: Serializer<MyObject> = struct([
  ['name', string()],
  ['publicKey', publicKey()],
  ['numbers', array(u32())],
]);

Each provided serializer defines its own arguments — e.g. the array function requires the item serializer as a first argument — but most of them have an optional options argument at the end that can be used to tweak the behaviour of the serializer. The attributes inside the options argument may vary from one serializer to the other but they all share one common attribute: description. This can be used to provide a specific description of the created serializer. Notice that, if omitted, a good-enough description will be created for you.

import { string } from '@metaplex-foundation/umi/serializers';

string().description; // -> 'string(utf8; u32(le))'.
string({ description: 'My custom string description' });

Serializer helpers

Now that we know how to import and use serializers, let's have a look at some of the helper methods Umi provide to transform them.

Mapping serializers

The mapSerializer can be used to transform a Serializer<A> into a Serializer<B> by providing two functions that transform B into A and A back into B.

For instance, imagine we want to transform a string serializer into a number serializer by storing the length of the string. Here's how we could use the mapSerializer function to do it.

const serializerA: Serializer<string> = ...;
const serializerB: Serializer<number> = mapSerializer(
  serializerA,
  (value: number): string => 'x'.repeat(value), // Create a mock string of the given length.
  (value: string): number => value.length, // Get the length of the string.
);

The mapSerializer can also be used to transform serializers that have different From and To types. Here's a similar example to the one above but with a different To type.

const serializerA: Serializer<string | null, string> = ...;
const serializerB: Serializer<number | null, number> = mapSerializer(
  serializerA,
  (value: number | null): string | null => value === null ? null : 'x'.repeat(value),
  (value: string): number => value.length,
);

Note that if we are only interested in transforming the From type of a serializer without changing its To type, we can use the mapSerializer function with only one function instead. Here's how we could loosen our Serializer<{ name: string, age: number }> instance to make the age attribute optional when serializing only.

type Person = { name: string, age: number };
type PersonWithOptionalAge = { name: string, age?: number };

const serializerA: Serializer<Person> = ...;
const serializerB: Serializer<PersonWithOptionalAge, Person> = mapSerializer(
  serializerA,
  (value: PersonWithOptionalAge): Person => ({
    name: value.name,
    age: value.age ?? 42,
  }),
);

Mapping serializers is a very powerful technique that can help build complex use-cases whilst still relying on the baked-in serializers.

Fixing serializers

The fixSerializer function is another helper that can transform any variable-size serializer into a fixed-size one by requesting a fixed size in bytes. It does so by padding or truncating the Uint8Array buffer to the requested size when necessary. The returned serializer will have the same From and To types as the original serializer.

const myFixedSerializer = fixSerializer(myVariableSerializer, 42);

Reversing serializers

The reverseSerializer function can be used to reverse the bytes of a fixed-size serializer. Applications of this function are less frequent but it can be useful when dealing with endianness for instance. Here again, the returned serializer will have the same From and To types as the original serializer.

const myReversedSerializer = reverseSerializer(mySerializer);

Byte helpers

It is worth noting that some low-level helper methods are also provided to manipulate bytes. These do not return serializers but can be useful when building custom ones.

// Merge multiple Uint8Array buffers into one.
mergeBytes([new Uint8Array([1, 2]), new Uint8Array([3, 4])]); // -> Uint8Array([1, 2, 3, 4])

// Pad a Uint8Array buffer to the given size.
padBytes(new Uint8Array([1, 2]), 4); // -> Uint8Array([1, 2, 0, 0])
padBytes(new Uint8Array([1, 2, 3, 4]), 2); // -> Uint8Array([1, 2, 3, 4])

// Pad and truncate a Uint8Array buffer to the given size.
fixBytes(new Uint8Array([1, 2]), 4); // -> Uint8Array([1, 2, 0, 0])
fixBytes(new Uint8Array([1, 2, 3, 4]), 2); // -> Uint8Array([1, 2])

Baked-in serializers

Let's now take a look at the various serializers that are shipped with Umi. Each of these primitives can be used to build more complex serializers as we've seen in the previous section.

Numbers

Umi ships with 12 number serializers: 5 unsigned integers, 5 signed integers and 2 floating point numbers. These can be used to serialize and deserialize numbers of different sizes. When the size of the number is greater than 32 bits, the serializer returned is a Serializer<number | bigint, bigint> instead of a Serializer<number> since JavaScript's native number type does not support numbers larger than 2^53 - 1.

// Unsigned integers.
u8(); // -> Serializer<number>
u16(); // -> Serializer<number>
u32(); // -> Serializer<number>
u64(); // -> Serializer<number | bigint, bigint>
u128(); // -> Serializer<number | bigint, bigint>

// Signed integers.
i8(); // -> Serializer<number>
i16(); // -> Serializer<number>
i32(); // -> Serializer<number>
i64(); // -> Serializer<number | bigint, bigint>
i128(); // -> Serializer<number | bigint, bigint>

// Floating point numbers.
f32(); // -> Serializer<number>
f64(); // -> Serializer<number>

Aside from the u8 and i8 serializers that use only one byte, all other number serializers are represented in little-endian by default and can be configured to use a different endianness. This can be done by passing the endian option to the serializer.

u64(); // Little-endian.
u64({ endian: Endian.Little }); // Little-endian.
u64({ endian: Endian.Big }); // Big-endian.

Note that, since number serializers are often reused in other serializers, Umi defines the following NumberSerializer type to include both number and bigint types.

type NumberSerializer =
  | Serializer<number>
  | Serializer<number | bigint, bigint>;

Booleans

The bool serializer can be used to create a Serializer<boolean>. By default, it uses a u8 number to store the boolean value but this can be changed by passing a NumberSerializer to the size option.

bool(); // -> Uses a u8.
bool({ size: u32() }); // -> Uses a u32.
bool({ size: u32({ endian: Endian.Big }) }); // -> Uses a big-endian u32.

String encodings

Umi ships with the following string serializers that can be used to serialize and deserialize strings in different formats: utf8, base10, base16, base58 and base64.

utf8.serialize('Hello World!');
base10.serialize('42');
base16.serialize('ff002a');
base58.serialize('LorisCg1FTs89a32VSrFskYDgiRbNQzct1WxyZb7nuA');
base64.serialize('SGVsbG8gV29ybGQhCg==');

It also ships with a baseX function that can create new string serializers for any given alphabet. For instance, this is how the base58 serializer is implemented.

const base58: Serializer<string> = baseX(
  '123456789ABCDEFGHJKLMNPQRSTUVWXYZabcdefghijkmnopqrstuvwxyz'
);

Strings

The string serializer returns a Serializer<string> that can be used to serialize strings using various encodings and size strategies. It contains the following options:

  • encoding: A Serializer<string> that represents the encoding to use when serializing and deserializing the string. It defaults to the built-in utf8 serializer. You might be wondering, why do we need to pass a Serializer<string> to create a Serializer<string>? This is because the purpose of the encoding serializer is only to convert some text to and from a byte array without worrying about anything else such as storing the size of the string. This allows us to plug in any encoding we want, whilst being able to leverage all other options provided by this string function.
  • size: In order to know how long the string goes on for in a given buffer, we need to know its size in bytes. To that end, one of the following size strategies may be used:
    • NumberSerializer: When a number serializer is passed, it will be used as a prefix to store and restore the size of the string. By default, the size is stored using a u32 prefix in little-endian — which is the default behaviour for borsh serialization.
    • number: The byte size can also be provided explicitly as a number. This will create a fixed-size serializer that does not use any size prefix and will always use the same number of bytes to store the string.
    • "variable": When the string "variable" is passed as a size, it will create a variable-size serializer that simply uses all the remaining bytes in the buffer when deserializing. When serializing, it will simply return the result of the encoding serializer without storing the size of the serialized string.
// Serialized values using different encodings for reference.
utf8.serialize('Hi'); // -> 0x4869
base58.serialize('Hi'); // -> 0x03c9

// Default behaviour: utf8 encoding and u32 (litte-endian) size.
string().serialize('Hi'); // -> 0x020000004869

// Custom encoding: base58.
string({ encoding: base58 }).serialize('Hi'); // -> 0x0200000003c9

// Custom size: u16 (big-endian) size.
string({ size: u16({ endian: Endian.Big }) }).serialize('Hi'); // -> 0x00024869

// Custom size: 5 bytes.
string({ size: 5 }).serialize('Hi'); // -> 0x4869000000

// Custom size: variable.
string({ size: 'variable' }).serialize('Hi'); // -> 0x4869

Bytes

The bytes serializer returns a Serializer<Uint8Array> which deserializes a Uint8Array into a... Uint8Array. Whilst this might seem a bit useless, it can be useful when composed into other serializers. For example, you could use it in a struct serializer to say that a particular field should be left unserialized.

Very similar to the string function, the bytes function contains a size option that configures how the size of the byte array is stored and restored. The same size strategies are supported as for the string function except that the default size here is the "variable" strategy. To recap:

  • NumberSerializer: Uses a prefixed number serializer to store and restore the size of the byte array.
  • number: Uses a fixed size to store the byte array.
  • "variable": Passes the buffer as-is when serializing and returns the remaining of the buffer when deserializing. Defaults behaviour.
// Default behaviour: variable size.
bytes().serialize(new Uint8Array([42])); // -> 0x2a

// Custom size: u16 (little-endian) size.
bytes({ size: u16() }).serialize(new Uint8Array([42])); // -> 0x01002a

// Custom size: 5 bytes.
bytes({ size: 5 }).serialize(new Uint8Array([42])); // -> 0x2a00000000

PublicKeys

The publicKey serializer returns a Serializer<PublicKey> that can be used to serialize and deserialize public keys. Here's an example of serializing and deserializing the same public key. Notice that the publicKey function is also exported by the main @metaplex-foundation/umi package and allows us to create public keys from various input. Therefore you may need to alias your imports to avoid conflicts.

import { publicKey } from '@metaplex-foundation/umi';
import { publicKey as publicKeySerializer } from '@metaplex-foundation/umi/serializers';

const myPublicKey = publicKey('...');
const buffer = publicKeySerializer().serialize(myPublicKey);
const [myDeserializedPublicKey, offset] = publicKeySerializer().deserialize(buffer);
myPublicKey === myDeserializedPublicKey; // -> true

Units

The unit serializer returns a Serializer<void> that serializes undefined into an empty Uint8Array and returns undefined without consuming any bytes when deserializing. This is more of a low-level serializer that can be used internally by other serializers. For instance, this is how dataEnum serializers describe empty variants internally.

unit().serialize(undefined); // -> new Uint8Array([])
unit().deserialize(new Uint8Array([42])); // -> [undefined, 0]

Arrays, Sets and Maps

Umi provides three functions to serialize lists and maps:

  • array: Serializes an array of items. It accepts a Serializer<T> as an argument and returns a Serializer<T[]>.
  • set: Serializes a set of unique items. It accepts a Serializer<T> as an argument and returns a Serializer<Set<T>>.
  • map: Serializes a map of key-value pairs. It accepts a Serializer<K> for the keys and a Serializer<V> for the values as arguments and returns a Serializer<Map<K, V>>.

All three functions accept the same size option that configures how the length of the array, set or map is stored and restored. This is very similar to how the string and bytes serializers work. Here are the supported strategies:

  • NumberSerializer: Uses a number serializer that prefixes the content with its size. By default, the size is stored using a u32 prefix in little-endian.
  • number: Returns an array, set or map serializer with a fixed number of items.
  • "remainder": Returns an array, set or map serializer that infers the number of items by dividing the rest of the buffer by the fixed size of its item. For instance, if a buffer has 64 bytes remaining and each item of an array is 16 bytes long, the array will be deserialized with 4 items. Note that this option is only available for fixed-size items. For maps, both the key serializer and the value serializer must have a fixed size.
// Arrays.
array(u8()) // Array of u8 items with a u32 size prefix.
array(u8(), { size: 5 }) // Array of 5 u8 items.
array(u8(), { size: 'remainder' }) // Array of u8 items with a variable size.

// Sets.
set(u8()) // Set of u8 items with a u32 size prefix.
set(u8(), { size: 5 }) // Set of 5 u8 items.
set(u8(), { size: 'remainder' }) // Set of u8 items with a variable size.

// Maps.
map(u8(), u8()) // Map of (u8, u8) entries with a u32 size prefix.
map(u8(), u8(), { size: 5 }) // Map of 5 (u8, u8) entries.
map(u8(), u8(), { size: 'remainder' }) // Map of (u8, u8) entries with a variable size.

Options and Nullables

Umi provides two functions to serialize optional values:

  • nullable: Serializes a value that can be null. It accepts a Serializer<T> as an argument and returns a Serializer<Nullable<T>> where Nullable<T> is a type alias for T | null.
  • option: Serializes an Option instance (See documentation). It accepts a Serializer<T> as an argument and returns a Serializer<OptionOrNullable<T>, Option<T>>. This means deserialized values will always be wrapped in an Option type but serialized values can either be an Option<T> or a Nullable<T>.

Both functions serialize optional values by prefixing them with a boolean value that indicates whether the value is present or not. If the prefixed boolean is false, the value is null (for nullables) or None (for options) and we can skip deserializing the actual value. Otherwise, the value is deserialized using the provided serializer and returned.

They both offer the same options to configure the behaviour of the created serializer:

  • prefix: The NumberSerializer to use to serialize and deserialize the boolean prefix. By default, it uses a u8 prefix in little-endian.
  • fixed: When this is true, it returns a fixed-size serializer by changing the serialization logic when the value is empty. In this case, the serialized value will be padded with zero such that empty values and filled values are serialized using the same amount of bytes. Note that this only works if the item serializer is of a fixed size.
// Options.
option(publicKey()) // Option<PublicKey> with a u8 prefix.
option(publicKey(), { prefix: u16() }) // Option<PublicKey> with a u16 prefix.
option(publicKey(), { fixed: true }) // Option<PublicKey> with a fixed size.

// Nullables.
nullable(publicKey()) // Nullable<PublicKey> with a u8 prefix.
nullable(publicKey(), { prefix: u16() }) // Nullable<PublicKey> with a u16 prefix.
nullable(publicKey(), { fixed: true }) // Nullable<PublicKey> with a fixed size.

Structs

The struct serializer allows us to serialize and deserialize a JavaScript object of generic type T.

It requires the name and the serializer of each field to be passed as an array on the first argument. This fields array is structured such that each field is a tuple where the first item is the name of the field and the second item is the serializer of the field. The order of the fields is important because it determines the order in which the fields are serialized and deserialized. Here's an example.

type Person = {
  name: string;
  age: number;
}

struct<Person>([
  ['name', string()],
  ['age', u32()],
]);

The struct function also accepts a second type parameter U in case some fields have different From and To type parameters. This allows us to create serializers of type Serializer<T, U>.

For instance, this is how we could create a struct serializer that offers a default value for the age field of the Person type.

type Person = { name: string; age: number; }
type PersonArgs = { name: string; age?: number; }

const ageOr42 = mapSerializer(
  u32(),
  (age: number | undefined): number => age ?? 42,
);

struct<PersonArgs, Person>([
  ['name', string()],
  ['age', ageOr42],
]);

Tuples

Umi offers a tuple serializer that can be used to serialize and deserialize tuples. Whilst tuples are not native in JavaScript, they can be represented in TypeScript using an array such that each item has its own defined type. For instance, a (String, u8) tuple in Rust can be represented as a [string, number] in TypeScript.

The tuple function accepts an array of serializers as its first argument that should match the items of the tuple in the same order. Here are a few examples.

tuple([bool()]); // Serializer<[bool]>
tuple([string(), u8()]); // Serializer<[string, number]>
tuple([publicKey(), u64()]); // Serializer<[PublicKey, number | bigint], [PublicKey, bigint]>

Scalar Enums

The scalarEnum function can be used to create serializers for scalar enums by storing the value (or index) of the variant as a u8 number.

It requires the enum constructor as its first argument. For instance, if an enum is defined as enum Direction { Left }, then the constructor Direction should be passed as the first argument. The serializer created will accept any variant of the enum as input, as well as its value or its name. Here is an example.

enum Direction { Left, Right, Up, Down };

const directionSerializer = scalarEnum(Direction); // Serializer<Direction>
directionSerializer.serialize(Direction.Left); // -> 0x00
directionSerializer.serialize(Direction.Right); // -> 0x01
directionSerializer.serialize('Left'); // -> 0x00
directionSerializer.serialize('Right'); // -> 0x01
directionSerializer.serialize(0); // -> 0x00
directionSerializer.serialize(1); // -> 0x01

// The deserialized value is always an instance of the enum.
directionSerializer.deserialize(new Uint8Array([1])); // -> [Direction.Right, 1]

Whilst the serialized value default to being stored using a u8 number serializer, a custom NumberSerializer can be provided as the size option to change that behaviour.

scalarEnum(Direction, { size: u32() }).serialize(Direction.Right); // -> 0x01000000

Note that if you use the scalarEnum function with a string enum — e.g. enum Direction { Left = 'LEFT' } — it will ignore the text value and only use the index of the variant.

enum Direction { Left = 'LEFT', Right = 'RIGHT', Up = 'UP', Down = 'DOWN' };

const directionSerializer = scalarEnum(Direction); // Serializer<Direction>
directionSerializer.serialize(Direction.Left); // -> 0x00
directionSerializer.serialize('Left'); // -> 0x00

// Note that the enum string value can be used as input.
directionSerializer.serialize('LEFT'); // -> 0x00

Data Enums

In Rust, enums are powerful data types whose variants can be one of the following:

  • An empty variant — e.g. enum Message { Quit }.
  • A tuple variant — e.g. enum Message { Write(String) }.
  • A struct variant — e.g. enum Message { Move { x: i32, y: i32 } }.

Whilst we do not have such powerful enums in JavaScript, we can emulate them in TypeScript using a union of objects such that each object is differentiated by a specific field. We call this a data enum.

In Umi, we use the __kind field to distinguish between the different variants of a data enum. Additionally, since all variants are objects, we use the fields property to wrap the array of tuple variants. Here is an example.

type Message = 
  | { __kind: 'Quit' } // Empty variant.
  | { __kind: 'Write'; fields: [string] } // Tuple variant.
  | { __kind: 'Move'; x: number; y: number }; // Struct variant.

The dataEnum function allows us to create serializers for data enums. It requires the name and serializer of each variant as a first argument. Similarly to the struct serializer, these are defined as an array of variant tuples where the first item is the name of the variant and the second item is the serializer of the variant. Since empty variants do not have data to serialize, they simply use the unit serializer. Here is how we can create a data enum serializer for our previous example.

const messageSerializer = dataEnum<Message>([
  // Empty variant.
  ['Quit', unit()],
  // Tuple variant.
  ['Write', struct<{ fields: [string] }>([
    ['fields', tuple([string()])]
  ])],
  // Struct variant.
  ['Move', struct<{ x: number; y: number }>([
    ['x', i32()],
    ['y', i32()]
  ])],
]);

Note that this serialization is compatible with the borsh serialization of Rust enums. First, it uses a u32 number in little-endian to store the index of the variant. If the selected variant is an empty variant, it stops there. Otherwise, it uses the serializer of the variant to serialize its data.

messageSerializer.serialize({ __kind: 'Quit' }); // -> 0x00000000
messageSerializer.serialize({ __kind: 'Write', fields: ['Hi'] }); // -> 0x01000000020000004869
messageSerializer.serialize({ __kind: 'Move', x: 5, y: 6 }); // -> 0x020000000500000006000000

The dataEnum function also accepts a prefix option that allows us to select a custom number serializer for the variant index — instead of the default u32 as mentioned above. Here's an example using a u8 instead of a u32.

const messageSerializer = dataEnum<Message>([...], {
  prefix: u8()
});

messageSerializer.serialize({ __kind: 'Quit' }); // -> 0x00
messageSerializer.serialize({ __kind: 'Write', fields: ['Hi'] }); // -> 0x01020000004869
messageSerializer.serialize({ __kind: 'Move', x: 5, y: 6 }); // -> 0x020500000006000000

Note that, when dealing with data enums, you may want to offer some helper methods to improve the developer experience so that it feels closer to the Rust way of handling enums. This is something that Kinobi offers to generated JavaScript clients out of the box.

// Example of helper methods.
message('Quit'); // -> { __kind: 'Quit' }
message('Write', ['Hi']); // -> { __kind: 'Write', fields: ['Hi'] }
message('Move', { x: 5, y: 6 }); // -> { __kind: 'Move', x: 5, y: 6 }
isMessage('Quit', message('Quit')); // -> true
isMessage('Write', message('Quit')); // -> false

Bit arrays

The bitArray serializer can be used to serialize and deserialize arrays of booleans such that each boolean is represented by a single bit. It requires the size of the serializer in bytes and an optional backward flag that can be used to reverse the order of the bits.

const booleans = [true, false, true, false, true, false, true, false];
bitArray(1).serialize(booleans); // -> Uint8Array.from([0b10101010]);
bitArray(1).deserialize(Uint8Array.from([0b10101010])); // -> [booleans, 1];
Previous
RPC