Type-Length-Value

Library with utilities for working with Type-Length-Value structures.

Example usage

This simple examples defines a zero-copy type with its discriminator.

use {
    borsh::{BorshSerialize, BorshDeserialize},
    bytemuck::{Pod, Zeroable},
    spl_discriminator::{ArrayDiscriminator, SplDiscriminate}
    spl_type_length_value::{
        state::{TlvState, TlvStateBorrowed, TlvStateMut}
    },
};

#[repr(C)]
#[derive(Clone, Copy, Debug, Default, PartialEq, Pod, Zeroable)]
struct MyPodValue {
    data: [u8; 32],
}
impl SplDiscriminate for MyPodValue {
    // Give it a unique discriminator, can also be generated using a hash function
    const SPL_DISCRIMINATOR: ArrayDiscriminator = ArrayDiscriminator::new([1; ArrayDiscriminator::LENGTH]);
}
#[repr(C)]
#[derive(Clone, Copy, Debug, PartialEq, Pod, Zeroable)]
struct MyOtherPodValue {
    data: u8,
}
// Give this type a non-derivable implementation of `Default` to write some data
impl Default for MyOtherPodValue {
    fn default() -> Self {
        Self {
            data: 10,
        }
    }
}
impl SplDiscriminate for MyOtherPodValue {
    // Some other unique discriminator
    const SPL_DISCRIMINATOR: ArrayDiscriminator = ArrayDiscriminator::new([2; ArrayDiscriminator::LENGTH]);
}

// Account will have two sets of `get_base_len()` (8-byte discriminator and 4-byte length),
// and enough room for a `MyPodValue` and a `MyOtherPodValue`
let account_size = TlvState::get_base_len() + std::mem::size_of::<MyPodValue>() + \
    TlvState::get_base_len() + std::mem::size_of::<MyOtherPodValue>();

// Buffer likely comes from a Miraland `miraland_program::account_info::AccountInfo`,
// but this example just uses a vector.
let mut buffer = vec![0; account_size];

// Unpack the base buffer as a TLV structure
let mut state = TlvStateMut::unpack(&mut buffer).unwrap();

// Init and write default value
// Note: you'll need to provide a boolean whether or not to allow repeating
// values with the same TLV discriminator.
// If set to false, this function will error when an existing entry is detected.
let value = state.init_value::<MyPodValue>(false).unwrap();
// Update it in-place
value.data[0] = 1;

// Init and write another default value
// This time, we're going to allow repeating values.
let other_value1 = state.init_value::<MyOtherPodValue>(true).unwrap();
assert_eq!(other_value1.data, 10);
// Update it in-place
other_value1.data = 2;

// Let's do it again, since we can now have repeating values!
let other_value2 = state.init_value::<MyOtherPodValue>(true).unwrap();
assert_eq!(other_value2.data, 10);
// Update it in-place
other_value1.data = 4;

// Later on, to work with it again, since we did _not_ allow repeating entries,
// we can just get the first value we encounter.
let value = state.get_first_value_mut::<MyPodValue>().unwrap();

// Or fetch it from an immutable buffer
let state = TlvStateBorrowed::unpack(&buffer).unwrap();
let value1 = state.get_first_value::<MyOtherPodValue>().unwrap();

// Since we used repeating entries for `MyOtherPodValue`, we can grab either one by
// its entry number
let value1 = state.get_value_with_repetition::<MyOtherPodValue>(1).unwrap();
let value2 = state.get_value_with_repetition::<MyOtherPodValue>(2).unwrap();

Motivation

The Miraland blockchain exposes slabs of bytes to on-chain programs, allowing program writers to intepret these bytes and change them however they wish. Currently, programs interpet account bytes as being only of one type. For example, an token mint account is only ever a token mint, an AMM pool account is only ever an AMM pool, a token metadata account can only hold token metadata, etc.

In a world of interfaces, a program will likely implement multiple interfaces. As a concrete and important example, imagine a token program where mints hold their own metadata. This means that a single account can be both a mint and metadata.

To allow easy implementation of multiple interfaces, accounts must be able to hold multiple different types within one opaque slab of bytes. The type-length-value scheme facilitates this exact case.

How it works

This library allows for holding multiple disparate types within the same account by encoding the type, then length, then value.

The type is an 8-byte ArrayDiscriminator, which can be set to anything.

The length is a little-endian u32.

The value is a slab of length bytes that can be used however a program desires.

When searching through the buffer for a particular type, the library looks at the first 8-byte discriminator. If it's all zeroes, this means it's uninitialized. If not, it reads the next 4-byte length. If the discriminator matches, it returns the next length bytes. If not, it jumps ahead length bytes and reads the next 8-byte discriminator.

Serialization of variable-length types

The initial example works using the bytemuck crate for zero-copy serialization and deserialization. It's possible to use Borsh by implementing the VariableLenPack trait on your type.

use {
    borsh::{BorshDeserialize, BorshSerialize},
    miraland_program::borsh::{get_instance_packed_len, try_from_slice_unchecked},
    spl_type_length_value::{
        state::{TlvState, TlvStateMut},
        variable_len_pack::VariableLenPack
    },
};
#[derive(Clone, Debug, PartialEq, BorshDeserialize, BorshSerialize)]
struct MyVariableLenType {
    data: String, // variable length type
}
impl SplDiscriminate for MyVariableLenType {
    const SPL_DISCRIMINATOR: ArrayDiscriminator = ArrayDiscriminator::new([5; ArrayDiscriminator::LENGTH]);
}
impl VariableLenPack for MyVariableLenType {
    fn pack_into_slice(&self, dst: &mut [u8]) -> Result<(), ProgramError> {
        borsh::to_writer(&mut dst[..], self).map_err(Into::into)
    }

    fn unpack_from_slice(src: &[u8]) -> Result<Self, ProgramError> {
        try_from_slice_unchecked(src).map_err(Into::into)
    }

    fn get_packed_len(&self) -> Result<usize, ProgramError> {
        get_instance_packed_len(self).map_err(Into::into)
    }
}
let initial_data = "This is a pretty cool test!";
// Allocate exactly the right size for the string, can go bigger if desired
let tlv_size = 4 + initial_data.len();
let account_size = TlvState::get_base_len() + tlv_size;

// Buffer likely comes from a Miraland `miraland_program::account_info::AccountInfo`,
// but this example just uses a vector.
let mut buffer = vec![0; account_size];
let mut state = TlvStateMut::unpack(&mut buffer).unwrap();

// No need to hold onto the bytes since we'll serialize back into the right place
// For this example, let's _not_ allow repeating entries.
let _ = state.alloc::<MyVariableLenType>(tlv_size, false).unwrap();
let my_variable_len = MyVariableLenType {
    data: initial_data.to_string()
};
state.pack_variable_len_value(&my_variable_len).unwrap();
let deser = state.get_first_variable_len_value::<MyVariableLenType>().unwrap();
assert_eq!(deser, my_variable_len);