Version 3.0 (2020-10-09)

Be prepared for the ink! 3.0 release notes because the whole version was basically a rewrite of
all the major components that make up ink!. With our experience gained from previous releases
of ink! we were able to detect weak spots of the design and provided ink! with more tools,
more features and more efficiency as ever. Read more below …

Just. Be. Rust. 3.0

In the 3.0 update we further explored the space for ink! to just feel like it was plain Rust.
With this in mind we changed the syntax slightly in order to better map from ink! to the generated
Rust code. So what users see is mostly what will be generated by ink! later.

In this vein #[ink(storage)] and #[ink(event)] structs as well as #[ink(message)] and
#[ink(constructor)] methods now need to be specified with public visibility (pub).

The #[ink(constructor)] syntax also changes and no longer uses a &mut self receiver but
now follows the natural Rust constructors scheme. So it is no longer possible to shoot
yourself in the foot by accidentally forgetting to initialize some important data structures.

Old ink! 2.0:

#[ink(constructor)]
fn new_erc20(&mut self, initial_supply: Balance) {
    let caller = self.env().caller();
    self.total_supply.set(initial_supply);
    self.balances.insert(caller, initial_supply);
}

New ink! 3.0:

#[ink(constructor)]
pub fn new_erc20(initial_supply: Balance) -> Self {
    let caller = self.env().caller();
    let mut balances = ink_storage::HashMap::new();
    balances.insert(caller, initial_supply);
    Self {
        total_supply: initial_supply,
        balances,
    }
}

Also ink! 3.0 no longer requires a mandatory version field in the header of the ink! module attribute.

Syntactically this is all it takes to port your current ink! smart contracts over to ink! 3.0 syntax.

New Storage Module

The storage module has been reworked entirely.
Also it no longer lives in the ink_core crate but instead is defined as its own ink_storage crate.

In a sense it acts as the standard storage library for ink! smart contracts in that it provides all the
necessary tools and data structures to organize and operate the contract's storage intuitively and efficiently.

Lazy

The most fundamental change in how you should think about data structures provided by the new ink_storage
crate is that they are inherently lazy. We will explain what this means below!
The ink_storage crate provides high-level and low-level lazy data structures.
The difference between high-level and low-level lies in the distinction in how these data structures are aware
of the elements that they operate on. For high-level data structures they are fully aware about the elements
they contains, do all the clean-up by themselves so the user can concentrate on the business logic.
For low-level data structures the responsibility about the elements lies in the hands of the contract author.
Also they operate on cells (Option<T>) instead of entities of type T.
But what does that mean exactly?

The new ink_storage::Lazy type is what corresponds the most to the old ink_core::storage::Value type. Both cache their entities and both act lazily on the storage. This means that a read or write operation is only performed when it really needs to in order to satisfy other inputs.
Data types such as Rust primitives i32 or Rust's very own Vec or data structures can also be used to operate on the contract's storage, however, they will load their contents eagerly which is often not what you want.

An example follows with the below contract storage and a message that operates on either of the two fields.

#[ink(storage)]
pub struct TwoValues {
    offset: i32,
    a: i32,
    b: i32,
}

impl TwoValues {
    #[ink(message)]
    pub fn set(&mut self, which: bool, new_value: i32) {
        match which {
            true  => { self.a = self.offset + new_value; },
            false => { self.b = self.offset + new_value; },
        }
    }
}

Whenever we call TwoValues::set always both a and b are loaded despite the fact the we only operate on one of them at a time. This is very costly since storage accesses are in fact database look-ups.
In order to prevent this eager loading of storage contents we can make use of ink_storage::Lazy or other lazy data structures defined in that crate:

#[ink(storage)]
pub struct TwoValues {
    offset: i32,
    a: ink_storage::Lazy<i32>,
    b: ink_storage::Lazy<i32>,
}

impl TwoValues {
    #[ink(message)]
    pub fn set(&mut self, which: bool, new_value: i32) {
        match which {
            true  => { self.a = offset + new_value; },
            false => { self.b = offset + new_value; },
        }
    }
}

Now a and b are only loaded when the contract really needs their values.
Note that offset remained i32 since it is always needed and could spare the minor overhead of the ink_storage::Lazy wrapper.

HashMap

In the follow we explore the differences between the high-level ink_storage::collections::HashMap
and the low-level ink_storage::lazy::LazyHashMap. Both provide very similar functionality in that they map some generic key to some storage entity.

However, their APIs look very different. Whereas the HashMap provides a rich and high-level API that is comparable to that of Rust's very own HashMap, the LazyHashMap provides only a fraction of the API and also operates on Option<T> values types instead of T directly. It is more similar Solidity mappings than to Rust's HashMap.

The fundamental difference of both data structures is that HashMap is aware of the keys that have been stored in it and thus can reconstruct exactly which elements and storage regions apply to it. This enables it to provide iteration and automated deletion as well as efficient way to defragment its underlying storage to free some storage space again. This goes very well in the vein of Substrate's storage rent model where contracts have to pay for the storage they are using.

Data Structure	level of abstraction	caching	lazy	element type	container
`T`	-	yes	no	`T`	primitive value
`Lazy<T>`	high-level	yes	yes	`T`	single element container
`LazyCell<T>`	low-level	yes	yes	`Option<T>`	single element, no container
`Vec<T>`	high-level	yes	yes	`T`	Rust vector-like container
`LazyIndexMap<T>`	low-level	yes	yes	`Option<T>`	similar to Solidity mapping
`HashMap<K, V>`	high-level	yes	yes	`V` (key type `K`)	Rust map-like container
`LazyHashMap<K, V>`	low-level	yes	yes	`Option<V>` (key type `K`)	similar to Solidity mapping

There are many more! For more information about the specifics please take a look into the ink_storage crate documentation.

Spread & Packed Modes

Storing or loading complex data structures to and from contract storage can be done in many different ways. You could store all information into a single storage cell or you could try to store all information into as many different cells as possible. Both strategies have pros and cons under different conditions.

For example it might be a very good idea to store all the information under the same cell if all the information is very compact. For example when we are dealing with a byte vector that is expected to never be larger than approx a thousand elements it would probably be more efficient if we store all those thousand bytes in the same cell and especially if we often access many of those (or all) in our contract messages.

On the other hand spreading information across as many cells as possible might be much more efficient if we are dealing with big data structures, a lot of information that is not compact, or when messages that operate on the data always only need a small fraction of the whole data.
An example for this use case is if you have a vector of user accounts where each account stores potentially a lot of information, e.g. a 32-byte hash etc and where our messages only every operate on only a few of those at a time.

The ink_storage crate provides the user full control over the strategy or a mix of these two root strategies through some fundamental abstractions that we are briefly presenting to you.

Default: Spreading Mode

By default ink! spreads information to as many cells as possible. For example if you have the following #[ink(storage)] struct every field will live in its own single storage cell. Note that for c all 32 bytes will share the same cell!

#[ink(storage)]
pub struct Spreaded {
    a: i32,
    b: ink_storage::Lazy<i32>,
    c: [u8; 32],
}

Packing Storage

We can alter this behaviour by using the ink_storage::Pack abstraction:

pub struct Spreaded {
    a: i32,
    b: ink_storage::Lazy<i32>,
    c: [u8; 32],
}

#[ink(storage)]
pub struct Packed {
    packed: ink_storage::Pack<Spreaded>,
}

Now all fields of Spreaded will share the same storage cell. This means whenever one of them is stored to or loaded from the contract storage, all of them are stored or loaded. A user has to choose wisely what mode of operation is more suitable for their contract.

These abstractions can be combined in various ways, yielding full control to the users. For example, in the following only a and b share a common storage cell while c lives in its own:

pub struct Spreaded {
    a: i32,
    b: ink_storage::Lazy<i32>,
}

#[ink(storage)]
pub struct Packed {
    packed: ink_storage::Pack<Spreaded>,
    c: [u8; 32],
}

Spreading Array Cells

If we prefer to store all bytes of c into their own storage cell we can make use of the SmallVec data structure. The SmallVec is a high-level data structure that allows to efficiently organize a fixed number of elements similar to a Rust array. However, unlike a Rust array it acts lazily upon the storage and spreads its elements into different cells.

use typenum::U32;

pub struct Spreaded {
    a: i32,
    b: ink_storage::Lazy<i32>,
}

#[ink(storage)]
pub struct Packed {
    packed: ink_storage::Pack<Spreaded>,
    c: SmallVec<u8, U32>,
}

Opting-out of Storage

If you are in need of storing some temporary information across method and message boundaries ink! will have your back with the ink_storage::Memory abstraction. It allows you to simply opt-out of using the storage for the wrapped entitiy at all and thus is very similar to Solidity's very own memory annotation.

An example below:

#[ink(storage)]
pub struct OptedOut {
    a: i32,
    b: ink_storage::Lazy<i32>,
    c: ink_storage::Memory<i32>,
}

The the above example a and b are normal storage entities, however, c on the other hand side will never load from or store to contract storage and will always be reset to the default value of its i32 type for every contract call.
It can be accesses from all ink! messages or methods via self.c but will never manipulate the contract storage and thus acts wonderfully as some shared local information.

Dynamic Storage Allocator

In the previous section we have seen how the default mode of operation is to spread information and how we can opt-in to packing information into single cells via ink_storage::Packed.

However, what if we wanted to store a vector of a vector of i32 for example?
Naturally a user would try to construct this as follows:

use ink_storage::Vec as StorageVec;

#[ink(storage)]
pub struct Matrix {
    values: StorageVec<StorageVec<i32>>,
}

However, this will fail compilation with an error indicating that StorageVec<T> requires for its T to be packed (T: PackedLayout) which StorageVec<T> itself does not since it always stores all of its elements into different cells. The same applies to many other storage data sturctures provided by ink_storage and is a trade-off the ink! team decided for the case of efficiency of the overall system.
Instead what a user can do in order to get their vector-of-vector to be working is to make use of ink!'s dynamic storage allocator capabilities.

For this the contract author has to first enable the feature via:

use ink_lang as ink;

#[ink::contract(dynamic_storage_allocator = true)]
mod matrix {
    // contract code ...
}

And then we can define our Matrix #[ink(storage)] as follows:

use ink_storage::{
    Vec as StorageVec,
    Box as StorageBox,
};

#[ink(storage)]
pub struct Matrix {
    values: StorageVec<StorageBox<StorageVec<i32>>>,
}

With ink_storage::Box<T> we can use a T: SpreadLayout as if it was T: PackedLayout since the ink_storage::Box<T> itself suffices the requirements and can be put into a single contract storage cell. The whole concept works quite similar to how Rust's Box works: by an indirection - contract authors are therefore advised to make use of dynamic storage allocator capabilities only if other ways of dealing with ones problems are not applicable.

Custom Data Sturctures

While the ink_storage crate provides tons of useful utilities and data structures to organize and manipulate the contract's storage contract authors are not limited by its capabilities. By implementing the core SpreadLayout and PackedLayout traits users are able to define their very own custom storage data structures with their own set of requirement and features that work along the ink_storage data structures as long as they fulfill the mere requirements stated by those two traits.

In the future we plan on providing some more ink! workshops and tutorials guiding the approach to design and implement a custom storage data structure.

In Summary

The new ink_storage crate provides everything you need to operate on your contract's storage.
There are low-level and high-level data structures depending on your need of control.
All provided data structures operate lazily on the contract's storage and cache their reads and writes for a more gas efficient storage access.
Users should prefer high-level data structures found in the collections module over the low-level data structures found in the lazy module.
For a list of all the new storage data structure visit ink_storage's documentation.

ink! Attributes

For ink! 3.0 we have added some more useful ink! specific attributes to the table.
All of these ink! attributes are available to specify inside an ink! module.
An ink! module is the module that is flagged by #[ink::contract] containing all the ink! definitions:

use ink_lang as ink;

#[ink::contract]
mod erc20 {
    #[ink(storage)]
    pub struct Erc20 { ... }

    impl Erc20 {
        #[ink(constructor)]
        pub fn new(initial_supply: Balance) -> Self { .. }

        #[ink(constructor)]
        pub fn total_supply(&self) -> Balance { .. }

        // etc. ...
    }
}

We won't be going into the details for any of those but will briefly present the entire set of ink! specific attributes below:

Attribute	Where Applicable	Description
`#[ink(storage)]`	On `struct` definitions.	Defines the ink! storage struct. There can only be one ink! storage definition per contract.
`#[ink(event)]`	On `struct` definitions.	Defines an ink! event. A contract can define multiple such ink! events.
`#[ink(anonymous)]` new	Applicable to ink! events.	Tells the ink! codegen to treat the ink! event as anonmyous which omits the event signature as topic upon emitting. Very similar to anonymous events in Solidity.
`#[ink(topic)]`	Applicate on ink! event field.	Tells the ink! codegen to provide a topic hash for the given field. Every ink! event can only have a limited number of such topic field. Similar semantics as to indexed event arguments in Solidity.
`#[ink(message)]`	Applicable to methods.	Flags a method for the ink! storage struct as message making it available to the API for calling the contract.
`#[ink(constructor)]`	Applicable to method.	Flags a method for the ink! storage struct as constructor making it available to the API for instantiating the contract.
`#[ink(payable)]` new	Applicable to ink! messages.	Allows receiving value as part of the call of the ink! message. ink! constructors are implicitely payable.
`#[ink(selector = "..")]` new	Applicable to ink! messages and ink! constructors.	Specifies a concrete dispatch selector for the flagged entity. This allows a contract author to precisely control the selectors of their APIs making it possible to rename their API without breakage.
`#[ink(namespace = "..")]` new	Applicable to ink! trait implementation blocks.	Changes the resulting selectors of all the ink! messages and ink! constructors within the trait implementation. Allows to disambiguate between trait implementations with overlapping message or constructor names. Use only with great care and consideration!
`#[ink(implementation)]` new	Applicable to ink! implementation blocks.	Tells the ink! codegen that some implementation block shall be granted access to ink! internals even without it containing any ink! messages or ink! constructors.

Merging of ink! Attributes

It is possible to merge attributes that share a common flagged entitiy.
The example below demonstrates this for a payable message with a custom selector.

#[ink(message)]
#[ink(payable)]
#[ink(selector = "0xCAFEBABE")]
pub fn transfer(&mut self, from: AccountId, to: AccountId, value: Balance) -> Result<(), Error> {
    // actual implementation
}

We can also write the above ink! message definition in the following way:

#[ink(message, payable, selector = "0xCAFEBABE")]
pub fn transfer(&mut self, from: AccountId, to: AccountId, value: Balance) -> Result<(), Error> {
    // actual implementation
}

Trait Support

One of the most anticipated features of ink! 3.0 is its Rust trait support.
Through the new #[ink::trait_definition] proc. macro it is now possible to define your very own trait definitions that are then implementable by ink! smart contracts.

This allows to defined shared smart contract interfaces to different concrete implementations.
Note that this ink! trait definition can be defined anywhere, even in another crate!

Example

Defined in the base_erc20.rs module.

use ink_lang as ink;

#[ink::trait_definition]
pub trait BaseErc20 {
    /// Creates a new ERC-20 contract and initializes it with the initial supply for the instantiator.
    #[ink(constructor)]
    pub fn new(initial_supply: Balance) -> Self;

    /// Returns the total supply.
    #[ink(message)]
    pub fn total_supply(&self) -> Balance;

    /// Transfers `amount` from caller to `to`.
    #[ink(message, payable)]
    pub fn transfer(&mut self, to: AccountId, amount: Balance);
}

An ink! smart contract definition can then implement this trait definition as follows:

use ink_lang as ink;

#[ink::contract]
mod erc20 {
    use base_erc20::BaseErc20;

    #[ink(storage)]
    pub struct Erc20 {
        total_supply: Balance,
        // more fields ...
    }

    impl BaseErc20 for Erc20 {
        #[ink(constructor)]
        pub fn new(initial_supply: Balance) -> Self {
            // implementation ...
        }

        #[ink(message)]
        pub fn total_supply(&self) -> Balance {
            // implementation ...
        }

        #[ink(message, payable)]
        pub fn transfer(&mut self, to: AccountId, amount: Balance) {
            // implementation ...
        }
    }
}

Calling the above Erc20 explicitely through its trait implementation can be done just as if it was normal Rust code:

// --- Instantiating the ERC-20 contract:
//
let mut erc20 = <Erc20 as BaseErc20>::new(1000);
// --- Is just the same as:
use base_erc20::BaseErc20;
let mut erc20 = Erc20::new(1000);

// --- Retrieving the total supply:
//
assert_eq!(<Erc20 as BaseErc20>::total_supply(&erc20), 1000);
// --- Is just the same as:
use base_erc20::BaseErc20;
assert_eq!(erc20.total_supply(), 1000);

There are still many limitations to ink! trait definitions and trait implementations.
For example it is not possible to define associated constants or types or have default implemented methods.
These limitations exist because of technical intricacies, however, please expect that many of those will be tackled in future ink! releases.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

ink! 3.0.0 RC 1