Pallet Coupling
The term coupling is often used to describe the degree to which two software modules depend on each other. For example, in object-oriented programming tight coupling and loose coupling are used to describe the relationship between objects classes:
Tight coupling is when two groups of classes are dependent on each other.
Loose coupling is when a class uses an interface that another class exposes.
In Substrate, tight pallet coupling and loose pallet coupling are used to describe how a pallet can call functions in another pallet. Both techniques achieve the same thing in different ways, each having certain trade-offs.
Tightly coupled pallets
Because tight coupling makes working with pallets less flexible and extensible, you would only use tight pallet coupling if a pallet requires inheriting its coupled counterpart as a whole rather than specific types or methods.
When writing a pallet that requires tight coupling, you explicitly specify the pallet's Config
trait to be bound by the Config
trait of the pallet to couple with.
All FRAME pallets are tightly coupled to the frame_system
pallet. The following example illustrates how to use the Config
trait of a pallet called some_pallet
to be tightly coupled with the frame_system
pallet:
This is very similar to using class inheritance in object-oriented programming. Supplying this trait bound implies that this pallet can only be installed in a runtime that also has some_pallet
pallet installed. Similar to frame_system
, the tight coupling in this example would require you to specify some_pallet
in the coupled pallet's Cargo.toml file.
Tight coupling has several disadvantages developers should take into account:
Maintainability: changes in one pallet will often result in needing to modify the other pallet.
Reusability: both modules must be included for one to be used, making it more difficult to integrate a tightly coupled pallet.
Loosely coupled pallets
In loose pallet coupling, you can specify the traits and function interfaces that certain types need to be bound by.
The actual implementation of such types happens outside of the pallet during the runtime configuration—typically as code defined in the /runtime/src/lib.rs
file. With loose coupling, you can use types and interfaces from another pallet that has implemented the traits, or you can declare a totally new struct, implement those traits, and assign it when implementing the pallet in runtime.
As an example, assume you have a pallet that can access account balances and make transfers to another account. This pallet defines a Currency
trait, which has an abstract function interface that will implement the actual transfer logic later.
In a second pallet, you define the MyCurrency
associated type and bind it by Currency<Self::AccountId>
trait so that you can use the balance transfer logic by calling T::MyCurrency::transfer(...)
.
Notice that, at this point, you have not specified how the Currency::transfer()
logic will be implemented. It is only agreed upon that it will be implemented somewhere.
Now, you can use the runtime configuration—runtime/src/lib.rs
—to implement the pallet and specify the type to be Balances
.
The Balances
type is specified in construct_runtime!
macro as part of the pallet_balances
that implements the Currency
trait.
With the implementation provided by the runtime, you can make use of Currency<AccountId>
trait in your loosely coupled pallet.
Many FRAME pallets are coupled to this Currency trait in this way.
Choosing a pallet coupling strategy
In general, loose coupling provides more flexibility than tight coupling and is considered a better practice from a system design perspective. It guarantees better maintainability, reusability, and extensibility of your code. However, tight coupling can be useful for pallets that are less complex or that have more overlap in methods and types than differences.
In FRAME, there are two pallets that are tightly coupled to pallet_treasury
:
As a general rule, the more complex a pallet is, the less desirable it is to tightly couple it. This evokes a concept in computer science called cohesion, a metric used to examine the overall quality of a software system.
Where to go next
Last updated