Attribute Macro frame_support::pallet

source ·
#[pallet]
Expand description

The pallet attribute macro defines a pallet that can be used with construct_runtime!. It must be attached to a module named pallet as follows:

#[pallet]
pub mod pallet {
	...
}

Note that various types can be automatically imported using frame_support::pallet_prelude and frame_system::pallet_prelude:

#[pallet]
pub mod pallet {
	use frame_support::pallet_prelude::*;
	use frame_system::pallet_prelude::*;
	...
}

pallet::* Attributes

The pallet macro will parse any items within your pallet module that are annotated with #[pallet::*] attributes. Some of these attributes are mandatory and some are optional, and they can attach to different types of items within your pallet depending on the attribute in question. The full list of #[pallet::*] attributes is shown below in the order in which they are mentioned in this document:

Note that at compile-time, the #[pallet] macro will analyze and expand all of these attributes, ultimately removing their AST nodes before they can be parsed as real attribute macro calls. This means that technically we do not need attribute macro definitions for any of these attributes, however, for consistency and discoverability reasons, we still maintain stub attribute macro definitions for all of these attributes in the pallet_macros module which is automatically included in all pallets as part of the pallet prelude. The actual “work” for all of these attribute macros can be found in the macro expansion for #[pallet].

Also note that in this document, pallet attributes are explained using the syntax of non-instantiable pallets. For an example of an instantiable pallet, see this example.

Dev Mode (#[pallet(dev_mode)])

Specifying the argument dev_mode on the #[pallet] or #[frame_support::pallet] attribute attached to your pallet module will allow you to enable dev mode for a pallet. The aim of dev mode is to loosen some of the restrictions and requirements placed on production pallets for easy tinkering and development. Dev mode pallets should not be used in production. Enabling dev mode has the following effects:

  • Weights no longer need to be specified on every #[pallet::call] declaration. By default, dev mode pallets will assume a weight of zero (0) if a weight is not specified. This is equivalent to specifying #[weight(0)] on all calls that do not specify a weight.
  • All storages are marked as unbounded, meaning you do not need to implement MaxEncodedLen on storage types. This is equivalent to specifying #[pallet::unbounded] on all storage type definitions.

Note that the dev_mode argument can only be supplied to the #[pallet] or #[frame_support::pallet] attribute macro that encloses your pallet module. This argument cannot be specified anywhere else, including but not limited to the #[pallet::pallet] attribute macro.

WARNING:
You should not deploy or use dev mode pallets in production. Doing so can break your chain
and therefore should never be done. Once you are done tinkering, you should remove the
'dev_mode' argument from your #[pallet] declaration and fix any compile errors before
attempting to use your pallet in a production scenario.

Pallet struct placeholder: #[pallet::pallet] (mandatory)

The pallet struct placeholder #[pallet::pallet] is mandatory and allows you to specify pallet information.

The struct must be defined as follows:

#[pallet::pallet]
pub struct Pallet<T>(_);

I.e. a regular struct definition named Pallet, with generic T and no where clause.

Macro expansion:

The macro adds this attribute to the struct definition:

#[derive(
	frame_support::CloneNoBound,
	frame_support::EqNoBound,
	frame_support::PartialEqNoBound,
	frame_support::RuntimeDebugNoBound,
)]

and replaces the type _ with PhantomData<T>. It also implements on the pallet:

  • GetStorageVersion
  • OnGenesis: contains some logic to write the pallet version into storage.
  • PalletErrorTypeInfo: provides the type information for the pallet error, if defined.

It declares type Module type alias for Pallet, used by construct_runtime.

It implements PalletInfoAccess on Pallet to ease access to pallet information given by frame_support::traits::PalletInfo. (The implementation uses the associated type frame_system::Config::PalletInfo).

It implements StorageInfoTrait on Pallet which give information about all storages.

If the attribute generate_store is set then the macro creates the trait Store and implements it on Pallet.

If the attribute set_storage_max_encoded_len is set then the macro calls StorageInfoTrait for each storage in the implementation of StorageInfoTrait for the pallet. Otherwise it implements StorageInfoTrait for the pallet using the PartialStorageInfoTrait implementation of storages.

Config trait: #[pallet::config] (mandatory)

The mandatory attribute #[pallet::config] defines the configurable options for the pallet.

Item must be defined as:

#[pallet::config]
pub trait Config: frame_system::Config + $optionally_some_other_supertraits
$optional_where_clause
{
...
}

I.e. a regular trait definition named Config, with the supertrait frame_system::pallet::Config, and optionally other supertraits and a where clause. (Specifying other supertraits here is known as tight coupling)

The associated type RuntimeEvent is reserved. If defined, it must have the bounds From<Event> and IsType<<Self as frame_system::Config>::RuntimeEvent>.

pallet::event must be present if RuntimeEvent exists as a config item in your #[pallet::config].

Also see pallet::config

pallet::constant

The #[pallet::constant] attribute can be used to add an associated type trait bounded by Get from pallet::config into metadata, e.g.:

#[pallet::config]
pub trait Config: frame_system::Config {
	#[pallet::constant]
	type Foo: Get<u32>;
}

Also see pallet::constant

pallet::disable_frame_system_supertrait_check

To bypass the frame_system::Config supertrait check, use the attribute pallet::disable_frame_system_supertrait_check, e.g.:

#[pallet::config]
#[pallet::disable_frame_system_supertrait_check]
pub trait Config: pallet_timestamp::Config {}

NOTE: Bypassing the frame_system::Config supertrait check is typically desirable when you want to write an alternative to the frame_system pallet.

Also see pallet::disable_frame_system_supertrait_check

Macro expansion:

The macro expands pallet constant metadata with the information given by #[pallet::constant].

pallet::generate_store($vis trait Store)

To generate a Store trait associating all storages, annotate your Pallet struct with the attribute #[pallet::generate_store($vis trait Store)], e.g.:

#[pallet::pallet]
#[pallet::generate_store(pub(super) trait Store)]
pub struct Pallet<T>(_);

More precisely, the Store trait contains an associated type for each storage. It is implemented for Pallet allowing access to the storage from pallet struct.

Thus when defining a storage named Foo, it can later be accessed from Pallet using <Pallet as Store>::Foo.

NOTE: this attribute is only valid when applied directly to your Pallet struct definition.

Also see pallet::generate_store.

pallet::generate_storage_info

To generate the full storage info (used for PoV calculation) use the attribute #[pallet::generate_storage_info], e.g.:

#[pallet::pallet]
#[pallet::generate_storage_info]
pub struct Pallet<T>(_);

This requires all storage items to implement the trait traits::StorageInfoTrait, thus all keys and value types must be bound by pallet_prelude::MaxEncodedLen. Individual storages can opt-out from this constraint by using #[pallet::unbounded] (see #[pallet::storage] for more info).

Also see pallet::generate_storage_info

pallet::storage_version

Because the pallet::pallet macro implements traits::GetStorageVersion, the current storage version needs to be communicated to the macro. This can be done by using the pallet::storage_version attribute:

const STORAGE_VERSION: StorageVersion = StorageVersion::new(5);

#[pallet::pallet]
#[pallet::storage_version(STORAGE_VERSION)]
pub struct Pallet<T>(_);

If not present, the current storage version is set to the default value.

Also see pallet::storage_version

Hooks: #[pallet::hooks] (optional)

The pallet::hooks attribute allows you to specify a Hooks implementation for Pallet that specifies pallet-specific logic.

The item the attribute attaches to must be defined as follows:

#[pallet::hooks]
impl<T: Config> Hooks<BlockNumberFor<T>> for Pallet<T> $optional_where_clause {
    ...
}

I.e. a regular trait implementation with generic bound: T: Config, for the trait Hooks<BlockNumberFor<T>> (they are defined in preludes), for the type Pallet<T> and with an optional where clause.

If no #[pallet::hooks] exists, then the following default implementation is automatically generated:

#[pallet::hooks]
impl<T: Config> Hooks<BlockNumberFor<T>> for Pallet<T> {}

Also see pallet::hooks

Call: #[pallet::call] (optional)

Implementation of pallet dispatchables.

Item must be defined as:

#[pallet::call]
impl<T: Config> Pallet<T> {
	/// $some_doc
	#[pallet::weight($ExpressionResultingInWeight)]
	pub fn $fn_name(
		origin: OriginFor<T>,
		$some_arg: $some_type,
		// or with compact attribute: #[pallet::compact] $some_arg: $some_type,
		...
	) -> DispatchResultWithPostInfo { // or `-> DispatchResult`
		...
	}
	...
}

I.e. a regular type implementation, with generic T: Config, on type Pallet<T>, with an optional where clause.

#[pallet::weight($expr)]

Each dispatchable needs to define a weight with #[pallet::weight($expr)] attribute, the first argument must be origin: OriginFor<T>.

Also see pallet::weight

#[pallet::compact] $some_arg: $some_type

Compact encoding for arguments can be achieved via #[pallet::compact]. The function must return a DispatchResultWithPostInfo or DispatchResult.

Also see pallet::compact

#[pallet::call_index($idx)]

Each dispatchable may also be annotated with the #[pallet::call_index($idx)] attribute, which explicitly defines the codec index for the dispatchable function in the Call enum.

All call indexes start from 0, until it encounters a dispatchable function with a defined call index. The dispatchable function that lexically follows the function with a defined call index will have that call index, but incremented by 1, e.g. if there are 3 dispatchable functions fn foo, fn bar and fn qux in that order, and only fn bar has a call index of 10, then fn qux will have an index of 11, instead of 1.

WARNING: modifying dispatchables, changing their order, removing some, etc., must be done with care. Indeed this will change the outer runtime call type (which is an enum with one variant per pallet), this outer runtime call can be stored on-chain (e.g. in pallet-scheduler). Thus migration might be needed. To mitigate against some of this, the #[pallet::call_index($idx)] attribute can be used to fix the order of the dispatchable so that the Call enum encoding does not change after modification. As a general rule of thumb, it is therefore adventageous to always add new calls to the end so you can maintain the existing order of calls.

Also see pallet::call_index

Extra constants: #[pallet::extra_constants] (optional)

Allows you to define some extra constants to be added into constant metadata.

Item must be defined as:

#[pallet::extra_constants]
impl<T: Config> Pallet<T> where $optional_where_clause {
	/// $some_doc
	$vis fn $fn_name() -> $some_return_type {
		...
	}
	...
}

I.e. a regular rust impl block with some optional where clause and functions with 0 args, 0 generics, and some return type.

Macro expansion

The macro add some extra constants to pallet constant metadata.

Also see: pallet::extra_constants

Error: #[pallet::error] (optional)

The #[pallet::error] attribute allows you to define an error enum that will be returned from the dispatchable when an error occurs. The information for this error type is then stored in metadata.

Item must be defined as:

#[pallet::error]
pub enum Error<T> {
	/// $some_optional_doc
	$SomeFieldLessVariant,
	/// $some_more_optional_doc
	$SomeVariantWithOneField(FieldType),
	...
}

I.e. a regular enum named Error, with generic T and fieldless or multiple-field variants.

Any field type in the enum variants must implement scale_info::TypeInfo in order to be properly used in the metadata, and its encoded size should be as small as possible, preferably 1 byte in size in order to reduce storage size. The error enum itself has an absolute maximum encoded size specified by MAX_MODULE_ERROR_ENCODED_SIZE.

(1 byte can still be 256 different errors. The more specific the error, the easier it is to diagnose problems and give a better experience to the user. Don’t skimp on having lots of individual error conditions.)

Field types in enum variants must also implement PalletError, otherwise the pallet will fail to compile. Rust primitive types have already implemented the PalletError trait along with some commonly used stdlib types such as Option and PhantomData, and hence in most use cases, a manual implementation is not necessary and is discouraged.

The generic T must not bound anything and a where clause is not allowed. That said, bounds and/or a where clause should not needed for any use-case.

Also see: pallet::error

Event: #[pallet::event] (optional)

Allows you to define pallet events. Pallet events are stored under the system / events key when the block is applied (and then replaced when the next block writes it’s events).

The Event enum must be defined as follows:

#[pallet::event]
#[pallet::generate_deposit($visibility fn deposit_event)] // Optional
pub enum Event<$some_generic> $optional_where_clause {
	/// Some doc
	$SomeName($SomeType, $YetanotherType, ...),
	...
}

I.e. an enum (with named or unnamed fields variant), named Event, with generic: none or T or T: Config, and optional w here clause.

Each field must implement Clone, Eq, PartialEq, Encode, Decode, and Debug (on std only). For ease of use, bound by the trait Member, available in frame_support::pallet_prelude.

Also see pallet::event

#[pallet::generate_deposit($visibility fn deposit_event)]

The attribute #[pallet::generate_deposit($visibility fn deposit_event)] generates a helper function on Pallet that handles deposit events.

NOTE: For instantiable pallets, the event must be generic over T and I.

Also see pallet::generate_deposit

Storage: #[pallet::storage] (optional)

The #[pallet::storage] attribute lets you define some abstract storage inside of runtime storage and also set its metadata. This attribute can be used multiple times.

Item should be defined as:

#[pallet::storage]
#[pallet::getter(fn $getter_name)] // optional
$vis type $StorageName<$some_generic> $optional_where_clause
	= $StorageType<$generic_name = $some_generics, $other_name = $some_other, ...>;

or with unnamed generic:

#[pallet::storage]
#[pallet::getter(fn $getter_name)] // optional
$vis type $StorageName<$some_generic> $optional_where_clause
	= $StorageType<_, $some_generics, ...>;

I.e. it must be a type alias, with generics: T or T: Config. The aliased type must be one of StorageValue, StorageMap or StorageDoubleMap. The generic arguments of the storage type can be given in two manners: named and unnamed. For named generic arguments, the name for each argument should match the name defined for it on the storage struct:

  • StorageValue expects Value and optionally QueryKind and OnEmpty,
  • StorageMap expects Hasher, Key, Value and optionally QueryKind and OnEmpty,
  • CountedStorageMap expects Hasher, Key, Value and optionally QueryKind and OnEmpty,
  • StorageDoubleMap expects Hasher1, Key1, Hasher2, Key2, Value and optionally QueryKind and OnEmpty.

For unnamed generic arguments: Their first generic must be _ as it is replaced by the macro and other generic must declared as a normal generic type declaration.

The Prefix generic written by the macro is generated using PalletInfo::name::<Pallet<..>>() and the name of the storage type. E.g. if runtime names the pallet “MyExample” then the storage type Foo<T> = ... should use the prefix: Twox128(b"MyExample") ++ Twox128(b"Foo").

For the CountedStorageMap variant, the Prefix also implements CountedStorageMapInstance. It also associates a CounterPrefix, which is implemented the same as above, but the storage prefix is prepend with "CounterFor". E.g. if runtime names the pallet “MyExample” then the storage type Foo<T> = CountedStorageaMap<...> will store its counter at the prefix: Twox128(b"MyExample") ++ Twox128(b"CounterForFoo").

E.g:

#[pallet::storage]
pub(super) type MyStorage<T> = StorageMap<Hasher = Blake2_128Concat, Key = u32, Value = u32>;

In this case the final prefix used by the map is Twox128(b"MyExample") ++ Twox128(b"OtherName").

Also see pallet::storage

#[pallet::getter(fn $my_getter_fn_name)] (optional)

The optional attribute #[pallet::getter(fn $my_getter_fn_name)] allows you to define a getter function on Pallet.

Also see pallet::getter

#[pallet::storage_prefix = "SomeName"] (optional)

The optional attribute #[pallet::storage_prefix = "SomeName"] allows you to define the storage prefix to use, see how Prefix generic is implemented above. This is helpful if you wish to rename the storage field but don’t want to perform a migration.

E.g:

#[pallet::storage]
#[pallet::storage_prefix = "foo"]
#[pallet::getter(fn my_storage)]
pub(super) type MyStorage<T> = StorageMap<Hasher = Blake2_128Concat, Key = u32, Value = u32>;

or

#[pallet::storage]
#[pallet::getter(fn my_storage)]
pub(super) type MyStorage<T> = StorageMap<_, Blake2_128Concat, u32, u32>;

Also see pallet::storage_prefix

#[pallet::unbounded] (optional)

The optional attribute #[pallet::unbounded] declares the storage as unbounded. When implementating the storage info (when #[pallet::generate_storage_info] is specified on the pallet struct placeholder), the size of the storage will be declared as unbounded. This can be useful for storage which can never go into PoV (Proof of Validity).

Also see pallet::unbounded

#[pallet::whitelist_storage] (optional)

The optional attribute #[pallet::whitelist_storage] will declare the storage as whitelisted from benchmarking.

See pallet::whitelist_storage for more info.

#[cfg(..)] (for storage)

The optional attributes #[cfg(..)] allow conditional compilation for the storage.

E.g:

#[cfg(feature = "my-feature")]
#[pallet::storage]
pub(super) type MyStorage<T> = StorageValue<Value = u32>;

All the cfg attributes are automatically copied to the items generated for the storage, i.e. the getter, storage prefix, and the metadata element etc.

Any type placed as the QueryKind parameter must implement frame_support::storage::types::QueryKindTrait. There are 3 implementations of this trait by default:

  1. OptionQuery, the default QueryKind used when this type parameter is omitted. Specifying this as the QueryKind would cause storage map APIs that return a QueryKind to instead return an Option, returning Some when a value does exist under a specified storage key, and None otherwise.
  2. ValueQuery causes storage map APIs that return a QueryKind to instead return the value type. In cases where a value does not exist under a specified storage key, the OnEmpty type parameter on QueryKindTrait is used to return an appropriate value.
  3. ResultQuery causes storage map APIs that return a QueryKind to instead return a Result<T, E>, with T being the value type and E being the pallet error type specified by the #[pallet::error] attribute. In cases where a value does not exist under a specified storage key, an Err with the specified pallet error variant is returned.

NOTE: If the QueryKind generic parameter is still generic at this stage or is using some type alias then the generation of the getter might fail. In this case the getter can be implemented manually.

NOTE: The generic Hasher must implement the StorageHasher trait (or the type is not usable at all). We use StorageHasher::METADATA for the metadata of the hasher of the storage item. Thus generic hasher is supported.

Macro expansion

For each storage item the macro generates a struct named _GeneratedPrefixForStorage$NameOfStorage, and implements StorageInstance on it using the pallet and storage name. It then uses it as the first generic of the aliased type. For CountedStorageMap, CountedStorageMapInstance is implemented, and another similar struct is generated.

For a named generic, the macro will reorder the generics, and remove the names.

The macro implements the function storage_metadata on the Pallet implementing the metadata for all storage items based on their kind:

  • for a storage value, the type of the value is copied into the metadata
  • for a storage map, the type of the values and the key’s type is copied into the metadata
  • for a storage double map, the type of the values, and the types of key1 and key2 are copied into the metadata.

Type value: #[pallet::type_value] (optional)

The #[pallet::type_value] attribute lets you define a struct implementing the Get trait to ease use of storage types. This attribute is meant to be used alongside #[pallet::storage] to define a storage’s default value. This attribute can be used multiple times.

Item must be defined as:

#[pallet::type_value]
fn $MyDefaultName<$some_generic>() -> $default_type $optional_where_clause { $expr }

I.e.: a function definition with generics none or T: Config and a returned type.

E.g.:

#[pallet::type_value]
fn MyDefault<T: Config>() -> T::Balance { 3.into() }

Also see pallet::type_value

Genesis config: #[pallet::genesis_config] (optional)

The #[pallet::genesis_config] attribute allows you to define the genesis configuration for the pallet.

Item is defined as either an enum or a struct. It needs to be public and implement the trait GenesisBuild with #[pallet::genesis_build]. The type generics are constrained to be either none, or T or T: Config.

E.g:

#[pallet::genesis_config]
pub struct GenesisConfig<T: Config> {
	_myfield: BalanceOf<T>,
}

Also see pallet::genesis_config

Genesis build: #[pallet::genesis_build] (optional)

The #[pallet::genesis_build] attribute allows you to define how genesis_configuration is built. This takes as input the GenesisConfig type (as self) and constructs the pallet’s initial state.

The impl must be defined as:

#[pallet::genesis_build]
impl<T: Config> GenesisBuild<T> for GenesisConfig<$maybe_generics> {
	fn build(&self) { $expr }
}

I.e. a trait implementation with generic T: Config, of trait GenesisBuild<T> on type GenesisConfig with generics none or T.

E.g.:

#[pallet::genesis_build]
impl<T: Config> GenesisBuild<T> for GenesisConfig {
	fn build(&self) {}
}

Also see pallet::genesis_build

Inherent: #[pallet::inherent] (optional)

The #[pallet::inherent] attribute allows the pallet to provide some inherent. An inherent is some piece of data that is inserted by a block authoring node at block creation time and can either be accepted or rejected by validators based on whether the data falls within an acceptable range.

The most common inherent is the timestamp that is inserted into every block. Since there is no way to validate timestamps, validators simply check that the timestamp reported by the block authoring node falls within an acceptable range.

Item must be defined as:

#[pallet::inherent]
impl<T: Config> ProvideInherent for Pallet<T> {
	// ... regular trait implementation
}

I.e. a trait implementation with bound T: Config, of trait ProvideInherent for type Pallet<T>, and some optional where clause.

Also see pallet::inherent

Validate unsigned: #[pallet::validate_unsigned] (optional)

The #[pallet::validate_unsigned] attribute allows the pallet to validate some unsigned transaction:

Item must be defined as:

#[pallet::validate_unsigned]
impl<T: Config> ValidateUnsigned for Pallet<T> {
	// ... regular trait implementation
}

I.e. a trait implementation with bound T: Config, of trait ValidateUnsigned for type Pallet<T>, and some optional where clause.

NOTE: There is also the sp_runtime::traits::SignedExtension trait that can be used to add some specific logic for transaction validation.

Also see pallet::validate_unsigned

Origin: #[pallet::origin] (optional)

The #[pallet::origin] attribute allows you to define some origin for the pallet.

Item must be either a type alias, an enum, or a struct. It needs to be public.

E.g.:

#[pallet::origin]
pub struct Origin<T>(PhantomData<(T)>);

WARNING: modifying origin changes the outer runtime origin. This outer runtime origin can be stored on-chain (e.g. in pallet-scheduler), thus any change must be done with care as it might require some migration.

NOTE: for instantiable pallets, the origin must be generic over T and I.

Also see pallet::origin

General notes on instantiable pallets

An instantiable pallet is one where Config is generic, i.e. Config<I>. This allows runtime to implement multiple instances of the pallet, by using different types for the generic. This is the sole purpose of the generic I, but because PalletInfo requires the Pallet placeholder to be static, it is important to bound by 'static whenever PalletInfo can be used. Additionally, in order to make an instantiable pallet usable as a regular pallet without an instance, it is important to bound by = () on every type.

Thus impl bound looks like impl<T: Config<I>, I: 'static>, and types look like SomeType<T, I=()> or SomeType<T: Config<I>, I: 'static = ()>.

Example of a non-instantiable pallet

pub use pallet::*; // reexport in crate namespace for `construct_runtime!`

#[frame_support::pallet]
// NOTE: The name of the pallet is provided by `construct_runtime` and is used as
// the unique identifier for the pallet's storage. It is not defined in the pallet itself.
pub mod pallet {
	use frame_support::pallet_prelude::*; // Import various types used in the pallet definition
	use frame_system::pallet_prelude::*; // Import some system helper types.

	type BalanceOf<T> = <T as Config>::Balance;

	// Define the generic parameter of the pallet
	// The macro parses `#[pallet::constant]` attributes and uses them to generate metadata
	// for the pallet's constants.
	#[pallet::config]
	pub trait Config: frame_system::Config {
		#[pallet::constant] // put the constant in metadata
		type MyGetParam: Get<u32>;
		type Balance: Parameter + MaxEncodedLen + From<u8>;
		type RuntimeEvent: From<Event<Self>> + IsType<<Self as frame_system::Config>::RuntimeEvent>;
	}

	// Define some additional constant to put into the constant metadata.
	#[pallet::extra_constants]
	impl<T: Config> Pallet<T> {
		/// Some description
		fn exra_constant_name() -> u128 { 4u128 }
	}

	// Define the pallet struct placeholder, various pallet function are implemented on it.
	#[pallet::pallet]
	#[pallet::generate_store(pub(super) trait Store)]
	pub struct Pallet<T>(_);

	// Implement the pallet hooks.
	#[pallet::hooks]
	impl<T: Config> Hooks<BlockNumberFor<T>> for Pallet<T> {
		fn on_initialize(_n: BlockNumberFor<T>) -> Weight {
			unimplemented!();
		}

		// can implement also: on_finalize, on_runtime_upgrade, offchain_worker, ...
		// see `Hooks` trait
	}

	// Declare Call struct and implement dispatchables.
	//
	// WARNING: Each parameter used in functions must implement: Clone, Debug, Eq, PartialEq,
	// Codec.
	//
	// The macro parses `#[pallet::compact]` attributes on function arguments and implements
	// the `Call` encoding/decoding accordingly.
	#[pallet::call]
	impl<T: Config> Pallet<T> {
		/// Doc comment put in metadata
		#[pallet::weight(0)] // Defines weight for call (function parameters are in scope)
		pub fn toto(
			origin: OriginFor<T>,
			#[pallet::compact] _foo: u32,
		) -> DispatchResultWithPostInfo {
			let _ = origin;
			unimplemented!();
		}
	}

	// Declare the pallet `Error` enum (this is optional).
	// The macro generates error metadata using the doc comment on each variant.
	#[pallet::error]
	pub enum Error<T> {
		/// doc comment put into metadata
		InsufficientProposersBalance,
	}

	// Declare pallet Event enum (this is optional).
	//
	// WARNING: Each type used in variants must implement: Clone, Debug, Eq, PartialEq, Codec.
	//
	// The macro generates event metadata, and derive Clone, Debug, Eq, PartialEq and Codec
	#[pallet::event]
	// Generate a funciton on Pallet to deposit an event.
	#[pallet::generate_deposit(pub(super) fn deposit_event)]
	pub enum Event<T: Config> {
		/// doc comment put in metadata
		// `<T as frame_system::Config>::AccountId` is not defined in metadata list, the last
		// Thus the metadata is `<T as frame_system::Config>::AccountId`.
		Proposed(<T as frame_system::Config>::AccountId),
		/// doc
		// here metadata will be `Balance` as define in metadata list
		Spending(BalanceOf<T>),
		// here metadata will be `Other` as define in metadata list
		Something(u32),
	}

	// Define a struct which implements `frame_support::traits::Get<T::Balance>` (optional).
	#[pallet::type_value]
	pub(super) fn MyDefault<T: Config>() -> T::Balance { 3.into() }

	// Declare a storage item. Any amount of storage items can be declared (optional).
	//
	// Is expected either `StorageValue`, `StorageMap` or `StorageDoubleMap`.
	// The macro generates the prefix type and replaces the first generic `_`.
	//
	// The macro expands the metadata for the storage item with the type used:
	// * for a storage value the type of the value is copied into the metadata
	// * for a storage map the type of the values and the type of the key is copied into the metadata
	// * for a storage double map the types of the values and keys are copied into the
	//   metadata.
	//
	// NOTE: The generic `Hasher` must implement the `StorageHasher` trait (or the type is not
	// usable at all). We use [`StorageHasher::METADATA`] for the metadata of the hasher of the
	// storage item. Thus generic hasher is supported.
	#[pallet::storage]
	pub(super) type MyStorageValue<T: Config> =
		StorageValue<Value = T::Balance, QueryKind = ValueQuery, OnEmpty = MyDefault<T>>;

	// Another storage declaration
	#[pallet::storage]
	#[pallet::getter(fn my_storage)]
	#[pallet::storage_prefix = "SomeOtherName"]
	pub(super) type MyStorage<T> =
		StorageMap<Hasher = Blake2_128Concat, Key = u32, Value = u32>;

	// Declare the genesis config (optional).
	//
	// The macro accepts either a struct or an enum; it checks that generics are consistent.
	//
	// Type must implement the `Default` trait.
	#[pallet::genesis_config]
	#[derive(Default)]
	pub struct GenesisConfig {
		_myfield: u32,
	}

	// Declare genesis builder. (This is need only if GenesisConfig is declared)
	#[pallet::genesis_build]
	impl<T: Config> GenesisBuild<T> for GenesisConfig {
		fn build(&self) {}
	}

	// Declare a pallet origin (this is optional).
	//
	// The macro accept type alias or struct or enum, it checks generics are consistent.
	#[pallet::origin]
	pub struct Origin<T>(PhantomData<T>);

	// Declare validate_unsigned implementation (this is optional).
	#[pallet::validate_unsigned]
	impl<T: Config> ValidateUnsigned for Pallet<T> {
		type Call = Call<T>;
		fn validate_unsigned(
			source: TransactionSource,
			call: &Self::Call
		) -> TransactionValidity {
			Err(TransactionValidityError::Invalid(InvalidTransaction::Call))
		}
	}

	// Declare inherent provider for pallet (this is optional).
	#[pallet::inherent]
	impl<T: Config> ProvideInherent for Pallet<T> {
		type Call = Call<T>;
		type Error = InherentError;

		const INHERENT_IDENTIFIER: InherentIdentifier = INHERENT_IDENTIFIER;

		fn create_inherent(_data: &InherentData) -> Option<Self::Call> {
			unimplemented!();
		}

		fn is_inherent(_call: &Self::Call) -> bool {
			unimplemented!();
		}
	}

	// Regular rust code needed for implementing ProvideInherent trait

	#[derive(codec::Encode, sp_runtime::RuntimeDebug)]
	#[cfg_attr(feature = "std", derive(codec::Decode))]
	pub enum InherentError {
	}

	impl sp_inherents::IsFatalError for InherentError {
		fn is_fatal_error(&self) -> bool {
			unimplemented!();
		}
	}

	pub const INHERENT_IDENTIFIER: sp_inherents::InherentIdentifier = *b"testpall";
}

Example of an instantiable pallet

pub use pallet::*;

#[frame_support::pallet]
pub mod pallet {
	use frame_support::pallet_prelude::*;
	use frame_system::pallet_prelude::*;

	type BalanceOf<T, I = ()> = <T as Config<I>>::Balance;

	#[pallet::config]
	pub trait Config<I: 'static = ()>: frame_system::Config {
		#[pallet::constant]
		type MyGetParam: Get<u32>;
		type Balance: Parameter + MaxEncodedLen + From<u8>;
		type RuntimeEvent: From<Event<Self, I>> + IsType<<Self as frame_system::Config>::RuntimeEvent>;
	}

	#[pallet::extra_constants]
	impl<T: Config<I>, I: 'static> Pallet<T, I> {
		/// Some description
		fn exra_constant_name() -> u128 { 4u128 }
	}

	#[pallet::pallet]
	#[pallet::generate_store(pub(super) trait Store)]
	pub struct Pallet<T, I = ()>(PhantomData<(T, I)>);

	#[pallet::hooks]
	impl<T: Config<I>, I: 'static> Hooks<BlockNumberFor<T>> for Pallet<T, I> {
	}

	#[pallet::call]
	impl<T: Config<I>, I: 'static> Pallet<T, I> {
		/// Doc comment put in metadata
		#[pallet::weight(0)]
		pub fn toto(origin: OriginFor<T>, #[pallet::compact] _foo: u32) -> DispatchResultWithPostInfo {
			let _ = origin;
			unimplemented!();
		}
	}

	#[pallet::error]
	pub enum Error<T, I = ()> {
		/// doc comment put into metadata
		InsufficientProposersBalance,
	}

	#[pallet::event]
	#[pallet::generate_deposit(pub(super) fn deposit_event)]
	pub enum Event<T: Config<I>, I: 'static = ()> {
		/// doc comment put in metadata
		Proposed(<T as frame_system::Config>::AccountId),
		/// doc
		Spending(BalanceOf<T, I>),
		Something(u32),
	}

	#[pallet::type_value]
	pub(super) fn MyDefault<T: Config<I>, I: 'static>() -> T::Balance { 3.into() }

	#[pallet::storage]
	pub(super) type MyStorageValue<T: Config<I>, I: 'static = ()> =
		StorageValue<Value = T::Balance, QueryKind = ValueQuery, OnEmpty = MyDefault<T, I>>;

	#[pallet::storage]
	#[pallet::getter(fn my_storage)]
	#[pallet::storage_prefix = "SomeOtherName"]
	pub(super) type MyStorage<T, I = ()> =
		StorageMap<Hasher = Blake2_128Concat, Key = u32, Value = u32>;

	#[pallet::genesis_config]
	#[derive(Default)]
	pub struct GenesisConfig {
		_myfield: u32,
	}

	#[pallet::genesis_build]
	impl<T: Config<I>, I: 'static> GenesisBuild<T, I> for GenesisConfig {
		fn build(&self) {}
	}

	#[pallet::origin]
	pub struct Origin<T, I = ()>(PhantomData<(T, I)>);

	#[pallet::validate_unsigned]
	impl<T: Config<I>, I: 'static> ValidateUnsigned for Pallet<T, I> {
		type Call = Call<T, I>;
		fn validate_unsigned(
			source: TransactionSource,
			call: &Self::Call
		) -> TransactionValidity {
			Err(TransactionValidityError::Invalid(InvalidTransaction::Call))
		}
	}

	#[pallet::inherent]
	impl<T: Config<I>, I: 'static> ProvideInherent for Pallet<T, I> {
		type Call = Call<T, I>;
		type Error = InherentError;

		const INHERENT_IDENTIFIER: InherentIdentifier = INHERENT_IDENTIFIER;

		fn create_inherent(_data: &InherentData) -> Option<Self::Call> {
			unimplemented!();
		}

		fn is_inherent(_call: &Self::Call) -> bool {
			unimplemented!();
		}
	}

	// Regular rust code needed for implementing ProvideInherent trait

	#[derive(codec::Encode, sp_runtime::RuntimeDebug)]
	#[cfg_attr(feature = "std", derive(codec::Decode))]
	pub enum InherentError {
	}

	impl sp_inherents::IsFatalError for InherentError {
		fn is_fatal_error(&self) -> bool {
			unimplemented!();
		}
	}

	pub const INHERENT_IDENTIFIER: sp_inherents::InherentIdentifier = *b"testpall";
}

Upgrade guidelines

  1. Export the metadata of the pallet for later checks

    • run your node with the pallet active
    • query the metadata using the state_getMetadata RPC and curl, or use subsee -p <PALLET_NAME> > meta.json
  2. Generate the template upgrade for the pallet provided by decl_storage with the environment variable PRINT_PALLET_UPGRADE: PRINT_PALLET_UPGRADE=1 cargo check -p my_pallet. This template can be used as it contains all information for storages, genesis config and genesis build.

  3. Reorganize the pallet to have the trait Config, decl_* macros, ValidateUnsigned, ProvideInherent, and Origin` all together in one file. Suggested order:

    • Config,
    • decl_module,
    • decl_event,
    • decl_error,
    • decl_storage,
    • origin,
    • validate_unsigned,
    • provide_inherent, so far it should compile and all be correct.
  4. start writing the new pallet module

    pub use pallet::*;
    
    #[frame_support::pallet]
    pub mod pallet {
    	use frame_support::pallet_prelude::*;
    	use frame_system::pallet_prelude::*;
    	use super::*;
    
    	#[pallet::pallet]
    	#[pallet::generate_store($visibility_of_trait_store trait Store)]
    	// NOTE: if the visibility of trait store is private but you want to make it available
    	// in super, then use `pub(super)` or `pub(crate)` to make it available in crate.
    	pub struct Pallet<T>(_);
    	// pub struct Pallet<T, I = ()>(PhantomData<T>); // for instantiable pallet
    }
  5. migrate Config: move trait into the module with

    • all const in decl_module to #[pallet::constant]
    • add the bound IsType<<Self as frame_system::Config>::RuntimeEvent> to type RuntimeEvent
  6. migrate decl_module: write:

    #[pallet::hooks]
    impl<T: Config> Hooks for Pallet<T> {
    }

    and write inside on_initialize, on_finalize, on_runtime_upgrade, offchain_worker, and integrity_test.

    then write:

    #[pallet::call]
    impl<T: Config> Pallet<T> {
    }

    and write inside all the calls in decl_module with a few changes in the signature:

    • origin must now be written completely, e.g. origin: OriginFor<T>
    • result type must be DispatchResultWithPostInfo, you need to write it and also you might need to put Ok(().into()) at the end or the function.
    • #[compact] must now be written #[pallet::compact]
    • #[weight = ..] must now be written #[pallet::weight(..)]
  7. migrate event: rewrite as a simple enum with the attribute #[pallet::event], use #[pallet::generate_deposit($vis fn deposit_event)] to generate deposit_event,

  8. migrate error: rewrite it with attribute #[pallet::error].

  9. migrate storage: decl_storage provide an upgrade template (see 3.). All storages, genesis config, genesis build and default implementation of genesis config can be taken from it directly.

    Otherwise here is the manual process:

    first migrate the genesis logic. write:

    #[pallet::genesis_config]
    struct GenesisConfig {
    	// fields of add_extra_genesis
    }
    impl Default for GenesisConfig {
    	// type default or default provided for fields
    }
    #[pallet::genesis_build]
    impl<T: Config> GenesisBuild<T> for GenesisConfig {
    // for instantiable pallet:
    // `impl<T: Config, I: 'static> GenesisBuild<T, I> for GenesisConfig {
    	fn build() {
    		// The add_extra_genesis build logic
    	}
    }

    for each storage, if it contains config(..) then add fields, and make it default to the value in = ..; or the type default if none, if it contains no build then also add the logic to build the value. for each storage if it contains build(..) then add the logic to genesis_build.

    NOTE: within decl_storage: the individual config is executed first, followed by the build and finally the add_extra_genesis build.

    Once this is done you can migrate storages individually, a few notes:

    • for private storage use pub(crate) type or pub(super) type or nothing,

    • for storages with get(fn ..) use #[pallet::getter(fn ...)]

    • for storages with value being Option<$something> make generic Value being $something and generic QueryKind being OptionQuery (note: this is default). Otherwise make Value the complete value type and QueryKind being ValueQuery.

    • for storages with default value: = $expr; provide some specific OnEmpty generic. To do so use of #[pallet::type_value] to generate the wanted struct to put. example: MyStorage: u32 = 3u32 would be written:

      #[pallet::type_value] fn MyStorageOnEmpty() -> u32 { 3u32 }
      #[pallet::storage]
      pub(super) type MyStorage<T> = StorageValue<_, u32, ValueQuery, MyStorageOnEmpty>;

      NOTE: decl_storage also generates the functions assimilate_storage and build_storage directly on GenesisConfig, and these are sometimes used in tests. In order not to break they can be implemented manually, one can implement those functions by calling the GenesisBuild implementation.

  10. migrate origin: move the origin to the pallet module to be under a #[pallet::origin] attribute

  11. migrate validate_unsigned: move the ValidateUnsigned implementation to the pallet module under a #[pallet::validate_unsigned] attribute

  12. migrate provide_inherent: move the ProvideInherent implementation to the pallet module under a #[pallet::inherent] attribute

  13. rename the usage of Module to Pallet inside the crate.

  14. migration is done, now double check the migration with the checking migration guidelines shown below.

Checking upgrade guidelines:

  • compare metadata. Use subsee to fetch the metadata and do a diff of the resulting json before and after migration. This checks for: * call, names, signature, docs
    • event names, docs
    • error names, docs
    • storage names, hasher, prefixes, default value
    • error, error, constant
  • manually check that:
    • Origin was moved inside the macro under #[pallet::origin] if it exists
    • ValidateUnsigned was moved inside the macro under #[pallet::validate_unsigned)] if it exists
    • ProvideInherent was moved inside the macro under #[pallet::inherent)] if it exists
    • on_initialize / on_finalize / on_runtime_upgrade / offchain_worker were moved to the Hooks implementation
    • storages with config(..) were converted to GenesisConfig field, and their default is = $expr; if the storage has a default value
    • storages with build($expr) or config(..) were built in GenesisBuild::build
    • add_extra_genesis fields were converted to GenesisConfig field with their correct default if specified
    • add_extra_genesis build was written into GenesisBuild::build
  • storage items defined with pallet use the name of the pallet provided by traits::PalletInfo::name as pallet_prefix (in decl_storage, storage items used the pallet_prefix given as input of decl_storage with the syntax as Example). Thus a runtime using the pallet must be careful with this change. To handle this change:
    • either ensure that the name of the pallet given to construct_runtime! is the same as the name the pallet was giving to decl_storage,
    • or do a storage migration from the old prefix used to the new prefix used.

NOTE: The prefixes used by storage items are in metadata. Thus, ensuring the metadata hasn’t changed ensures that the pallet_prefixs used by the storage items haven’t changed.

Notes when macro fails to show proper error message spans:

Rustc loses span for some macro input. Some tips to fix it:

  • do not use inner attribute:
    #[pallet]
    pub mod pallet {
    	//! This inner attribute will make span fail
    	..
    }
  • use the newest nightly possible. The pallet struct placeholder #[pallet::pallet] is mandatory and allows you to specify pallet information.

The struct must be defined as follows:

#[pallet::pallet]
pub struct Pallet<T>(_);

I.e. a regular struct definition named Pallet, with generic T and no where clause.

Macro expansion:

The macro adds this attribute to the struct definition:

#[derive(
	frame_support::CloneNoBound,
	frame_support::EqNoBound,
	frame_support::PartialEqNoBound,
	frame_support::RuntimeDebugNoBound,
)]

and replaces the type _ with PhantomData<T>. It also implements on the pallet:

  • GetStorageVersion
  • OnGenesis: contains some logic to write the pallet version into storage.
  • PalletErrorTypeInfo: provides the type information for the pallet error, if defined.

It declares type Module type alias for Pallet, used by construct_runtime.

It implements PalletInfoAccess on Pallet to ease access to pallet information given by frame_support::traits::PalletInfo. (The implementation uses the associated type frame_system::Config::PalletInfo).

It implements StorageInfoTrait on Pallet which give information about all storages.

If the attribute generate_store is set then the macro creates the trait Store and implements it on Pallet.

If the attribute set_storage_max_encoded_len is set then the macro calls StorageInfoTrait for each storage in the implementation of StorageInfoTrait for the pallet. Otherwise it implements StorageInfoTrait for the pallet using the PartialStorageInfoTrait implementation of storages.

Dev Mode (#[pallet(dev_mode)])

Specifying the argument dev_mode will allow you to enable dev mode for a pallet. The aim of dev mode is to loosen some of the restrictions and requirements placed on production pallets for easy tinkering and development. Dev mode pallets should not be used in production. Enabling dev mode has the following effects:

  • Weights no longer need to be specified on every #[pallet::call] declaration. By default, dev mode pallets will assume a weight of zero (0) if a weight is not specified. This is equivalent to specifying #[weight(0)] on all calls that do not specify a weight.
  • All storages are marked as unbounded, meaning you do not need to implement MaxEncodedLen on storage types. This is equivalent to specifying #[pallet::unbounded] on all storage type definitions.

Note that the dev_mode argument can only be supplied to the #[pallet] or #[frame_support::pallet] attribute macro that encloses your pallet module. This argument cannot be specified anywhere else, including but not limited to the #[pallet::pallet] attribute macro.

WARNING:
You should not deploy or use dev mode pallets in production. Doing so can break your chain
and therefore should never be done. Once you are done tinkering, you should remove the
'dev_mode' argument from your #[pallet] declaration and fix any compile errors before
attempting to use your pallet in a production scenario.

See frame_support::pallet docs for more info.