Capabilities/1.16

From Forge Community Wiki
Revision as of 04:13, 27 July 2021 by ShrimpBot (talk | contribs) (Copy Capabilities to MC1.16 archive)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

Capabilities are a Forge system that allows cross-mod interactions by allowing capability providers to dynamically respect contracts and provide specialized behavior without requiring the implementation of many interfaces or hard dependencies on mods.

History

In an ideal world, all that would be needed for a mod to provide the equivalent of a capability would be implementing an interface. This is in fact how cross-mod interaction used to work prior to the introduction of capabilities.

The real world, though, is often much more complicated: users wanted to be free to combine mods the way they wanted and saw fit, and developers wanted to be able to declare soft dependencies on other mods, thus reducing the need of having a huge mod pack just for testing.

The first approach used by Forge was conditional stripping of interfaces and methods, but this proved to be problematic. While the idea works well in theory, in practice the ASM editing of classes relied on complex mechanics and could lead to hard to spot bugs.

For this reason, the entire system was redesigned and the concept of capabilities was born.

The Concept

A capability allows any capability provider to conditionally expose a certain ability to do something, e.g. accepting power or handling items. A capability provider, moreover, can decide to expose a capability only on certain sides, allowing for easy interactions with hoppers, cables, etc.

Capabilities may also be added and removed dynamically both from the "owner" of the capability provider and other mods, allowing even easier cross-mod interaction. For example, a mod that isn't compatible with Forge Energy could be converted into one by dynamically attaching the Forge Energy capability and handling the conversion to a third-party energy system without having to alter the original mod.

Terminology

The high flexibility of the system comes with a cost, though, which is terminology. The following section wants to be a dictionary of sorts, defining all the terms that you may come across when dealing with capabilities.

In the rest of this article, we will refer to these terms frequently, so make sure you are familiar with them.

Capability
the ability to perform something. In-code this is represented by the Capability class.
Capability Provider
something that is able to support capabilities and provides a mean of accessing them. In-code they are represented by implementations of ICapabilityProvider. There are multiple kinds of capability providers:
Volatile Provider
a provider that doesn't persist data to disk; once the provider ceases to exist for any number of reasons, all capability data gets deleted.
Persistent Provider
a provider that requires all capabilities to serialize data to disk, in order to persist data even across game restarts. They implement the INBTSerializable interface.
Agnostic Provider
a provider that isn't neither volatile nor persistent, rather delegates the decision either to the capability directly or to sub-implementations. They also implement the INBTSerializable interface.
Capability Interface
the interface that defines the contract of the capability, so what operations the capability exposes.
Capability Implementation
one of the possibly many implementations of the capability interface, that actually carries out the work; one of the various implementations may also be considered the default capability implementation.
Capability Storage
the manager that handles loading and storing persistent capabilities data from and to disk, guaranteeing preservation of information; in-code this is represented by an implementation of the Capability.IStorage interface.

The wary reader may note that both persistent and agnostic providers are represented the same way in code. In fact, the only difference between them comes from pure semantics in how their serialization methods are designed. This will be further discussed in their respective sections.

Moreover, it is also common to refer to the capability interface as simply the capability. While not strictly correct, due to common usage we will also use this convention. So, to refer to the capability interface MyCapability, we will usually talk about the "MyCapability capability".

Forge-provided Capabilities and Providers

In order to ensure mods can work together seamlessly, Forge provides a set of default capabilities and capability providers.

The default capability providers in a Forge environment are: TileEntity, Entity, ItemStack, World, and Chunk. These are all agnostic providers, since they don't mandate any sort of capability persistency requirements. Rather, it is the job of whoever subclasses these providers to deal with either volatile or non-volatile capabilities.

The default capabilities that forge provides are represented by the interfaces IItemHandler, IFluidHandler, IFluidHandlerItem, IEnergyStorage, and IAnimationStateMachine. Each one of these capabilities will be discussed in the corresponding section.

IItemHandler

The IItemHandler capability refers to the ability for any capability provider to have some sort of internal inventory with a certain number of slots, from which items can be inserted and extracted. It is also possible, though, to expose this capability even if no such inventory is present as long as the capability provider can emulate its presence (e.g. tools that allow accessing remote inventories).

This effectively replaces the vanilla interfaces IInventory and ISidedInventory. These interfaces are in fact retained only to allow vanilla code to compile and should not be used in mod code. This extends to anything that implements those vanilla interfaces, such as LockableLootTileEntity.

A default reference implementation for this capability interface is provided in ItemStackHandler.

IFluidHandler

The IFluidHandler capability refers to the ability for any capability provider to handle and store fluids in one or multiple fluid tanks. It is effectively the equivalent in terms of fluids of the IItemHandler capability.

A default reference implementation for this capability interface is provided in TileFluidHandler.

IFluidHandlerItem

The IFluidHandlerItem capability referes to the ability for an ItemStack capability provider to handle and store fluids in one or multiple fluid tanks. It is basically a specialized version of the IFluidHandler capability that allows ItemStacks to define a custom container.

IEnergyStorage

The IEnergyStorage capability refers to the ability for any capability provider to store, consume, and produce energy. This capability is the base capability for what's commonly known in the modded world as Forge Energy (or FE), i.e. the energy system most mods use. Its internal design is heavily based on the (now defunct) Redstone Flux Energy API, supporting both a push and pull system.

A default reference implementation for this capability interface is provided in EnergyStorage.

IAnimationStateMachine

The IAnimationStateMachine capability refers to the ability for any capability provider to leverage the Forge Animation State Machine API for animations.

Working with Capabilities

Both capability providers and users need to be able to provide and access capabilities through a common framework, otherwise the ideal of dynamic and mod-agnostic would not really exist. For this reason, both capability providers and capability accessors (which we define as everything that wants to access a capability), also known as clients or users, need to work together and with Forge to ensure that the common interface is used sensibly and correctly by all parties.

Obtaining a Capability

Before being able to work with a capability, it is necessary to obtain an instance of the Capability object itself. Since these objects are created by Forge and there is only one unique instance for each capability that may exist, this instance cannot be obtained by "common" means. Forge provides two different methods of obtaining such instances: injecting into a field, or a callback method.

Injecting into a Field

A Capability can be injected automatically into a field as soon as it gets created by Forge, following the principle commonly known as dependency injection. This provides less flexibility, since it doesn't notify the user that the capability has been injected nor runs arbitrary code. Nevertheless, it is suggested to use this method instead of the callback approach.

To inject the Capability into a field, all that's needed is to declare a static field of type Capability<T>, where T represents the capability interface, and annotate it with @CapabilityInject(T.class).

For a more practical example, consider the following snippet:

@CapabilityInject(IItemHandler.class)
public static Capability<IItemHandler> ITEM_HANDLER_CAPABILITY = null;

The above code will let Forge know that the field ITEM_HANDLER_CAPABILITY should be injected with the unique instance of the IItemHandler capability. Assigning the field to null allows us to provide a reasonable fallback in case the capability we want hasn't been registered yet.

This injection is, for obvious reasons, redundant, since that capability is also available through CapabilityItemHandler.

Declaring a Callback

Another option is to declare a callback method, meaning a method that will be called with the value of the desired Capability once the instance is available. This gives more flexibility since the method may perform a number of arbitrary actions with the received instance prior to storing it in a field, or may even discard the capability entirely if wanted. Nevertheless, the usage of a field instead of a method is encouraged as a matter of style.

To use a method as a callback, the method must be declared as static and accepting a single parameter of type Capability<T>, where T represents the capability interface. The method should also be annotated with @CapabilityInject(T.class).

For a more practical example, consider the following snippet:

public static Capability<IEnergyStorage> ENERGY = null;

@CapabilityInject(IEnergyStorage.class)
private static void onEnergyStorageInit(Capability<IEnergyStorage> capability) {
    LOGGER.info("Received IEnergyStorage capability '{}': enabling Forge Energy support", capability);
    ENERGY = capability;
}

The above code declares a callback method that will be invoked when a Capability instance for IEnergyStorage is available. The callback then prints a log message and stores the capability into a public field for accessibility. The field is initialized to null to provide a reasonable fallback in case the capability does not exist.

This callback is, for obvious reasons, redundant, since that capability is also available through CapabilityEnergy.

Exposing a Capability

Exposing a capability is a voluntary act by a capability provider that allows the capability to be discovered and accessed by users.

To do so, a capability provider needs to juggle a couple more moving pieces to ensure that the capability state remains consistent and that the lookup remains fast. It is in fact possible for a capability provider to be asked to provide many capabilities many times in the same tick. For this reason, a provider is asked to do the following:

  • the LazyOptionals that get returned must be cached;
  • if a capability changes exposure state (more on this later), all listeners must be notified;
  • if a capability gets invalidated (more on this later), all listeners must be notified
  • the lookup inside getCapability must be performed with an if-else chain;
  • all unexposed but still present capabilities should be available if the provider is queried with a null direction (see Accessing a Capability for more information);
  • if no capability of a given type is available or accessible, the provider must call super as long as it is possible to do so.

Capability providers must also reflect changes in the exposure state of a capability, meaning that if the accessibility of a capability from a certain Direction changes (refer to Accessing a Capability for more information), it is the provider's responsibility to trigger a state response by invalidating the returned LazyOptional and caching a new one. This should also be performed when a capability gets invalidated, such as when a capability provider gets removed.

With all of the above in mind, part of a capability provider implementation may be similar to the following snippet:

// suppose the presence of a field 'inventory' of type 'IItemHandler'

private final LazyOptional<IItemhandler> inventoryOptional = LazyOptional.of(() -> this.inventory);

@Override
public <T> LazyOptional<T> getCapability(Capability<T> capability, @Nullable Direction direction) {
    if (capability == CapabilityItemHandler.ITEM_HANDLER_CAPABILITY
            && (direction == null || direction == Direction.UP || direction == Direction.DOWN)) {
        return this.inventoryOptional.cast();
    }
    return super.getCapability(capability, direction); // See note after snippet
}

@Override
protected void invalidateCaps() {
    super.invalidateCaps();
    this.inventoryOptional.invalidate();
}

This possible implementation of a capability provider exposes an IItemHandler capability and restricts access only to the UP and DOWN directions. If we assume this capability provider is a TileEntity, then we may also say that the inventory is only accessible from the top and the bottom of the block.

Moreover, the capability gets automatically invalidated when the provider gets invalidated. Assuming this is a TileEntity, this usually happens when the block gets removed from the world or unloaded due to distance.

The super call at the end of the getCapability method is extremely important, since it's what allows Attaching external Capabilities to capability providers. Nevertheless, it is not always possible to invoke super: in those cases, an empty LazyOptional should be returned.

Attaching a Capability

Attaching a Capability is a process by which external agents "modify" a Capability Provider, making it expose additional capabilities other than the already available ones.

To do so, the attaching agent (which means the thing that wants to attach a capability to another provider) must listen to the AttachCapabilitiesEvent<T>. The T in this case represents the capability provider you want to attach the capability to. Note that the type of T must be the base type of the capability provider, not a subclass. As an example, if you want to attach a capability to a MyTileEntity, which extends TileEntity, you'll have to listen to AttachCapabilitiesEvent<TileEntity>, NOT to AttachCapabilitiesEvent<MyTileEntity>, since the latter will never fire.

The attaching agent can use the provided methods getObject, addCapability, and addListener to check whether the capability should be attached to the current object and perform the desired action.

When attaching a capability, the attaching agent should also provide a name in the form of a ResourceLocation. The name must be under the attaching agent's namespace, but no restrictions are placed on the actual name, as long as it is unique inside the given namespace.

Maybe a little counter-intuitively, the process of attaching does not attach a capability nor a capability interface directly. Rather, the attaching agent should create its own implementation of a Capability Provider and attach it via the event. This is done so that the attaching agent can have control over when, how, and where its capabilities are exposed, instead of relying on the game itself deciding these parameters. For this reason, all considerations given in the Exposing a Capability section on how to correctly create a Capability Provider.

With the above in mind, part of an attaching agent may be similar to the following snippet of code:

@SubscribeEvent
public void onAttachingCapabilities(final AttachCapabilitiesEvent<TileEntity> event) {
    if (!(event.getObject() instanceof EnergyBasedTileEntity)) return;

    EnergyStorage backend = new EnergyStorage(((EnergyBasedTileEntity) event.getObject()).capacity);
    LazyOptional<IEnergyStorage> optionalStorage = LazyOptional.of(() -> backend);

    ICapabilityProvider provider = new ICapabilityProvider() {
        @Override
        public <T> LazyOptional<T> getCapability(Capability<T> cap, @Nullable Direction direction) {
            if (cap == CapabilityEnergy.ENERGY) {
                return optionalStorage.cast();
            }
            return LazyOptional.empty();
        }
    };

    event.addCapability(new ResourceLocation("examplemod", "fe_compatibility"), provider);
    event.addListener(optionalStorage::invalidate);
}

This example implementation of an attaching agent attaches a IEnergyStorage capability to all TileEntity instance that are a subclass of EnergyBasedTileEntity. It also sets up the LazyOptional for invalidation if the parent capability provider gets invalidated.

Note also the call of LazyOptional.empty() rather than a super. This is needed because when attaching a capability, the parent capability provider isn't known. For this reason, it is necessary to return an empty LazyOptional. The game will then handle automatic merging of the various different providers into a single one.

The above example is one of a Volatile Capability Provider. On the other hand, mods may want to persist their data across sessions. In this case, they should attach a Persistent Capability Provider: this can be done either by implementing the INBTSerializable interface along with ICapabilityProvider or by implementing the ICapabilitySerializable interface.

The previous example reworked to use a Persistent Capability Provider may be similar to the following snippet:

@SubscribeEvent
public void onAttachingCapabilities(final AttachCapabilitiesEvent<TileEntity> event) {
    if (!(event.getObject() instanceof EnergyBasedTileEntity)) return;

    EnergyStorage backend = new EnergyStorage(((EnergyBasedTileEntity) event.getObject()).capacity);
    LazyOptional<IEnergyStorage> optionalStorage = LazyOptional.of(() -> backend);
    Capability<IEnergyStorage> capability = CapabilityEnergy.ENERGY;

    ICapabilityProvider provider = new ICapabilitySerializable<IntNBT>() {
        @Override
        public <T> LazyOptional<T> getCapability(Capability<T> cap, @Nullable Direction direction) {
            if (cap == capability) {
                return optionalStorage.cast();
            }
            return LazyOptional.empty();
        }

        @Override
        public IntNBT serializeNBT() {
            return capability.getStorage().writeNbt(capability, backend, null);
        }

        @Override
        public void deserializeNBT(IntNBT nbt) {
            capability.getStorage().readNBT(capability, backend, null, nbt);
        }
    };

    event.addCapabilities(new ResourceLocation("examplemod", "fe_compatibility"), provider);
    event.addListener(optionalStorage::invalidate);
}


Accessing a Capability

Accessing a Capability is the process by which a user is able to query a Capability Provider for a specific instance of a capability.

This is perhaps the second most important part of the entire capability system, since it is what allows cross-mod interaction. To obtain an instance of a Capability, the user must first get a hold of the Capability Provider that should be queried. This can be done in a variety of ways and is outside the scope of this article. The user should then invoke the getCapability method passing the unique instance of the capability that should be queried (see Obtaining a Capability for more information) and the querying Direction, if applicable.

The returned object is a LazyOptional wrapping the queried Capability, if the capability provider exposes it, otherwise it will be empty. The LazyOptional can be either unwrapped via an orElse or used directly via ifPresent.

It is highly suggested to cache the returned LazyOptional to avoid querying the same provider every time, in order to improve performance. The user should thus register itself to the invalidation listener via the addListener method. This ensures that the user will be able to react to the invalidation of the LazyOptional and remove it from the cache.

With the above in mind, part of an user may be similar to the following snippet of code:

private final Map<Direction, LazyOptional<IEnergyStorage>> cache = new HashMap<>();

private void sendPowerTo(int power, Direction direction) {
    LazyOptional<IEnergyStorage> targetCapability = cache.get(direction);

    if (targetCapability == null) {
        ICapabilityProvider provider = world.getTileEntity(pos.offset(direction));
        targetCapability = provider.getCapability(CapabilityEnergy.ENERGY, direction.getOpposite());
        cache.put(direction, targetCapability);
        targetCapability.addListener(self -> cache.put(direction, null));
    }

    targetCapability.ifPresent(storage -> storage.receiveEnergy(power, false));
}

This example implementation of an user is querying via a TileEntity the neighboring capability provider for an IEnergyStorage capability. Before obtaining the provider, the code performs a cache lookup for the targeted capability. If the check succeeds, then no lookup is performed; if the check fails, the targeted Capability Provider is obtained and queried for the Capability. The obtained LazyOptional is then cached and a listener is attached to it so that the cache would be emptied on invalidation. The code then continues with the interaction with the capability, which is outside the scope of this article.

Creating Custom Capabilities

While the various capabilities provided by Forge may satisfy the most common use cases, there is always the chance that a mod may require a custom solution. For this reason, Forge provides a way to define a custom Capability.

Defining a custom Capability requires the user to provide three main components: the Capability Interface, at least one Capability Implementation, and the Capability Storage. Optionally, a Capability Provider can also be created. In this case, the provider will be used as described in Attaching a Capability. The various details for all these components are described in the respective sections of this article.

In this section, we will refer to the implementation of a MyCapability capability, that can be used to store a single mutable String.

Refer also to Code Examples for an example on how the various components may be implemented in a real-world scenario.

The Capability Interface and the Capability Implementation

The Capability Interface is one of the most important parts of a Capability: without it, the Capability effectively does not exist. Designing a Capability Interface is exactly like designing any Java interface, so the particular details will be glossed over in this section.

The Capability Implementation, on the other hand, is the implementation of the previously defined Capability Interface. Usual rules for interface implementations follow. There can be more than one Capability Implementation for each capability, but no less than one.

Note that a well-formed capability implementation should not store the Capability Provider inside of it: we call the capability implementation provider-agnostic. This is not a hard-requirement, though, rather it should act more as a guideline. There are in fact certain situations where this cannot be avoided (e.g. attaching a client-synced capability to an ItemStack).

One of the various Capability Implementation should also act as the default implementation. Other mods can ask the capability to create an instance of the default implementation without ever referring to such an implementation themselves. This guarantees separation of API code from implementation code, which is also one of the goals of the capability system.

Given all of the above information, this may be an example implementation of both a Capability Interface and a Capability Implementation:

public interface MyCapability {
    String getValue();
    void setValue(String value);
}

public class MyCapabilityImplementation implements MyCapability {
    private String value;

    @Override
    public String getValue() {
        return this.value;
    }

    @Override
    public void setValue(String value) {
        this.value = value;
    }
}

Note that in this case, only a single implementation is provided, which will also act as the default implementation for the MyCapability capability.

The Capability Storage

The Capability Storage is that component of the Capability System that is responsible for serializing and deserializing a capability. All capabilities must provide one, since certain providers may or may not require that their capabilities are serializable.

The Capability Storage implements the Capability.IStorage<T> interface, where T is the generic type of the Capability Interface the storage refers to. Each capability must have one and exactly one Capability Storage.

The Storage is usually called by the Capability Provider when serialization or deserialization needs to happen. The Storage is then responsible of reading the data from the given capability instance and convert that into an NBT-based structure that can be serialized. A Storage may also return null to indicate that no serialization is necessary, although some providers may require an empty tag to be supplied instead. At the same time, the Storage is also responsible for restoring the original state of the capability when deserialization happens. In this case, the given NBT structure is guaranteed not to be null.

In all cases, a Direction is provided for context, if available.

Although discouraged as a matter of code cleanliness, it is legal for a Capability Storage to require a specific Capability Implementation for the serialization and deserialization to be successful. If this is the case, this requirement must be documented, though the code should be refactored wherever possible to remove this requirement.

Given the above information, an example of an implementation of a Capability Storage may be the following:

public class MyCapabilityStorage implements Capability.IStorage<MyCapability> {
    @Override
    @Nullable
    INBT writeNBT(Capability<MyCapability> capability, MyCapability instance, Direction direction) {
        return StringNBT.valueOf(instance.getValue());
    }

    @Override
    public void readNBT(Capability<MyCapability> capability, MyCapability instance, Direction direction, INBT nbtData) {
        if (!(nbtData instanceof StringNBT)) {
            throw new IllegalArgumentException("Unable to deserialize 'MyCapability' from a non-String NBT structure");
        }
        instance.setValue(((StringNBT) nbtData).getString());
    }
}

Note the instanceof check needed to ensure that the given INBT instance is valid for deserialization. A Capability Storage should always perform this check prior to the cast in order to provide a meaningful error message, rather than a cryptic ClassCastException.

The Capability Provider

The Capability Provider is an optional component of the capability that allows the Capability to be attached to a component. The details on how a Capability Provider should behave have already been discussed in the two previous sections Exposing a Capability and Attaching a Capability: refer to those for more information.

Tying it All Together

Once all components of a Capability have been created, they must be registered so that the game is aware of the capability's presence. The registration requires specifying the Capability Interface, the Capability Storage, and a factory for the default capability implementation.

The registration can be performed by calling the register method on the CapabilityManager. This needs to happen when the FMLCommonSetupEvent is fired on the MOD event bus. The registration will also automatically inject the created Capability into all relevant fields and methods: refer to Obtaining a Capability for more information.

An example of registration can be found in the snippet that follows:

public void onCommonSetup(FMLCommonSetupEvent event) {
    CapabilityManager.INSTANCE.register(MyCapability.class, new MyCapabilityStorage(), MyCapabilityImplementation::new);
}

Custom Capability Providers

Much like custom Capabilities, Forge also allows the creation of custom Capability Providers. The main advantage of this is allowing mods to create custom providers for their custom objects, in order to promote not only cross-mod compatibility, but also uniformity in the way users may interact with different mod APIs.

This section will only give the basic outline of what is necessary to implement a custom Capability Provider: for more in-depth explanation, people are referred to the game code.

By definition, a custom Capability Provider is everything that implements the ICapabilityProvider interface. In this section, though, we will only cover people that may want to replicate the functionality of one of the default providers, such as TileEntity or Chunk.

The easiest way of doing this is extending the CapabilityProvider class provided by Forge. This will automatically set up an agnostic Capability Provider. To fully initialize the capability provider, the subclass should then invoke the gatherCapabilities method as the last instruction in its constructor, to ensure that the game is able to recollect and attach all capabilities that other mods may want to attach to the capability provider.

Code Examples