The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/) and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html). Additional documentation and release notes are available at [Multiplayer Documentation](https://docs-multiplayer.unity3d.com). ## [2.0.0-exp.2] - 2024-04-02 ### Added - Added updates to all internal messages to account for a distributed authority network session connection. (#2863) - Added `NetworkRigidbodyBase` that provides users with a more customizable network rigidbody, handles both `Rigidbody` and `Rigidbody2D`, and provides an option to make `NetworkTransform` use the rigid body for motion. (#2863) - For a customized `NetworkRigidbodyBase` class: - `NetworkRigidbodyBase.AutoUpdateKinematicState` provides control on whether the kinematic setting will be automatically set or not when ownership changes. - `NetworkRigidbodyBase.AutoSetKinematicOnDespawn` provides control on whether isKinematic will automatically be set to true when the associated `NetworkObject` is despawned. - `NetworkRigidbodyBase.Initialize` is a protected method that, when invoked, will initialize the instance. This includes options to: - Set whether using a `RigidbodyTypes.Rigidbody` or `RigidbodyTypes.Rigidbody2D`. - Includes additional optional parameters to set the `NetworkTransform`, `Rigidbody`, and `Rigidbody2d` to use. - Provides additional public methods: - `NetworkRigidbodyBase.GetPosition` to return the position of the `Rigidbody` or `Rigidbody2d` (depending upon its initialized setting). - `NetworkRigidbodyBase.GetRotation` to return the rotation of the `Rigidbody` or `Rigidbody2d` (depending upon its initialized setting). - `NetworkRigidbodyBase.MovePosition` to move to the position of the `Rigidbody` or `Rigidbody2d` (depending upon its initialized setting). - `NetworkRigidbodyBase.MoveRotation` to move to the rotation of the `Rigidbody` or `Rigidbody2d` (depending upon its initialized setting). - `NetworkRigidbodyBase.Move` to move to the position and rotation of the `Rigidbody` or `Rigidbody2d` (depending upon its initialized setting). - `NetworkRigidbodyBase.Move` to move to the position and rotation of the `Rigidbody` or `Rigidbody2d` (depending upon its initialized setting). - `NetworkRigidbodyBase.SetPosition` to set the position of the `Rigidbody` or `Rigidbody2d` (depending upon its initialized setting). - `NetworkRigidbodyBase.SetRotation` to set the rotation of the `Rigidbody` or `Rigidbody2d` (depending upon its initialized setting). - `NetworkRigidbodyBase.ApplyCurrentTransform` to set the position and rotation of the `Rigidbody` or `Rigidbody2d` based on the associated `GameObject` transform (depending upon its initialized setting). - `NetworkRigidbodyBase.WakeIfSleeping` to wake up the rigid body if sleeping. - `NetworkRigidbodyBase.SleepRigidbody` to put the rigid body to sleep. - `NetworkRigidbodyBase.IsKinematic` to determine if the `Rigidbody` or `Rigidbody2d` (depending upon its initialized setting) is currently kinematic. - `NetworkRigidbodyBase.SetIsKinematic` to set the `Rigidbody` or `Rigidbody2d` (depending upon its initialized setting) current kinematic state. - `NetworkRigidbodyBase.ResetInterpolation` to reset the `Rigidbody` or `Rigidbody2d` (depending upon its initialized setting) back to its original interpolation value when initialized. - Now includes a `MonoBehaviour.FixedUpdate` implementation that will update the assigned `NetworkTransform` when `NetworkRigidbodyBase.UseRigidBodyForMotion` is true. (#2863) - Added `RigidbodyContactEventManager` that provides a more optimized way to process collision enter and collision stay events as opposed to the `Monobehaviour` approach. (#2863) - Can be used in client-server and distributed authority modes, but is particularly useful in distributed authority. - Added rigid body motion updates to `NetworkTransform` which allows users to set interolation on rigid bodies. (#2863) - Extrapolation is only allowed on authoritative instances, but custom class derived from `NetworkRigidbodyBase` or `NetworkRigidbody` or `NetworkRigidbody2D` automatically switches non-authoritative instances to interpolation if set to extrapolation. - Added distributed authority mode support to `NetworkAnimator`. (#2863) - Added session mode selection to `NetworkManager` inspector view. (#2863) - Added distributed authority permissions feature. (#2863) - Added distributed authority mode specific `NetworkObject` permissions flags (Distributable, Transferable, and RequestRequired). (#2863) - Added distributed authority mode specific `NetworkObject.SetOwnershipStatus` method that applies one or more `NetworkObject` instance's ownership flags. If updated when spawned, the ownership permission changes are synchronized with the other connected clients. (#2863) - Added distributed authority mode specific `NetworkObject.RemoveOwnershipStatus` method that removes one or more `NetworkObject` instance's ownership flags. If updated when spawned, the ownership permission changes are synchronized with the other connected clients. (#2863) - Added distributed authority mode specific `NetworkObject.HasOwnershipStatus` method that will return (true or false) whether one or more ownership flags is set. (#2863) - Added distributed authority mode specific `NetworkObject.SetOwnershipLock` method that locks ownership of a `NetworkObject` to prevent ownership from changing until the current owner releases the lock. (#2863) - Added distributed authority mode specific `NetworkObject.RequestOwnership` method that sends an ownership request to the current owner of a spawned `NetworkObject` instance. (#2863) - Added distributed authority mode specific `NetworkObject.OnOwnershipRequested` callback handler that is invoked on the owner/authoritative side when a non-owner requests ownership. Depending upon the boolean returned value depends upon whether the request is approved or denied. (#2863) - Added distributed authority mode specific `NetworkObject.OnOwnershipRequestResponse` callback handler that is invoked when a non-owner's request has been processed. This callback includes a `NetworkObjet.OwnershipRequestResponseStatus` response parameter that describes whether the request was approved or the reason why it was not approved. (#2863) - Added distributed authority mode specific `NetworkObject.DeferDespawn` method that defers the despawning of `NetworkObject` instances on non-authoritative clients based on the tick offset parameter. (#2863) - Added distributed authority mode specific `NetworkObject.OnDeferredDespawnComplete` callback handler that can be used to further control when deferring the despawning of a `NetworkObject` on non-authoritative instances. (#2863) - Added `NetworkClient.SessionModeType` as one way to determine the current session mode of the network session a client is connected to. (#2863) - Added distributed authority mode specific `NetworkClient.IsSessionOwner` property to determine if the current local client is the current session owner of a distributed authority session. (#2863) - Added distributed authority mode specific client side spawning capabilities. When running in distributed authority mode, clients can instantiate and spawn `NetworkObject` instances (the local client is authomatically the owner of the spawned object). (#2863) - This is useful to better visually synchronize owner authoritative motion models and newly spawned `NetworkObject` instances (i.e. projectiles for example). - Added distributed authority mode specific client side player spawning capabilities. Clients will automatically spawn their associated player object locally. (#2863) - Added distributed authority mode specific `NetworkConfig.AutoSpawnPlayerPrefabClientSide` property (default is true) to provide control over the automatic spawning of player prefabs on the local client side. (#2863) - Added distributed authority mode specific `NetworkManager.OnFetchLocalPlayerPrefabToSpawn` callback that, when assigned, will allow the local client to provide the player prefab to be spawned for the local client. (#2863) - This is only invoked if the `NetworkConfig.AutoSpawnPlayerPrefabClientSide` property is set to true. - Added distributed authority mode specific `NetworkBehaviour.HasAuthority` property that determines if the local client has authority over the associated `NetworkObject` instance (typical use case is within a `NetworkBehaviour` script much like that of `IsServer` or `IsClient`). (#2863) - Added distributed authority mode specific `NetworkBehaviour.IsSessionOwner` property that determines if the local client is the session owner (typical use case would be to determine if the local client can has scene management authority within a `NetworkBehaviour` script). (#2863) - Added support for distributed authority mode scene management where the currently assigned session owner can start scene events (i.e. scene loading and scene unloading). (#2863) ### Fixed - Fixed issue where the host was not invoking `OnClientDisconnectCallback` for its own local client when internally shutting down. (#2822) - Fixed issue where NetworkTransform could potentially attempt to "unregister" a named message prior to it being registered. (#2807) - Fixed issue where in-scene placed `NetworkObject`s with complex nested children `NetworkObject`s (more than one child in depth) would not synchronize properly if WorldPositionStays was set to true. (#2796) ### Changed - Changed client side awareness of other clients is now the same as a server or host. (#2863) - Changed `NetworkManager.ConnectedClients` can now be accessed by both server and clients. (#2863) - Changed `NetworkManager.ConnectedClientsList` can now be accessed by both server and clients. (#2863) - Changed `NetworkTransform` defaults to owner authoritative when connected to a distributed authority session. (#2863) - Changed `NetworkVariable` defaults to owner write and everyone read permissions when connected to a distributed authority session (even if declared with server read or write permissions). (#2863) - Changed `NetworkObject` no longer implements the `MonoBehaviour.Update` method in order to determine whether a `NetworkObject` instance has been migrated to a different scene. Instead, only `NetworkObjects` with the `SceneMigrationSynchronization` property set will be updated internally during the `NetworkUpdateStage.PostLateUpdate` by `NetworkManager`. (#2863) - Changed `NetworkManager` inspector view layout where properties are now organized by category. (#2863) - Changed `NetworkTransform` to now use `NetworkTransformMessage` as opposed to named messages for NetworkTransformState updates. (#2810) - Changed `CustomMessageManager` so it no longer attempts to register or "unregister" a null or empty string and will log an error if this condition occurs. (#2807)
345 lines
14 KiB
C#
345 lines
14 KiB
C#
using System;
|
|
using Unity.Collections;
|
|
using Unity.Collections.LowLevel.Unsafe;
|
|
using Unity.Networking.Transport;
|
|
|
|
namespace Unity.Netcode.Transports.UTP
|
|
{
|
|
/// <summary>Queue for batched messages meant to be sent through UTP.</summary>
|
|
/// <remarks>
|
|
/// Messages should be pushed on the queue with <see cref="PushMessage"/>. To send batched
|
|
/// messages, call <see cref="FillWriterWithMessages"/> or <see cref="FillWriterWithBytes"/>
|
|
/// with the <see cref="DataStreamWriter"/> obtained from <see cref="NetworkDriver.BeginSend"/>.
|
|
/// This will fill the writer with as many messages/bytes as possible. If the send is
|
|
/// successful, call <see cref="Consume"/> to remove the data from the queue.
|
|
///
|
|
/// This is meant as a companion to <see cref="BatchedReceiveQueue"/>, which should be used to
|
|
/// read messages sent with this queue.
|
|
/// </remarks>
|
|
internal struct BatchedSendQueue : IDisposable
|
|
{
|
|
// Note that we're using NativeList basically like a growable NativeArray, where the length
|
|
// of the list is the capacity of our array. (We can't use the capacity of the list as our
|
|
// queue capacity because NativeList may elect to set it higher than what we'd set it to
|
|
// with SetCapacity, which breaks the logic of our code.)
|
|
private NativeList<byte> m_Data;
|
|
private NativeArray<int> m_HeadTailIndices;
|
|
private int m_MaximumCapacity;
|
|
private int m_MinimumCapacity;
|
|
|
|
/// <summary>Overhead that is added to each message in the queue.</summary>
|
|
public const int PerMessageOverhead = sizeof(int);
|
|
|
|
internal const int MinimumMinimumCapacity = 4096;
|
|
|
|
// Indices into m_HeadTailIndicies.
|
|
private const int k_HeadInternalIndex = 0;
|
|
private const int k_TailInternalIndex = 1;
|
|
|
|
/// <summary>Index of the first byte of the oldest data in the queue.</summary>
|
|
private int HeadIndex
|
|
{
|
|
get { return m_HeadTailIndices[k_HeadInternalIndex]; }
|
|
set { m_HeadTailIndices[k_HeadInternalIndex] = value; }
|
|
}
|
|
|
|
/// <summary>Index one past the last byte of the most recent data in the queue.</summary>
|
|
private int TailIndex
|
|
{
|
|
get { return m_HeadTailIndices[k_TailInternalIndex]; }
|
|
set { m_HeadTailIndices[k_TailInternalIndex] = value; }
|
|
}
|
|
|
|
public int Length => TailIndex - HeadIndex;
|
|
public int Capacity => m_Data.Length;
|
|
public bool IsEmpty => HeadIndex == TailIndex;
|
|
public bool IsCreated => m_Data.IsCreated;
|
|
|
|
/// <summary>Construct a new empty send queue.</summary>
|
|
/// <param name="capacity">Maximum capacity of the send queue.</param>
|
|
public BatchedSendQueue(int capacity)
|
|
{
|
|
// Make sure the maximum capacity will be even.
|
|
m_MaximumCapacity = capacity + (capacity & 1);
|
|
|
|
// We pick the minimum capacity such that if we keep doubling it, we'll eventually hit
|
|
// the maximum capacity exactly. The alternative would be to use capacities that are
|
|
// powers of 2, but this can lead to over-allocating quite a bit of memory (especially
|
|
// since we expect maximum capacities to be in the megabytes range). The approach taken
|
|
// here avoids this issue, at the cost of not having allocations of nice round sizes.
|
|
m_MinimumCapacity = m_MaximumCapacity;
|
|
while (m_MinimumCapacity / 2 >= MinimumMinimumCapacity)
|
|
{
|
|
m_MinimumCapacity /= 2;
|
|
}
|
|
|
|
m_Data = new NativeList<byte>(m_MinimumCapacity, Allocator.Persistent);
|
|
m_HeadTailIndices = new NativeArray<int>(2, Allocator.Persistent);
|
|
|
|
m_Data.ResizeUninitialized(m_MinimumCapacity);
|
|
|
|
HeadIndex = 0;
|
|
TailIndex = 0;
|
|
}
|
|
|
|
public void Dispose()
|
|
{
|
|
if (IsCreated)
|
|
{
|
|
m_Data.Dispose();
|
|
m_HeadTailIndices.Dispose();
|
|
}
|
|
}
|
|
|
|
/// <summary>Write a raw buffer to a DataStreamWriter.</summary>
|
|
private unsafe void WriteBytes(ref DataStreamWriter writer, byte* data, int length)
|
|
{
|
|
#if UTP_TRANSPORT_2_0_ABOVE
|
|
writer.WriteBytesUnsafe(data, length);
|
|
#else
|
|
writer.WriteBytes(data, length);
|
|
#endif
|
|
}
|
|
|
|
/// <summary>Append data at the tail of the queue. No safety checks.</summary>
|
|
private void AppendDataAtTail(ArraySegment<byte> data)
|
|
{
|
|
unsafe
|
|
{
|
|
#if UTP_TRANSPORT_2_0_ABOVE
|
|
var writer = new DataStreamWriter(m_Data.GetUnsafePtr() + TailIndex, Capacity - TailIndex);
|
|
#else
|
|
var writer = new DataStreamWriter((byte*)m_Data.GetUnsafePtr() + TailIndex, Capacity - TailIndex);
|
|
#endif
|
|
|
|
writer.WriteInt(data.Count);
|
|
|
|
fixed (byte* dataPtr = data.Array)
|
|
{
|
|
WriteBytes(ref writer, dataPtr + data.Offset, data.Count);
|
|
}
|
|
}
|
|
|
|
TailIndex += sizeof(int) + data.Count;
|
|
}
|
|
|
|
/// <summary>Append a new message to the queue.</summary>
|
|
/// <param name="message">Message to append to the queue.</param>
|
|
/// <returns>
|
|
/// Whether the message was appended successfully. The only way it can fail is if there's
|
|
/// no more room in the queue. On failure, nothing is written to the queue.
|
|
/// </returns>
|
|
public bool PushMessage(ArraySegment<byte> message)
|
|
{
|
|
if (!IsCreated)
|
|
{
|
|
return false;
|
|
}
|
|
|
|
// Check if there's enough room after the current tail index.
|
|
if (Capacity - TailIndex >= sizeof(int) + message.Count)
|
|
{
|
|
AppendDataAtTail(message);
|
|
return true;
|
|
}
|
|
|
|
// Move the data at the beginning of of m_Data. Either it will leave enough space for
|
|
// the message, or we'll grow m_Data and will want the data at the beginning anyway.
|
|
if (HeadIndex > 0 && Length > 0)
|
|
{
|
|
unsafe
|
|
{
|
|
#if UTP_TRANSPORT_2_0_ABOVE
|
|
UnsafeUtility.MemMove(m_Data.GetUnsafePtr(), m_Data.GetUnsafePtr() + HeadIndex, Length);
|
|
#else
|
|
UnsafeUtility.MemMove(m_Data.GetUnsafePtr(), (byte*)m_Data.GetUnsafePtr() + HeadIndex, Length);
|
|
#endif
|
|
}
|
|
|
|
TailIndex = Length;
|
|
HeadIndex = 0;
|
|
}
|
|
|
|
// If there's enough space left at the end for the message, now is a good time to trim
|
|
// the capacity of m_Data if it got very large. We define "very large" here as having
|
|
// more than 75% of m_Data unused after adding the new message.
|
|
if (Capacity - TailIndex >= sizeof(int) + message.Count)
|
|
{
|
|
AppendDataAtTail(message);
|
|
|
|
while (TailIndex < Capacity / 4 && Capacity > m_MinimumCapacity)
|
|
{
|
|
m_Data.ResizeUninitialized(Capacity / 2);
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
// If we get here we need to grow m_Data until the data fits (or it's too large).
|
|
while (Capacity - TailIndex < sizeof(int) + message.Count)
|
|
{
|
|
// Can't grow m_Data anymore. Message simply won't fit.
|
|
if (Capacity * 2 > m_MaximumCapacity)
|
|
{
|
|
return false;
|
|
}
|
|
|
|
m_Data.ResizeUninitialized(Capacity * 2);
|
|
}
|
|
|
|
// If we get here we know there's now enough room for the message.
|
|
AppendDataAtTail(message);
|
|
return true;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Fill as much of a <see cref="DataStreamWriter"/> as possible with data from the head of
|
|
/// the queue. Only full messages (and their length) are written to the writer.
|
|
/// </summary>
|
|
/// <remarks>
|
|
/// This does NOT actually consume anything from the queue. That is, calling this method
|
|
/// does not reduce the length of the queue. Callers are expected to call
|
|
/// <see cref="Consume"/> with the value returned by this method afterwards if the data can
|
|
/// be safely removed from the queue (e.g. if it was sent successfully).
|
|
///
|
|
/// This method should not be used together with <see cref="FillWriterWithBytes"> since this
|
|
/// could lead to a corrupted queue.
|
|
/// </remarks>
|
|
/// <param name="writer">The <see cref="DataStreamWriter"/> to write to.</param>
|
|
/// <param name="softMaxBytes">
|
|
/// Maximum number of bytes to copy (0 means writer capacity). This is a soft limit only.
|
|
/// If a message is larger than that but fits in the writer, it will be written. In effect,
|
|
/// this parameter is the maximum size that small messages can be coalesced together.
|
|
/// </param>
|
|
/// <returns>How many bytes were written to the writer.</returns>
|
|
public int FillWriterWithMessages(ref DataStreamWriter writer, int softMaxBytes = 0)
|
|
{
|
|
if (!IsCreated || Length == 0)
|
|
{
|
|
return 0;
|
|
}
|
|
|
|
softMaxBytes = softMaxBytes == 0 ? writer.Capacity : Math.Min(softMaxBytes, writer.Capacity);
|
|
|
|
unsafe
|
|
{
|
|
var reader = new DataStreamReader(m_Data.AsArray());
|
|
var readerOffset = HeadIndex;
|
|
|
|
reader.SeekSet(readerOffset);
|
|
var messageLength = reader.ReadInt();
|
|
var bytesToWrite = messageLength + sizeof(int);
|
|
|
|
// Our behavior here depends on the size of the first message in the queue. If it's
|
|
// larger than the soft limit, then add only that message to the writer (we want
|
|
// large payloads to be fragmented on their own). Otherwise coalesce all small
|
|
// messages until we hit the soft limit (which presumably means they won't be
|
|
// fragmented, which is the desired behavior for smaller messages).
|
|
|
|
if (bytesToWrite > softMaxBytes && bytesToWrite <= writer.Capacity)
|
|
{
|
|
writer.WriteInt(messageLength);
|
|
#if UTP_TRANSPORT_2_0_ABOVE
|
|
WriteBytes(ref writer, m_Data.GetUnsafePtr() + reader.GetBytesRead(), messageLength);
|
|
#else
|
|
WriteBytes(ref writer, (byte*)m_Data.GetUnsafePtr() + reader.GetBytesRead(), messageLength);
|
|
#endif
|
|
|
|
return bytesToWrite;
|
|
}
|
|
else
|
|
{
|
|
var bytesWritten = 0;
|
|
|
|
while (readerOffset < TailIndex)
|
|
{
|
|
reader.SeekSet(readerOffset);
|
|
messageLength = reader.ReadInt();
|
|
bytesToWrite = messageLength + sizeof(int);
|
|
|
|
if (bytesWritten + bytesToWrite <= softMaxBytes)
|
|
{
|
|
writer.WriteInt(messageLength);
|
|
#if UTP_TRANSPORT_2_0_ABOVE
|
|
WriteBytes(ref writer, m_Data.GetUnsafePtr() + reader.GetBytesRead(), messageLength);
|
|
#else
|
|
WriteBytes(ref writer, (byte*)m_Data.GetUnsafePtr() + reader.GetBytesRead(), messageLength);
|
|
#endif
|
|
|
|
readerOffset += bytesToWrite;
|
|
bytesWritten += bytesToWrite;
|
|
}
|
|
else
|
|
{
|
|
break;
|
|
}
|
|
}
|
|
|
|
return bytesWritten;
|
|
}
|
|
}
|
|
}
|
|
|
|
/// <summary>
|
|
/// Fill the given <see cref="DataStreamWriter"/> with as many bytes from the queue as
|
|
/// possible, disregarding message boundaries.
|
|
/// </summary>
|
|
/// <remarks>
|
|
/// This does NOT actually consume anything from the queue. That is, calling this method
|
|
/// does not reduce the length of the queue. Callers are expected to call
|
|
/// <see cref="Consume"/> with the value returned by this method afterwards if the data can
|
|
/// be safely removed from the queue (e.g. if it was sent successfully).
|
|
///
|
|
/// This method should not be used together with <see cref="FillWriterWithMessages"/> since
|
|
/// this could lead to reading messages from a corrupted queue.
|
|
/// </remarks>
|
|
/// <param name="writer">The <see cref="DataStreamWriter"/> to write to.</param>
|
|
/// <param name="maxBytes">Max number of bytes to copy (0 means writer capacity).</param>
|
|
/// <returns>How many bytes were written to the writer.</returns>
|
|
public int FillWriterWithBytes(ref DataStreamWriter writer, int maxBytes = 0)
|
|
{
|
|
if (!IsCreated || Length == 0)
|
|
{
|
|
return 0;
|
|
}
|
|
|
|
var maxLength = maxBytes == 0 ? writer.Capacity : Math.Min(maxBytes, writer.Capacity);
|
|
var copyLength = Math.Min(maxLength, Length);
|
|
|
|
unsafe
|
|
{
|
|
#if UTP_TRANSPORT_2_0_ABOVE
|
|
WriteBytes(ref writer, m_Data.GetUnsafePtr() + HeadIndex, copyLength);
|
|
#else
|
|
WriteBytes(ref writer, (byte*)m_Data.GetUnsafePtr() + HeadIndex, copyLength);
|
|
#endif
|
|
}
|
|
|
|
return copyLength;
|
|
}
|
|
|
|
/// <summary>Consume a number of bytes from the head of the queue.</summary>
|
|
/// <remarks>
|
|
/// This should only be called with a size that matches the last value returned by
|
|
/// <see cref="FillWriter"/>. Anything else will result in a corrupted queue.
|
|
/// </remarks>
|
|
/// <param name="size">Number of bytes to consume from the queue.</param>
|
|
public void Consume(int size)
|
|
{
|
|
// Adjust the head/tail indices such that we consume the given size.
|
|
if (size >= Length)
|
|
{
|
|
HeadIndex = 0;
|
|
TailIndex = 0;
|
|
|
|
// This is a no-op if m_Data is already at minimum capacity.
|
|
m_Data.ResizeUninitialized(m_MinimumCapacity);
|
|
}
|
|
else
|
|
{
|
|
HeadIndex += size;
|
|
}
|
|
}
|
|
}
|
|
}
|