* Making ReaderWriterProcessor return if it found custom functions (not counting ones in mirror.dll)
* Always writing generateContainerClass if modified
* processing PropertySiteProcessor if either ReaderWriterProcessor or WeaveModule are successful
* creating generateContainerClass in WeaverList constructor
* Moving new WeaverList to after WeaverTypes.SetupTargetTypes
* feat: adding MiddlewareTransport
* Allows Middleware to be created at the transport layer
* Middleware can choice which methods they want to override and which to let inner handle
* using expression bodies
* fix: improving errors in SendTargetRPCInternal
* now give error when connection is null
* updating error message to accurately explain what went wrong
* clientOwnedObjects is a hash set which doesn't need to be disposed
* DestroyOwnedObjects already clears list which is called by networkmanager when connection disconnects
Co-authored-by: vis2k <info@noobtuts.com>
* moving check to base connection because it could be used for either
* renaming to IsAlive
* passing in timeout so it doesn't need to check NetworkServer
* updating doc comment
The code to serialize lists, arrays, and array segments is now in the runtime instead of generated in the weaver.
removes about 300 LOC of weaver witchcraft
* perf: use built in == instead of unity one
should be the same performance as #2325 without the complexity
* more readable version
* more readable version
* Intial commit
see https://github.com/MirrorNetworking/SimpleWebTransport for full git histroy
* Update main.yml
Adding SimpleWebTransport to coverage exclusions
* fully ignoring SimpleWebTransport from sonarcloud
SimpleWebTransport has sonarcloud running on its own repo
* using Buffer.BlockCopy
* SimpleWebTransport v0.4.1
* making error messages use LogError (MirrorNetworking/SimpleWebTransport@5f0096c)
* messages with length of 1 can be recieved by the client (MirrorNetworking/SimpleWebTransport@4237fc4)
* stopping un-needed error logs after disconnect (MirrorNetworking/SimpleWebTransport@25ff190)
* updating readme
* SimpleWebTransport v0.5.0
* adding missing warn level to Log.Levels (MirrorNetworking/SimpleWebTransport@0f68498)
* Allowing multiple instances of javascript client (MirrorNetworking/SimpleWebTransport#14) (MirrorNetworking/SimpleWebTransport@cc6e513)
* SimpleWebTransport v0.5.8
* Stopping send error when client disconnects
* Adding buffer pool
* Using ArrayBuffer for receive
* Removing info and verbose logging by default so performance isnt effected in editor or development builds.
* Making WebSocketClientStandAlone use timeouts from inspector
* Removing incorrect header from inspector
* Removing readonly from logger so that it can be set
* Fixing use of Interlocked in buffer pools
* using ManualResetEventSlim
* making Log.Exception always be logged
* SimpleWebTransport v0.6.0
* Trying to fix error when connection is closed
* Increase arraybuffer speed
* Removing shared buffers from server handshake
* Improving debug logging
* fixing code smells
* SimpleWebTransport v0.6.1
* making sure pending connection work when stop serve is called
* removing exception that isnt directly thrown by this method
this breaks docfx for some reason
* Adding new websocket transport soon that doesn't depend on library
* Code for old transport can now be found here https://github.com/MirrorNetworking/NinjaWebSocketsTransport
* We shouldn't need to replace with empty files from store as having this in project should not break things
* perf: NetworkIdentity DirtyComponentsMask code removed entirely. OnSerialize now includes the component index as byte before serializing each component. Faster because we avoid GetDirtyComponentsMask() and GetSyncModeObserversMask() calculations. Increases allowed NetworkBehaviour components from 64 to 255. Bandwidth is now smaller if <8 components, and larger if >8 components. Code is significantly more simple.
* Update Assets/Mirror/Runtime/NetworkIdentity.cs
Co-authored-by: James Frowen <jamesfrowendev@gmail.com>
Co-authored-by: James Frowen <jamesfrowendev@gmail.com>
* Users must initialize syncobjects (#391)
Previously we initialized syncobjects, so this is valid:
```cs
public class Pepe : NetworkBehavior {
public SyncList<int> mylist;
}
```
With this change, users must initialize their own fields:
```cs
public class Pepe : NetworkBehavior {
public SyncList<int> mylist = new SyncList<int>();
}
```
BREAKING CHANGE: You must initialize all your SyncLists
* Add null check
* This is no longer a weaver error
* Update Assets/Mirror/Runtime/NetworkBehaviour.cs
Co-authored-by: James Frowen <jamesfrowendev@gmail.com>
* Remove unnecesary using
Co-authored-by: James Frowen <jamesfrowendev@gmail.com>
* Server Teleport Enchancement for NetworkTransform
Related to this issue: https://github.com/vis2k/Mirror/issues/2200
* Improvements on NetworkTransformBase
We disable clientAuthority until teleportation is finished and acknowledged. We set the goal and start data points to null before teleportation.
We set the clientAuthority back to it's initial value registered on server, both on the server and client side; it seems like we don't need to change it to anything on Client side because to my understanding clientAuthortiy is not affecting anything on client side. I may be wrong but still it is a great practice to set it back to it's initial value just in case...
* Added teleportation with rotation
I forgot to add a rotation functionality on my previous commit. It wouldn't be a teleportation without also changing the rotation. If provided a rotation input it will teleport with a rotation, if not it will simply protect the original rotation before the teleportation.
* Removed my debug statement
I forgot to remove the debug log statement lol silly me
* fixing up code
* resetting goal and last pos on server and client
* renaming variables
* making sure TeleportFinished can't request authority at any time
* moving teleport block
* removing duplicate lines
Co-authored-by: Emre Bugday <47198270+EmreB99@users.noreply.github.com>
* breaking: no need to override Serialize/Deserialize in messages
Messages no longer serilize themselves. This has been decoupled. Serializing a message is now done
via readers and writers, which can be either generated or user provided.
This lifts some restrictions,
* you no longer need to have a default constructor in messages
* Messages types can be recursive
* struct Messages don't need to provide an empty Serialize and Deserialize method
Before:
```cs
public struct ReadyMessage : IMessageBase
{
public void Deserialize(NetworkReader reader) { }
public void Serialize(NetworkWriter writer) { }
}
```
After:
```cs
public struct ReadyMessage : IMessageBase
{
}
```
BREAKING CHANGE: Messages must be public
BREAKING CHANGE: Use custom reader and writer instead of Serialize/Deserialize methods
* Remove unused method
* remove unused methods
* remove unused methods
* make all messages struct
* Fix test code generator
* Get rid of MessageBase
* Rename IMessageBase -> NetworkMessage
* add MessageBase as obsolete
* Use a default request
* Empty file to make asset store happy
* Apply suggestions from code review
Co-authored-by: James Frowen <jamesfrowendev@gmail.com>
Co-authored-by: James Frowen <jamesfrowendev@gmail.com>
* Fixed OnStopClient not being called on Client Side
* Moved identity.OnStopClient() two scopes above
Co-authored-by: Emre Bugday <47198270+EmreB99@users.noreply.github.com>
* feat: Use SyncLists directly
Previously, you had to write an intermediary class to use synclists, syncsets and syncdictionaries.
The weaver would populate that intermediary class with a serialization and deserialization method
This PR gets rid of 90% of the weaver code for synclists.
There is no need to generate these methods anymore.
Instead the lists use `writer.Write<T>` and `read.read<T>` to serialize their Data.
Since there is no code generate in synclists, you can now use the synclists directly instead
of subclassing them.
BEFORE:
```cs
public class MyComponent : NetworkBehaviour {
// nonsense class to make the weaver happy
class SyncListData : Synclist<Data> {}
SyncListData mySyncList;
}
```
AFTER:
```cs
public class MyComponent : NetworkBehaviour {
Synclist<Data> mySyncList;
}
```
* linting
* feat: Use SyncLists directly (no overrides)
Previously, you had to write an intermediary class to use synclists and syncdictionaries.
The weaver would populate that intermediary class with a serialization and deserialization method
This PR gets rid of 90% of the weaver code for synclists.
There is no need to generate these methods anymore.
Instead the lists use `writer.Write<T>` and `read.read<T>` to serialize their Data.
Since there is no code generate in synclists, you can now use the synclists directly instead
of subclassing them.
Same as #2305 ,but it removes the deprecated Serialize and Deserialize methods from syncobjects,
This way you get a nice compiler error for the code that no longer works, instead of a warning you might accidentally ignore.
BEFORE:
```cs
public class MyComponent : NetworkBehaviour {
// nonsense class to make the weaver happy
class SyncListData : Synclist<Data> {}
SyncListData mySyncList;
}
```
AFTER:
```cs
public class MyComponent : NetworkBehaviour {
Synclist<Data> mySyncList;
}
```
BREAKING CHANGE: Serialize and Deserialize methods in synclists don't do anything anymore
* Remove old comment
* Fix compilatio error
Currently in mirror, there is no way for Mirror itself to invoke
readers and writers for types it does not know about.
This causes us to have to generate code for lists, dictionaries and messages.
This PR makes it so that if there is a reader and writer anywhere in the
code for type X, then the writer can be invoked like this:
```cs
X x = ...;
writer.Write<X>(x);
```
and the reader can be invoked like this:
```cs
X x = reader.Read<X>();
```
* fix: generic arguments lookup
The weaver was not able to figure out the synclist type in code like this:
```cs
public class SomeList<G, T> : SyncList<T> { }
public class SomeListInt : SomeList<string, int> { }
```
This code fixes that problem, and greatly reduces the complexity
of argument lookup.
* linting
Remove all the recursionCount nonsense.
This was added to prevent infinite recursion with types that reference themselves.
No need to check anymore, the weaver can generate readers and writers for types that reference themselves such as:
```cs
class Tree {
Tree child1;
Tree child2;
}
```
This works by the weaver doing it the way the compiler does: Create a function first, memoize it, then write the body. If the body needs the function, it will get itself and issue a call to itself.
Currently, whenever we want to serialize an enum,
we simply serialize the underlying type (byte, short, int)
This works fine, but in order to get the reader and writer
at runtime I need a function for each type.
With this PR, we generate a reader and writer function for enums too,
the function body simply calls the underlying reader and writer.
* throwing Exception instead of returning null
* re-adding null check until later PR
* removing error from test
* removing extra error
* adding comment
* replacing error with throw for abstact
The plan is to remove WeaverLists at some point so moving the functions
out of weaver to here for now and then move them closer to where they
are actaully used when we start to remove WeaverLists.
# Before
This is hard to read:
```
if (ca.AttributeType.FullName == typeof(Mirror.CommandAttribute).FullName) {
...
}
```
# After
```
if (ca.AttributeType.Is<Mirror.CommandAttribute>()) {
...
}
```
# Before
WeaverTypes declared a static variable like this:
```cs
public static TypeReference int32Type;
```
Weavertypes then loaded the variable in SetupTargetTypes like this:
```cs
int32Type = ImportSystemModuleType(currentAssembly, systemModule, "System.Int32");
```
Note we lookup the type with a string, any typo there won't be detected until the code is executed. It will also not be found by the IDE when you click "find references", or when you refactor.
Then we use it like this:
```cs
var intType = WeaverTypes.int32Type;
```
This is a DRY violation: I have to modify 3 different code sections just to say that I want the int32 type.
# After
Get rid of all the duplication, and use the type instead of a string:
```cs
var intType = WeaverTypes.Import<int>();
```
No need for static variable, or to add it in SetupTargetTypes.
* fixing null ref error when local player is destroyed
* removing null objects
* fixing InvalidOperationException
* moving code to its own function
* Update Assets/Mirror/Runtime/ClientScene.cs
* Update ClientScene.cs
Improving logging
- making sure sceneId is always hex
- only logging "Client spawn for ..." if it was successful
* Update ClientSceneTests_OnSpawn.cs
- Adding handlers in Awake instead of Start to make sure they are added before start on headless
- using GetComponent for NetworkManagerListServer
- Adding asserts to make sure values are set
* moving FakeNetworkConnection to common test folder
* adding test for local player destroy
* adding method to clear local player
* temp move test functions
* adding test for destroying non player
* moving asset to setup
* adding tests for destroy message and host mode
* adding ClearLocalPlayer to reset
* Removing manual invoke for Cmd and RPC
There should be no reason to manaully invoke remote code
they are marked with EditorBrowsableState.Never which implies they should not be called by the user
* removing tests
* adding note to guides about removed SyncEvent
* updating Obsolete
making Obsolete error whe used
updating text to say it has been removed
linking to issue
* adding Obsolete note to other references to SyncEvent
* Remove ScriptTemplates to clean up folder
* Update main.yml
Removing script templates from CI
Co-authored-by: James Frowen <jamesfrowendev@gmail.com>
* fix adding error for gernated read writer for abstract class
weaver can not initialize class abstract class so cant not create a reader
this gives a helpful error telling the server to make a custom reader
* adding tests for error message
* fixing typo
* renaming
* fixing expected error messages
Reader code doesn't run on Rpc now if a valid writer can not be found
In future PR write and read functions should be generated together as they share most of the same checks
* replacing assert with haserror
* adding IsSuccess and HasWarning functions
* using IsSuccess
* checking weaverWarnings empty in success
* using HasWarning
* replacing asserts missed by regex
* fixing typo
* adding HasNoErrors function to be used when tests only have warnings
* fixing typo
* making fields public to stop warnings
* simplifying more test files
