* Move RobustXaml to a shared package
In a near-future change, I'll make it possible to optionally link to
this from Robust.Client, which will allow JIT compiling XAML.
Also upgrade it to a version of .NET that supports nullability
annotations.
* Re-namespace packages
* Add a JIT compiler, plus hooks that call into it
In Debug, after this change, all XAML will be hot reloaded once every
time an assembly is reloaded.
The new code is compiled with SRE and is _not_ sandboxed -- this is not
suitable to run against prod.
In Release, the hot reload path is totally skipped, using the same trick
as SmugLeaf used in an earlier attempt to implement this functionality.
* Hot reload: watcher
This is a bit of a horror, but there's not in-engine support for
identifying the source tree or the XAML files in it.
* Put everything dangerous behind conditional comp
* Code cleanup, docs
* Fix a bad comment
* Deal a little better with crashes in the watcher
* Make reload failures Info, since they're expected
They were previously causing the integration tests to flag, even though
"a few types fail hot reloading because they're internal" is expected
behavior.
* Fix an unnecessary null check
I removed the ability for CompileCore to return null.
* injectors: null! strings, default primitives
* Tidy documentation (thanks, PJB!)
* Reinstate netstandard2.0, abolish Pidgin
* Internal-ize all of Robust.Xaml
* Add a cautionary note to Sandbox.yml
* Shuffle around where conditional compilation occurs
* Privatize fields in XamlImplementationStorage
* Internalize XamlJitDelegate
* Inline some remarks. No cond. comp in Robust.Xaml
* Use file-scoped namespaces
They aren't allowed at Language Level 8.0. (which I arbitrarily picked
for Robust.Xaml because it's the oldest one that would work)
* Bump language level for R.Xaml, file namespaces
* Force hot reloading off for integration tests
* Fix bizarre comment/behavior in XamlImplementationStorage
* Consistently use interfaces, even in generated code
* Update Robust.Client/ClientIoC.cs
---------
Co-authored-by: Pieter-Jan Briers <pieterjan.briers@gmail.com>
* Isolate net messages in integration tests.
Integration tests don't use Lidgren to connect client and send, instead they just use some in-process channels to communicate. Because of this, the original implementation of net messages *directly* passed the net message instances between client and server instances. This caused issues whenever content would mutate data in a NetMessage after it "passed through".
Now we run the messages through WriteToBuffer() and ReadFromBuffer() so they pass through binary serialization. This means there's no more implicit sharing of the objects.
Note that this requires some trickery: Lidgren's message types have internal constructors. Really ideally we'd change the engine to make this more testable... but that's a content breaking change. Instead I just added InternalsVisibleTo to Lidgren so we can mess with it. We maintain the library ourselves anyways I can do what I want.
Fixes#4836
* Register Robust.UnitTesting as assembly for reflection.
This is necessary so that serialized types in the assembly can be picked up by NetSerializer.
Have to disable automatic reflection on all entity systems/components that tests register manually right now, because otherwise tests break.
* Stop shallow cloning specific net messages in integration tests.
This isn't necessary anymore now that we have a thorough fix.
* Wow I really forgot to copy-paste that line to the other side huh.
* Add test that serializer hash matches.
* Another test one I missed earlier.
* Changelog
* Trying to get res:// and usr:// cookies to wrok
* Update CefGlue
* Bump CEF
* Seal types in WebView
Fixes warnings
* Move most of client cleanup to game thread.
This used to run in the windowing thread which broke CEF shutdown, most notably cookie saving.
* "flushcookies" command for CEF.
* Remove unecessary res:// code.
* Fix tests
* More request handler fixes.
* Good thing I don't have to care about commit quality in PRs like this.
* Add test pooling and global test setup
* WIP test pooling changes
* Make asynchronous tests the default again
* Finish fixing tests, make test threads background threads
* Un-pool tests with custom cvars
* Fix not changing FailureLogLevel cvar on instance return
* Fix error when overriding already overriden cvar
* Don't pool some physics integration tests
* Unpool engine tests, the technology just isn't there yet
* Remove explicit Pool = false from physics tests
* Change default pooling setting to false
* Didn't need this anyway
* Remove ConfigurationManager.ResetOverrides
* Bring back enum cvar override parsing
* Make integrationInstances name clearer > notPooledInstaces
Make clients ready and servers ready internal
* Add logging test instances
* Give more info on ran tests
* Show total clients and servers in test output
* Wipe LayerMap on SpriteComponent.AfterDeserialize
* Fix server not properly kicking clients
* Rider shut
* Make all test metrics report totals
* Format tests ran better
* Replace Console.WriteLine with TestContext.Out.WriteLine
* Fix two server test pooling info prints using total clients instead of total servers
Co-authored-by: metalgearsloth <comedian_vs_clown@hotmail.com>
* Adds ServerOptions, improve GameControllerOptions, fix engine integration tests
* Do component auto-registration in engine integration tests by default
* Fix integration tests on content, register components ONLY if not contentstarted or options are null
* Add integration test for engine integration tests working correctly
* Move cvar overrides out of content and into engine.
* Removed the Interfaces folder.
* All objects inside the GameObjects subfolders are now in the GameObjects namespace.
* Added a Resharper DotSettings file to mark the GameObjects subfolders as not providing namespaces.
* Simplified Robust.client.Graphics namespace.
* Automated remove redundant using statements.
* Renamed shared ICommand to IConsoleCommand.
* Lots of refactoring into a shared context.
* Removed ICommonSession from server concmd Execute.
* Added argStr parameter to concmd execute.
* The execute function of client concmds now returns void, use the new shell.RemoteExecuteCommand function to forward commands.
# Conflicts:
# Robust.Client/Console/Commands/Debug.cs
* Finally move shells and commands into shared.
* Console commands can now be registered directly without a class in a shared context.
* Pulled up ConsoleHost and Console shell into a shared context.
* Pulled up half the functions of ConsoleHost into a shared context.
* Repair rebase damage.
* Make LoadConsoleCommands function not remove any previously registered commands.
1. Set timeBeginPeriod(3) on the server to reduce scheduler latency in the lidgren thread.
2. Add 16ms of guaranteed lag bias to client prediction calculations to account for scheduler latency.
Both of these changes are to account for how the windows scheduler seems to handle time periods in related to socket polls. See this Discord conversation for why, details down below as well: https://discord.com/channels/310555209753690112/770682801607278632/798309250291204107
Basically Windows has this thing called time periods which determines the precision of sleep operations and such. By default it's like 16ms so a sleep will only be accurate to within 16ms.
Problem: Lidgren polls the socket with a timeout of 1ms.
The way Windows seems to handle this is that:
1. if a message comes into the socket, the poll immediately ends and Lidgren can handle it.
2. If nothing comes in, it takes the whole 16ms time period to actually process stuff.
Oh yeah, and Lidgren's thread needs to keep pumping at a steady rate or else it *won't flush its send queue*. On Windows it seems to normally pump at 65/125 Hz. On Linux it goes like 950 Hz as intended.
Now, the worst part is that (1) causes Lidgren's latency calculation to always read 0 (over localhost) instead of the 30~ms it SHOULD BE (assuming client and server localhost).
That 30ms of unaccounted delay worst caseis enough to cause prediction undershoot and have messages arrive too late. Yikes.
So, to fix this...
On the server we just decrease the tick period and call it a day. Screw your battery life players don't have local servers running anyways.
On the client we bias the prediction calculations to account for this "unmeasurable" lag.
Of course, all this can be configured via CVars.
