You cannot select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
* Fix code that deregisters the Atom Scene subsystem from the AzFramework Scene The AzFramework Scene subsystem API is a generic container based on the type of argument that is passed to it. It maintains a vector of typeids, and only one object of any type is stored at a time. The Bootstrap system component registers the Atom scene as a `ScenePtr` (aka `AZStd::shared_ptr<RPI:Scene>`) with the AzFramework Scene's generic subsystem. However, the component was previously deregistering the type by value, `RPI::Scene`. Since no subsystem for the type `RPI::Scene` was set, unsetting this type did nothing. The result was that the `RPI::Scene` object would still be around by the time that all the Atom `InstanceDatabse`s were being destroyed, resulting in a large number of errors reported about leaked instances during global shutdown. This fixes the above issue by passing the `m_defaultScene` as a parameter to `AzFramework::Scene::UnsetSubsystem`, the same value that is passed to `SetSubsystem`. This is better, because instead of providing explicit template arguments (which were specifying the incorrect type), this now allows the compiler to deduce the correct type, and the syntax is symmetric with the call to `SetSubsystem`. Signed-off-by: Chris Burel <burelc@amazon.com> * Correctly release the AWS API from the `HttpRequestManager` module This code was incorrectly assuming that `AWSNativeSDKInit::InitializationManager::Shutdown()` would be called automatically by the `InitializationManager` itself. However, all that `InitAwsApi()` does is create an `AZ::EnvironmentVariable`, which is a ref-counted type, and stores it in a global static. That global static is defined in a static library (namely `AWSNativeSDKInit`), which is linked in to the `HttpRequestManager` dynamic lib. Because it is a global static, it has to be explicitly cleared with the call to `Shutdown()`. Otherwise the destructor of the EnvironmentVariable doesn't happen until global destruction, by which time the allocator that is supplied to the AWS SDK has already been destroyed, and the shutdown of the AWS SDK attempts to use the already-destroyed allocator. Signed-off-by: Chris Burel <burelc@amazon.com> * Avoid blocking the remote console server thread if there are no connections The Remote console server runs in a separate thread. Previously, it would directly call `AzSock::Accept()` and block the server thread until some client connected to it. However, if no client connected, the thread would continue to be blocked, even if the game launcher tried to exit. This adds a check to see if there's a client on the socket before calling `Accept()`, to avoid the deadlock on launcher exit. Signed-off-by: Chris Burel <burelc@amazon.com> * Fix a log message to print one message per line Signed-off-by: Chris Burel <burelc@amazon.com> * Allow pumping the event loop to close the launcher window Events from the OS are handled in the game's main loop. The general loop looks like this: * Read events from the OS * Tick the game application One of the events that can come from the OS is that the window hosting the game is closed. When this event happens, many resources provided by the renderer are freed, and the game application's `shouldExit` bit is set. However, when the game's `Tick()` is called, there is lots of code that assumes the renderer is still there. To avoid crashing in the `Tick()` call, check if the game should exit after pumping the system events. Signed-off-by: Chris Burel <burelc@amazon.com> * Unload the level when exiting the launcher This ensures that any resources held onto by the level are freed before the launcher exits. Signed-off-by: Chris Burel <burelc@amazon.com> * Add an explicit bus `Disconnect()` call to `AZCoreLogSink` This is necessary because this bus has virtual functions and can be called from multiple threads. Signed-off-by: Chris Burel <burelc@amazon.com> * Allow normal cleanup to take place when exiting the game launcher Previously, global cleanup was side-stepped by calling `TerminateProcess` or `exit`, when quitting the game launcher. This is in contrast to the call to `_exit` on Linux and Mac when exiting the Editor. That leading `_` makes a big difference: the former runs object destruction, the latter does not. Instead of making the launcher exit with `_exit` on Linux, instead, remove that call and actually run all the atexit code. This does not modify the Editor's behavior however. It still uses `_exit` and `TerminateProcess`. Signed-off-by: Chris Burel <burelc@amazon.com> |
4 years ago | |
|---|---|---|
| .. | ||
| Platform | 4 years ago | |
| Resources | 5 years ago | |
| Tests | 5 years ago | |
| CMakeLists.txt | 4 years ago | |
| FindLauncherGenerator.cmake | 5 years ago | |
| Game.cpp | 5 years ago | |
| Launcher.cpp | 4 years ago | |
| Launcher.h | 4 years ago | |
| LauncherProject.cpp | 5 years ago | |
| Server.cpp | 5 years ago | |
| StaticModules.in | 5 years ago | |
| launcher_files.cmake | 5 years ago | |
| launcher_game_files.cmake | 5 years ago | |
| launcher_generator.cmake | 4 years ago | |
| launcher_project_files.cmake | 4 years ago | |
| launcher_server_files.cmake | 5 years ago | |
| launcher_test_files.cmake | 5 years ago | |