Fixing crash when Editor Application is quitting in Runtime Mode

.github/workflows/pre-commit.yml

@@ -13,7 +13,11 @@ jobs:
   pre-commit:
     runs-on: ubuntu-24.04
     env:
-        DOTNET_NOLOGO: 1
+      DOTNET_NOLOGO: 1
+      DOTNET_SKIP_FIRST_TIME_EXPERIENCE: "1"
+      DOTNET_CLI_HOME: "${{ github.workspace }}/.dotnet"
+      NUGET_PACKAGES: "${{ github.workspace }}/.nuget/packages"
+      TMPDIR: "${{ github.workspace }}/.tmp"
     steps:
     - uses: actions/checkout@v4
     - uses: actions/setup-python@v5
@@ -25,15 +29,20 @@ jobs:
       with:
         ruby-version: '2.7'
         bundler-cache: true # runs 'bundle install' and caches installed gems automatically
+    - name: Create writable dirs
+      run: |
+        mkdir -p "$DOTNET_CLI_HOME" "$NUGET_PACKAGES" "$TMPDIR"
     - uses: actions/setup-dotnet@v4
       with:
         dotnet-version: '8.0.202'
-    -   name: Clean dotnet shared memory
-        run: sudo rm -rf /tmp/.dotnet/shm
     - name: Install manual dependencies
       run: |
         python -m pip install pre-commit
         pre-commit install
+    - name: Pre-warm dotnet
+      run: |
+        dotnet --info
+        dotnet restore || true
     - name: Run pre-commit
       run: |
         pre-commit run --all-files

.pre-commit-config.yaml

@@ -94,7 +94,8 @@ repos:
     -   id: dotnet-format
         name: dotnet-format
         language: system
-        entry: dotnet format whitespace --folder --include
+        entry: dotnet format whitespace --folder --include . --verify-no-changes
+        pass_filenames: false
         types_or: ["c#"]
     -   id: markdown-link-check
         name: markdown-link-check

.yamato/com.unity.ml-agents-test.yml

@@ -8,7 +8,7 @@ test_editors:
       enableNoDefaultPackages: !!bool true
 
 trunk_editor:
-    - version: trunk
+    - version: 6000.4
       testProject: DevProject
 
 test_platforms:

.yamato/test_versions.metafile

@@ -7,5 +7,5 @@ test_editors:
     extra_test: gym
   - version: 6000.0
     extra_test: sensor
-  - version: trunk
+  - version: 6000.3
     extra_test: llapi

CLAUDE.md

@@ -0,0 +1,178 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## Project Overview
+
+The Unity ML-Agents Toolkit is a sophisticated machine learning infrastructure project that bridges Unity game development (C#) with Python-based ML training. It enables games and simulations to serve as environments for training intelligent agents using reinforcement learning, imitation learning, and other ML techniques.
+
+This is a **monorepo** with multiple interconnected packages:
+- `com.unity.ml-agents/` - Unity Package (C# SDK)
+- `ml-agents/` - Python Trainer Package
+- `ml-agents-envs/` - Python Environment Interface
+- `ml-agents-trainer-plugin/` - Plugin framework
+- `protobuf-definitions/` - Communication protocol definitions
+- `config/` - Training configuration examples
+- `Project/` and `DevProject/` - Example Unity projects
+
+## Common Development Commands
+
+### Python Package Installation (Development Mode)
+```bash
+# Install packages in editable mode for development
+pip install -e ./ml-agents-envs
+pip install -e ./ml-agents
+pip install -e ./ml-agents-trainer-plugin
+pip install -e ./ml-agents-plugin-examples
+
+# Install test dependencies
+pip install -r test_requirements.txt
+```
+
+### Testing
+```bash
+# Run all tests except slow ones (default for development)
+pytest -m "not slow"
+
+# Run tests with parallel execution (8 workers)
+pytest -m "not slow" -n 8
+
+# Run tests with coverage reporting
+pytest --cov=ml-agents --cov=ml-agents-envs --cov-report html
+
+# Run slow tests (training-related, takes much longer)
+pytest -m slow
+
+# Run specific test file
+pytest ml-agents/mlagents/trainers/tests/test_buffer.py
+
+# Run specific test function
+pytest ml-agents/mlagents/trainers/tests/test_buffer.py::test_buffer_functionality
+```
+
+### Code Quality and Formatting
+```bash
+# Run pre-commit hooks on all files
+pre-commit run --all-files
+
+# Run specific hooks
+pre-commit run black --all-files
+pre-commit run mypy --all-files
+pre-commit run flake8 --all-files
+
+# Format C# code
+dotnet format whitespace --folder --include
+
+# Check markdown links (manual stage)
+pre-commit run --hook-stage manual markdown-link-check
+```
+
+### Training Commands
+```bash
+# Basic training command
+mlagents-learn config/ppo/3DBall.yaml --env=Project/3DBall --run-id=my_run
+
+# Training with curriculum learning
+mlagents-learn config/curricula/Wall.yaml --env=Project/WallJump --run-id=wall_curriculum
+
+# Resume training from checkpoint
+mlagents-learn config/ppo/3DBall.yaml --env=Project/3DBall --run-id=my_run --resume
+
+# Push trained model to Hugging Face
+mlagents-push-to-hf --run-id=my_run --repo-name=my-org/my-model
+```
+
+### Building and Unity Integration
+```bash
+# Unity Package Manager operations (from Unity Editor)
+# Window > Package Manager > + > Add package from git URL
+# https://github.com/Unity-Technologies/ml-agents.git?path=com.unity.ml-agents
+
+# Build Unity project (from Unity Editor or command line)
+Unity -batchmode -quit -projectPath ./Project -buildTarget StandaloneLinux64
+```
+
+## Architecture Overview
+
+### Communication Flow
+The system uses a **gRPC-based architecture**:
+```
+Unity Environment ←→ gRPC ←→ Python Trainer
+    (C# SDK)                    (PyTorch/ML)
+```
+
+### Key Components
+
+**Python Side:**
+- **Trainers** (`ml-agents/mlagents/trainers/`): Core ML algorithms (PPO, SAC, MA-POCA, GAIL, BC)
+- **Environment Interface** (`ml-agents-envs/`): Communication with Unity environments
+- **Plugin System**: Extensible trainer types and stats writers via entry points
+
+**Unity Side:**
+- **Agent** (`com.unity.ml-agents/Runtime/Agent.cs`): Base class for trainable agents
+- **Sensors**: Observation collection (vision, raycast, grid sensors)
+- **Actuators**: Action execution (continuous/discrete)
+- **Communicator**: gRPC communication layer
+
+### Multi-Package Structure
+
+The project uses **separate Python packages** with careful version management:
+- Packages can be installed independently
+- Version constraints are strictly managed (see `setup.py` files)
+- Plugin system uses entry points for extensibility
+
+## Development Workflow
+
+### Code Quality Standards
+- **Python**: Black formatting (120 char line length), mypy type checking, flake8 linting
+- **C#**: dotnet-format for whitespace formatting
+- **Testing**: pytest with 60% minimum coverage requirement
+- **Pre-commit hooks**: Enforced via `.pre-commit-config.yaml`
+
+### Important Constraints
+- **Python Version**: 3.10.1-3.10.12 only
+- **PyTorch**: ≤2.8.0 (strict constraint to prevent ONNX breaking changes)
+- **Unity**: 6000.0+ (Unity 6)
+- **Banned modules**: Direct tensorflow/torch imports (use utils instead)
+
+### Testing Strategy
+- **Parallel execution**: Tests run with 8 workers using pytest-xdist
+- **Port allocation**: Custom `base_port` fixture prevents port collisions
+- **Slow tests**: Training-related tests marked with `@pytest.mark.slow`
+- **Coverage**: HTML reports available via `--cov-report html`
+
+### Plugin Development
+Extend functionality via entry points in `setup.py`:
+```python
+entry_points={
+    ML_AGENTS_STATS_WRITER: ["custom_writer=my_package:MyStatsWriter"],
+    ML_AGENTS_TRAINER_TYPE: ["custom_trainer=my_package:MyTrainer"]
+}
+```
+
+## Key Files for Development
+
+**Configuration:**
+- `config/` - Training configuration YAML files for different algorithms
+- `.pre-commit-config.yaml` - Code quality enforcement
+- `pytest.ini`, `setup.cfg` - Test and coverage configuration
+
+**Architecture Entry Points:**
+- `ml-agents/mlagents/trainers/learn.py` - Main training script
+- `ml-agents-envs/mlagents_envs/environment.py` - Environment interface
+- `com.unity.ml-agents/Runtime/Agent.cs` - Unity agent base class
+
+**Documentation:**
+- Primary docs: [Unity Package Documentation](https://docs.unity3d.com/Packages/com.unity.ml-agents@latest)
+- Legacy docs: `docs/` directory (deprecated)
+
+## Release and Versioning
+
+- **Main branch**: `develop` (unstable development)
+- **Release branch pattern**: `release_XX`
+- **Version validation**: Pre-commit hooks ensure version consistency across packages
+- **Git tags**: Must match `__release_tag__` in `__init__.py` files
+
+## Port Management
+
+Tests use a **file-lock based port allocation system** (`conftest.py`) to enable parallel execution without port conflicts. Ports are dynamically allocated starting from a base port calculated per test file.

com.unity.ml-agents/CHANGELOG.md

@@ -11,6 +11,7 @@ and this project adheres to
 #### com.unity.ml-agents (C#)
 - Fixed tensor indexing to use correct CHW layout (#6239)
 - Updated the installation doc (#6242)
+- Fixed Unity Editor crashing when quitting in play mode (#6267)
 
 #### ml-agents / ml-agents-envs
 - Set the Torch version constraint to 2.8 (#6251)

com.unity.ml-agents/Runtime/Academy.cs

@@ -261,6 +261,7 @@ protected Academy()
             LazyInitialize();
 
 #if UNITY_EDITOR
+            EditorApplication.quitting += Dispose;
             EditorApplication.playModeStateChanged += HandleOnPlayModeChanged;
 #endif
         }

ml-agents-envs/mlagents_envs/registry/unity_env_registry.py

@@ -121,6 +121,6 @@ def __iter__(self) -> Iterator[Any]:
 
 default_registry = UnityEnvRegistry()
 # TODO restore when a new registry is available.
-default_registry.register_from_yaml(
-    "https://storage.googleapis.com/mlagents-test-environments/1.1.0/manifest.yaml"
-)  # noqa E501
+# default_registry.register_from_yaml(
+#    "https://storage.googleapis.com/mlagents-test-environments/1.1.0/manifest.yaml"
+# )  # noqa E501