- Add SceneNode and EntityNode classes for editor-only metadata storage
- Implement SceneGraph view-model with O(1) entity lookup via internal caching
- Create IdRemapTable for file-local to global entity ID remapping on load
- Implement SceneSerializationContext for load/save operation tracking
- Add JSON-serializable SceneAssetData, EntityData, and ComponentData models
- Implement SceneSerializer for save/load with validation and reference remapping
- Add comprehensive documentation: README.md, IMPLEMENTATION_GUIDE.md, SYSTEM_SUMMARY.md
- Update Ghost.Editor.Core.csproj to reference Ghost.Entities assembly
- Support parent-child relationships via Hierarchy component
- Enforce no cross-scene entity references
- Keep runtime minimal: only SceneID, Hierarchy, LocalToWorld components
- All editor metadata (names, UI state) stored in editor-only SceneNode/EntityNode classes
This implements the architecture from SceneGraph Plan.md with clean separation of concerns,
minimal runtime footprint, and AOT compatibility.
Replaces ErrorStatus with Error across all systems for consistency.
Renames ResourceBarrierData fields to camelCase.
Adds BindlessAccess enum and updates GetBindlessIndex API.
Updates method signatures, result types, and error checks.
Modernizes HLSL mesh shader syntax and fixes naming.
Improves code style and updates comments for clarity.
Major overhaul of render graph system for modularity and performance:
- Split compilation and execution logic into dedicated classes (Compiler, Executor, NativePassBuilder, Barriers)
- Overhauled barrier system: now uses CompiledBarrier with target state only, querying before state at execution
- Resource size/alignment now queried from D3D12 device for accurate heap allocation
- ResourceDesc now includes Type field and asserts correct union access
- Centralized D3D12 interop logic in D3D12Utility extensions
- Added RenderGraphHasher for structural graph hashing and cache invalidation
- RenderGraph class simplified to orchestrate specialized components
- ResourceAliasingManager now uses allocator for size queries
- Compilation cache now stores compiled barriers, reducing memory usage
- Improved comments, debug assertions, and removed redundant code
Result: more maintainable, efficient, and robust render graph pipeline.
Major overhaul of resource barrier and state tracking in RenderGraph:
- Introduce ResourceBarrierData for explicit (layout, access, sync) tracking.
- Separate aliasing and transition barriers; explicit aliasing support.
- Remove BufferHint; infer buffer usage from BufferUsage flags.
- Update TextureAccess/BufferAccess to include usage requirements.
- Improve enums (BarrierSync, BarrierAccess, BarrierLayout) for D3D12 alignment.
- Update D3D12CommandBuffer to use new barrier data and error handling.
- Make D3D12DescriptorHeap a class; add ReleaseSampler to IResourceDatabase.
- Reset resource pools and aliasing managers each frame.
- Batch and flush barriers efficiently per pass.
- Update HLSL mesh shader macros to [NumThreads].
- Remove obsolete code and improve documentation.
This refactor improves correctness, extensibility, and prepares for advanced features.
Introduces a full-featured render graph system with pass culling, resource aliasing, and automatic barrier generation. Refactors resource and barrier APIs, improves error handling, and unifies result types. Renderer and render passes now use the new graph-based workflow. Updates shader includes, adds a blit shader, and improves HLSL parsing. Removes dynamic descriptor heaps in favor of persistent ones. Project file now includes the render graph module. Lays the foundation for advanced rendering features and improved memory efficiency.
Major architecture upgrade:
- Add native render pass merging (hardware pass grouping, load/store op inference)
- Implement heap-based aliasing for textures & buffers (D3D12-style)
- Unify resource model: buffers and textures in one registry
- Extend builder API for buffer creation/usage, access flags, hints
- Improve barrier/state tracking (buffer hints, indirect argument state)
- Update caching, hashing, and debug output for new model
- Add enums/structs: AttachmentLoadOp, StoreOp, BufferHint, etc.
- D3D12 backend: support named resources, temp upload buffers, correct heap usage
- Update docs, benchmarks, and project files for new features
Brings render graph closer to AAA engine standards, enabling efficient memory usage, lower driver overhead, and a more flexible API.
Major overhaul of Render Graph system:
- Replaced texture handles with generic Identifier<T> for unified, type-safe resource management (textures, buffers, etc.)
- Refactored resource registry and pooling for performance and extensibility
- Added AccessFlags and TextureAccess for precise resource usage tracking
- Split passes into Raster and Compute types; introduced builder interfaces for safer pass construction
- Modernized pass setup API (SetColorAttachment, UseTexture, etc.)
- Updated command buffer and context structs to use new resource system
- Refactored barrier and aliasing logic for improved correctness
- Integrated BenchmarkDotNet for performance/memory benchmarking
- Improved blackboard type safety and removed obsolete code/extensions
- Added BenchmarkDotNet NuGet package
These changes make the Render Graph more extensible, efficient, and ready for future resource types and advanced features.
- Major architectural refactor for performance, extensibility, and feature completeness: resource pooling, pass culling, aliasing, and compilation caching.
- Introduces type-safe builder and context APIs, blackboard pattern, and unified resource management.
- Adds detailed documentation and cleans up obsolete files and APIs.
- Includes (commented) Unity Render Graph source for reference; not compiled, for parity and future extension.
- Major optimization of Ghost.RenderGraph.Concept: pooled resources, zero-allocation hot paths, explicit queue types, and batch barrier APIs.
- Migrated Ghost.DSL shader compiler to ANTLR4-based parser; removed hand-written parser, added grammar files and semantic model conversion.
- Added CollectionPool/ListPool for pooled list management.
- Updated documentation for new architecture and performance.
- Removed Ghost.Shader.Concept (material/material system) from repo and solution.
- README.md replaced with a brief project statement.
Major refactor of shader compiler and related systems:
- Switch ShaderDescriptor/PassDescriptor to arrays; remove IPassDescriptor
- Rewrite keywords block parser/semantic analysis for flexible syntax
- Change property initializers to brace syntax `{ ... }`
- Simplify TokenStream API (remove ref index params)
- Make GetBindlessIndex return uint (~0u for not found)
- Update shader compilation and variant logic for new descriptors
- Update test shader syntax to match new property/keyword formats
- Add AGENTS.md agent development guide
- Add Antlr4 dependency to Ghost.DSL
- Miscellaneous code style and error handling improvements
Major architectural update to graphics/material/shader system:
- Introduced strongly-typed key structs (Key64/Key128) for passes, variants, and pipelines; removed legacy key types.
- Implemented robust hashing and key generation utilities for efficient variant and pipeline lookup/caching.
- Shader compiler now compiles/caches all keyword variants using new key system; includes handled as lists.
- Switched to push constant root signature for per-draw data; updated HLSL and C# codegen accordingly.
- Refactored Material, Shader, and Pass data structures for cache efficiency and variant support.
- Pipeline library and PSO management now use 128-bit keys and variant-specific caching.
- Replaced WorldNode with SceneNode in editor/scene graph; introduced ComponentManager for archetype/query management.
- Migrated math utilities to Misaki.HighPerformance.Mathematics; updated editor controls.
- Updated all HLSL and codegen for new buffer/push constant layouts and macros.
- Misc: project reference cleanup, D3D12 Work Graph support, doc updates, and code modernization.
Introduces a new Ghost.Shader.Concept project implementing a modern, data-oriented material and shader system with:
- Global/local keyword bitsets (fast O(1) ops, 64 bytes)
- Multi-pass shader program and per-pass render state overrides
- Thread-safe, 16-byte aligned material property blocks
- Material pooling to reduce GC pressure
- Batch renderer for efficient PSO grouping and async variant warmup
- Full demo (Program.cs) and extensive documentation (ARCHITECTURE.md, README.md, PROJECT_SUMMARY.md)
- Minor integration: new enums, doc updates, and keyword handling in existing code
No breaking changes to the existing engine; all new code is isolated. This serves as a reference implementation for high-performance, extensible material/shader architectures.
- Replace old pipeline enums/structs with new strongly-typed PipelineState and enums (ZTest, ZWrite, Cull, Blend, ColorWriteMask)
- Redesign pipeline keying: introduce 128-bit GraphicsPipelineKey, MaterialPipelineKey, and PassPipelineKey for robust PSO caching
- Replace IRenderTargetStrategy with IRenderOutput; add SwapChainRenderOutput and TextureRenderOutput
- Update renderer and window code to use new render output abstraction and handle viewport/scissor updates
- Make ShaderPass a readonly struct and Shader a struct; use ID-based pass lookup for efficiency
- Materials now support per-pass pipeline overrides with new keying
- Add defensive checks in D3D12CommandBuffer; update D3D12PipelineLibrary for new keying/state
- Move test shader to test.gsdef and update for new pipeline state syntax
- Remove obsolete files/interfaces and perform general code cleanups
- Update all usages and parsing logic for new pipeline state system
Major refactor of graphics infrastructure:
- Introduce ICommandAllocator and D3D12CommandAllocator for explicit command buffer management.
- Change ICommandBuffer.Begin to require an allocator.
- Add IRenderTargetStrategy abstraction with swap chain and texture implementations.
- Update IRenderer to use RenderTargetStrategy instead of direct handle.
- Add DPI scaling support to swap chains (ScaleX/ScaleY, SetScale).
- RenderSystem now supports thread-safe swap chain resize requests.
- Remove persistent copy command buffer; use per-frame allocators.
- Make Logger public/static and clean up API visibility.
- Update .editorconfig and debug layer enablement.
These changes improve modularity, DPI-awareness, and future extensibility.
- Updated Misaki.HighPerformance package versions in Core and Graphics projects.
- Added IsTrimmable to Ghost.Engine.csproj for trimming support.
- Renamed GetOrRegisterComponent to GetOrRegisterComponentID and updated all usages.
- Component registration codegen now uses a static class with [ModuleInitializer], no longer requires [EngineEntry].
- Improved JSON serialization: added string support, introduced Utf8JsonObjectScope/ArrayScope, and new extension methods for cleaner JSON writing.
- Removed [SkipLocalsInit] from Hierarchy and LocalToWorld.
- Fixed Entity.Invalid to use INVALID_ID for both fields.
- Minor cleanup: clarified comments, reorganized Ghost.Generator in solution, and disabled component serialization generator.
Major ECS API overhaul: added ComponentSet, refactored ComponentRegistry, and updated all entity/component creation methods. Introduced robust custom serialization infrastructure and per-component source generators for registration and (de)serialization. Updated editor, engine, and test code to use new APIs. Improved code quality, naming, and performance throughout. Removed obsolete code and updated dependencies.
Introduce per-component versioning in chunks and world for efficient change detection.
- Add version arrays to chunks and global version to world.
- Update queries and ForEach to mark written components as changed.
- Extend QueryBuilder with WithAllRW/WithPresentRW for write access.
- Expose change tracking API in ChunkView.
- Improve thread safety and debug code.
- Update tests and examples to demonstrate new features.
Added implicate conversion to Result<T, E> and RefResult<T, E>;
Added new ScheduleChunkParallel in EntityQuery;
Remove Ghost.SparseEntity from solution file. It's now completlty replaced by Ghost.Entities;
Updated packages version;
Removed Ghost.SparseEntities;
Added new EntityQuery.EntityComponentIterator;
Added new thread local command buffer in World;
Changed commands in EntityCommandBuffer from UnsafeList<Command> to UnsafeList<byte> for better performance;
Changed the name of IJobEntityParallel to IJobEntity;
- Upgraded `Misaki.HighPerformance.LowLevel` to v1.2.8.
- Added `IEquatable` to `Handle<T>` and `Identifier<T>`.
- Improved `Result` extensions with `[CallerArgumentExpression]`.
- Introduced `SetEnabled` in `EntityManager` to toggle components.
- Refactored `Chunk` and `Archetype` for enableable components.
- Added `EntityQueryMask` for filtering enabled/disabled components.
- Enhanced `QueryBuilder` with new filtering methods (`WithAll`, etc.).
- Improved `EntityQuery.ForEach` with entity validation.
Refactored `ArcEntityTest` to use updated `Transform` and `Mesh` components, improving query logic with `GetChunkIterator` and introducing `ForEach` methods for better modularity.
Enhanced `Chunk` and `Archetype` structs with `readonly` properties and memory optimizations. Fixed bugs in memory copy logic and entity relocation.
Improved `EntityManager` with proper disposal handling, added a destructor, and fixed pointer usage in `AddComponent` and `SetComponentData`.
Refactored `EntityQuery` to use `ChunkIterator` and `ChunkView` for better abstraction. Simplified `EntityQueryMask` logic for performance.
Introduced templated `ForEach` methods and `ForEachWithEntity` methods, dynamically generated using T4 templates for scalability.
Added disposal logic for archetypes and queries in `World`. Updated `Program.cs` to include memory debugging setup.
Integrated T4 templates for dynamic code generation and added helper functions for template generation. Updated project file to include templates and generated outputs.
General improvements include enforcing immutability, optimizing memory management, and adding debugging/logging for better traceability.
