Use C# arrays as pluggable storage for managed types
Opened this issue · 0 comments
When pluggable storages are added to flecs, pinned C# arrays should be used as storage for managed types.
Performance - Special unboxing code is required per every iteration in the current design. Using plain C# arrays will remove this overhead and bring the performance of managed component iteration on par with native C# ECS frameworks/libraries.
Memory Allocations - This will greatly reduce memory consumption due to only needing a single GCHandle
and Box<T>
object to be allocated per column instead of per component.
Debugging - With the current setup, it is not possible to inspect the contents of a column for both managed and unmanaged types inside a debugger due to the GCHandle
indirection and unboxing step required for managed objects. Switching to C# array storage will replace the Column<T>
type with the Span<T>
type and allow the contents of components to be displayed in the debugger.
Pointer APIs - Pointer-based APIs in Flecs.NET are currently restricted to unmanaged types only. This restriction can be removed entirely as using pinned C# arrays will allow us to retrieve pointers to managed component references. This is a prerequisite to being able to properly implement serialization and deserialization for managed types.
// Pointer-based iteration
Query.Iter((Iter it, Position* p, Velocity* v) => { });
// Span-based iteration
Query.Iter((Iter it, Span<Position> p, Span<Velocity> v) => { });
// Get raw pointers to managed objects.
Position* p = e.GetPtr<Position>();