Updated the organization of engine libraries to group features by architectural role rather than by individual feature. Enhanced documentation in CLAUDE.md and architecture.md to reflect these changes. Removed obsolete project references and added new modules for content management and simulation. Adjusted project files for MrGameEng.Graphics and MrGameEng.UI.Tests to align with the new structure.
This refactor aims to streamline the development process and improve clarity in the engine's architecture.
Add MrGameEng.AI utility-AI module; format codebase with CSharpier
New MrGameEng.AI module (ResponseCurve, Consideration, UtilityAction,
UtilityAi selector, Blackboard) plus CSharpier formatting applied across
the whole engine. Documents the CSharpier convention in CLAUDE.md.
@
Collisions: init bucket heads to -1 (QueryAabb hung before the first
rebuild), reset query stamps on truncated QueryAabb (later queries
silently dropped entities), inside-origin raycasts hit at fraction 0 for
circles too, exactly-touching boxes now pair like touching circles.
Graphics: render into the letterbox viewport so the picture matches
ScreenToWorld/WorldToScreen instead of stretching; Y-sort by the
transform pivot rather than the quad center; lock-free snapshot
LayerRegistry (parallel submit read it unsynchronized); validate
InitialCapacity; warn when UseRenderer2D drops options of a later scene.
Core: scenes are explicitly single-use (re-loading threw silently
duplicated systems/entities before — now it throws), Scene.RegisterUnload
for per-scene resources, a switch requested during the reveal phase
covers again instead of hard-swapping, borderless fullscreen
(HardwareModeSwitch off), InputCapture service for input-suppressing
overlays, host disposes the transition renderer and IDisposable services
on shutdown.
Input: game input reads as released while InputCapture is held; mouse
position and wheel freeze so deltas stay zero.
DevConsole: holds InputCapture while open (typing no longer drives the
camera), Revision increments only under the lock, quoted command
arguments, history capped at 256.
UI: scene Desktop skips Myra input processing while the console is open
(clicks no longer fall through), is disposed on scene unload, and Myra
init no longer depends on a process-static flag.
Audio: validate channel count/sample rate before stopping the previous
track, empty looped oggs no longer hang FillBuffers, the instance stops
when a non-looping track drains (IsPlaying was stuck true).
Atlases: metadata v2 stores per-source size+mtime snapshots, so
timestamp-preserving copies and renames invalidate correctly; loader
checks the version and disposes pages on partial load failure; shared
pages never exceed a non-POT MaxPageSize; oversized items pack first
onto exact-size pages instead of splitting an open shared page; the CLI
validates numeric options.
Assets.Generator: file names are escaped in XML docs and string
literals, members no longer collide with the enclosing class (CS0542),
and the Assets root is resolved against build_property.projectdir so
nested "Assets" directories do not shift region paths.
Pathfinding: queries throw when the grid was resized after construction;
generation stamps survive int overflow.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Five optimizations measured on the 100k-entity stress scene (Release,
vsync off): 103 FPS baseline -> 297 FPS.
- Sprite submission and vertex building run on all cores above
Renderer2DOptions.ParallelThreshold (default 8192). Work is sliced
into 4096-entity segments: a Friflo chunk holds a whole archetype,
so per-chunk parallelism degenerates to one thread. Segments merge
in deterministic order, preserving radix sort stability.
- Vertex buffer is ring-written with SetDataOptions.NoOverwrite
(GPU buffer 2x frame size); Discard only on wrap-around.
- Texture2DRegion precomputes UVs - four float divisions per sprite
per frame removed.
- Renderer2D exposes per-phase timings (submit/sort/build/upload/draw),
shown in the sample HUD - all further optimization is data-driven.
- Sample BounceSystem parallelized the same segmented way.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
SpriteBatcher now sorts with a stable LSD radix sort: equal-key sprites
keep submission order across frames (no flicker) and passes over digits
identical in all keys are skipped, making the common single-layer case
nearly free. Hot paths avoid per-sprite trig and square roots: SinCos is
skipped for unrotated sprites and the culling radius comes from the
region's precomputed diagonal.
Stress scene (100k entities, ~61k on screen, Release): 103 -> 124 FPS.
Sample now runs with VSync off to show real frame rates.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>