Skip to content

engine_analysis.md

Latest commit

 

History

History
87 lines (76 loc) · 4.7 KB

engine_analysis.md

File metadata and controls

87 lines (76 loc) · 4.7 KB

Epoch Engine Analysis

Snapshot

Epoch is now documented as a module-first engine with the active runtime living under Engine/modules/ and Engine/src/, while older compatibility/archive surfaces have been retired and mapped into active replacements.

Current public version: v0.82.51

Architecture highlights

  • Multi-context runtime: the shared context layer and multiplexer coordinate backend-owned windows, command queues, and render-thread work.
  • Backend stack: OpenGL, SDL, Raylib, SFML, software, and noop/headless are all represented in the active engine tree; Vulkan remains experimental.
  • Atlas-driven UI/rendering: atlas upload, sprite submission, and GUI layout are shared concerns across the active render paths.
  • Launcher/editor split: project and game entry now live in the launcher, while the editor uses a more traditional desktop-style menu flow.
  • Task graph + scripting: reload and background work are funneled through task scheduling rather than ad hoc threaded entry points, and editor-triggered compiled scripts now run through an explicit host API instead of a loose filewatch-first loop.
  • Migration map: formerly archived compatibility surfaces are now either preserved in active modules or called out explicitly in Engine/docs/legacy_feature_map.md.

Current strengths

  • Strong module-first organization for the active engine.
  • Broad backend coverage for desktop experimentation and tooling.
  • Good separation between active code and archived compatibility material after moving the legacy tree under Engine/.
  • Startup and shutdown behavior are getting more disciplined as backend-local diagnostics and hot-path logging are trimmed back out of the render loop.
  • SDL and Software now share the same editor preview-grid geometry/palette path as the GPU editor previews, which tightens backend parity for scene-view smoke tests.
  • Parent-window shutdown now behaves more like a real engine host lifecycle: docked children are marked for close and the session exits instead of leaving a dead console/process behind.
  • The parented docking layout now stays intact during shutdown instead of undocking backend panes as the host window closes.
  • Live pane management is back in a better place too: undocked windows keep the information they need to redock cleanly, and only still-docked panes are considered part of the parent grid layout.
  • Manual pane management is now usable again without modifier keys: a dedicated drag strip provides left-drag docking while leaving the rest of the pane free for normal backend/editor interaction.
  • Parent-close ownership now behaves like a real multi-window host lifecycle: undocked promoted backend windows can outlive the parent host, but once they are redocked they return to normal parent-owned shutdown behavior.
  • The last surviving undocked pane now shuts the process down cleanly when it closes, even if that backend window is owned by a render thread.
  • The editor shell now maps more cleanly to production-tool expectations: launcher responsibilities are separated from editor responsibilities, and the updater is confirmation-gated before it can run from the UI.
  • The source-update path is back to behaving like an engine updater instead of a passive download tool: it now restores manifest dependencies with vcpkg, rebuilds from the downloaded snapshot, and replaces the running runtime from the rebuilt output.
  • Backend presentation is more visually coherent now that the active renderer base colors are being pulled toward the same darker Vulkan-style baseline, and SFML’s shared preview path is clipped back to the intended scene view.

Current cautions

  • Vulkan is present but should still be treated as a migration/integration path, not the default renderer, even though the editor palette, editor grid preview, and GUI presentation have been brought closer to the OpenGL baseline.
  • A few minor archival and compatibility surfaces still exist and should be changed carefully.
  • Backend fixes are usually safest when applied locally to the affected backend instead of globally in the multiplexer.
  • Raylib parent/docking work should continue to respect GLFW/raylib ownership of the native GL context instead of swapping in fresh Win32 DC handles after initialization.

Recommended priorities

  1. Keep renderer/backend smoke coverage improving, especially around startup, resize, docking, and shutdown behavior.
  2. Continue reducing duplicated compatibility naming in public surfaces while preserving compile compatibility where needed.
  3. Promote only active, tested functionality into the main engine surface and keep migration notes honest when a retired system still needs a modern home.