Skip to content

A fast 2D and 3D game engine — TypeScript SDK, C++/WebAssembly core, visual editor. Ship one project to web, desktop, WeChat MiniGames, playable ads, and native Android / iOS.

docs(changelog): geometry that does not have to be re-uploaded every frame

esengine21affaea081c
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 88718f1c4..cf5ec7793 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -53,6 +53,26 @@ published separately; it ships inside the editor.
### Added
+- **Geometry can stay on the GPU.** Every triangle the engine drew had its
+ vertices written into the frame's transient pool by the CPU first — the draw
+ list resolved both the layout and the buffers from that pool, so there was no
+ way to draw geometry that simply lives on the device. That is right for a
+ sprite stream that changes every frame, and a per-frame re-upload of anything
+ that does not.
+
+ A draw can now name its own vertex buffer, index buffer and layout, with the
+ frame streaming only the per-object transforms. The arrangement is not new —
+ the particle path has always been a static quad in slot 0 with a per-instance
+ stream in slot 1 — so the same mesh drawn twice costs one more transform
+ rather than a second copy of its vertices. A mesh is buffers, layout and
+ bounds as one record, and it declares only its own channels: the transform is
+ appended by the engine, so nothing above has to know how one reaches a shader.
+
+ `mesh2d_makeResident` freezes a component's inline geometry onto the GPU —
+ the same vertices, uploaded once. Its gate is the mesh2d scene and the mesh2d
+ assertions, run again with the geometry frozen mid-run: equal pixels is the
+ whole claim, on both backends.
+
- **A game that loses its GPU comes back on its own.** Every piece of a recovery
was in place and nothing called it: the only caller outside the engine was a
test probe, so a shipped game that lost its device stayed on the placeholder