@motionscript/renderer 0.1.0 → 0.1.1-alpha.0

package/CHANGELOG.md

@@ -1,5 +1,18 @@
 # @motionscript/renderer
 
+## 0.1.1-alpha.0
+
+### Patch Changes
+
+- 6b89575: The renderer's prewarm now also builds the drawing pipelines for the canvas format, so their shader compile overlaps a host's start-up instead of delaying its first frame.
+- e279864: A 3D scene's surface and image textures stay their own across frames. A frame's image ids are given out in the order they are met, so when a scene's set of textures changed size (a token appearing, a plane entering) the scene's id moved on and its old id became a surface, while the renderer kept the old scene and drew it into that surface: a texture showed a picture of the whole scene, and sampling the target being drawn raised a WebGPU validation error. The renderer now drops a scene whose id becomes something else, and binds every image and surface a scene uses to the current frame's id on every sync.
+- e9db554: An asset is downloaded and decoded once while it is loading, however many ask for it: the weights of one font family share its files (each weight used to fetch the whole family), two draws of one image or video share one decode, and every session, still and export on a page shares one request per source (`loadAssetBytes`, which the audio mix now goes through as well). A load that fails is asked for again next time.
+- 7cd4e53: A Text with a Write On reveal (`start`/`end` short of the whole string) or styled `runs` drew nothing at all: the renderer ignored the text's pieces while core left their painting to them. Each piece now draws with its own fill and stroke, the unrevealed characters held in place but clear, so a Write On types the text on character by character.
+- Updated dependencies [3630b74]
+- Updated dependencies [1c43a37]
+- Updated dependencies [e279864]
+  - @motionscript/core@0.1.1-alpha.0
+
 ## 0.1.0
 
 ### Patch Changes

package/README.md

@@ -71,7 +71,10 @@ defaults to beside the module.
 - `WgpuStorageAdapter` decodes images to `ImageBitmap`s, video frames with
   mediabunny, and hands font bytes to ms-text. Bytes come through
   `loadBytes` (a plain `fetch` by default), which a platform points at its own
-  asset store.
+  asset store. A load in flight is joined, never repeated: per adapter by asset
+  (a font by family, since every weight registers the family's files) and
+  page-wide by source, so the sessions, stills and exports asking for one file
+  while it arrives share one download (`sharedLoader`).
 - `WgpuSurfaceContext` is the adapter's render context mounted on a canvas,
   with the preview knobs a host sets (`pixelRatio`, `view`, `frame`,
   `clipToFrame`, `frameBackground`). Mounting is asynchronous (WebGPU device