fix(core): read texture dimensions from the plain object image form

[mhayk]

Goal

Fixes #10371 — the documented plain object image form renders as opaque black, painting black over everything beneath the layer's bounds.

BitmapLayer's docs list {data: <Uint8Array>, width, height} as a supported image value (bitmap-layer.md), but in 9.x it does not render.

Root cause

createTexture() in modules/core/src/utils/texture.ts reads the dimensions off image.data:

const {width, height} = image.data;

Browser image objects (ImageData, ImageBitmap, canvas, …) hit the branch above that wraps them as image = {data: browserImage}, so image.data.width/height exist and everything works. The plain object form has constructor.name === 'Object', so it is not wrapped — it stays {data, width, height}, where data is the raw typed array. Both dimensions come back undefined, so:

device.getMipLevelCount(undefined, undefined)
// -> 1 + Math.floor(Math.log2(Math.max(undefined, undefined, 1)))  ->  NaN

The texture is created with mipLevels: NaN. Combined with the default mipmapFilter: 'linear' in DEFAULT_TEXTURE_PARAMETERS, the texture is incomplete and samples as (0, 0, 0, 1).

This runs through the shared image prop type in modules/core/src/lifecycle/prop-types.ts, so it affects every image-typed prop (BitmapLayer.image, IconLayer.iconAtlas, TextLayer, …), not just BitmapLayer.

Changes

modules/core/src/utils/texture.ts — read the dimensions from whichever level actually carries them:

const {width, height} = image.data.width === undefined ? image : image.data;

One-line behavioral change; the browser-object path is untouched.

Validation

New test/modules/core/utils/texture.spec.ts covering both forms, asserting width, height, and a valid mipLevels. Confirmed it actually catches the bug — with the fix reverted:

AssertionError: derives a valid mip level count: expected NaN to be 7

Also verified end-to-end by pixel readback, rendering a solid-yellow bitmap through both forms and sampling the canvas centre:

	plain object	equivalent ImageData
before	[0, 0, 0, 255] — black	[255, 255, 0, 255]
after	[255, 255, 0, 255]	[255, 255, 0, 255]

This matches the reporter's measurements. That probe was a throwaway and is not included — the committed test asserts the root cause (mipLevels) instead, since the repo keeps visual checks in test/render with golden images.

Suites run: core + layers headless — 84 files, 360 tests, all passing. biome check clean on all three files.

Not run locally: yarn test-render. The render project needs headed chromium, and the Playwright build this checkout pins (1217) currently 400s on every CDN mirror, so I could not install it. No existing golden image exercises the plain-object path — all bitmap-layer render cases load images by URL — so none should shift, but that is worth confirming in CI.

Also note the full headless suite is flaky on my machine independently of this change: repeated runs on unmodified master fail a varying 2–3 tests (LoadingWidget consistently, plus rotating WebGLAggregator/MapboxOverlay/react-mount failures). None are in core or layers.