Description

Skwasm's Canvas.drawAtlas previously hardcoded SamplingOptionsForQuality(FilterQuality::medium) regardless of Paint.filterQuality, so pixel-art tilemaps (e.g. flame_tiled atlases at FilterQuality.none) rendered blurry on web/skwasm while every other backend correctly honored the paint.

Use case / why this matters

Canvas.drawAtlas exists to batch many sprite quads sourced from a single atlas image into ~one GPU draw call (one vertex buffer, one texture bind, one shader pipeline). For a Flame + Tiled pixel-art game, a typical 64x64 tilemap forwards ~4 096 sprites/frame: rendering them via drawImageRect would mean 4 096 draw calls + state changes/frame, while drawAtlas collapses that to roughly one. That's the path flame_tiled takes for tilemap rendering, and it's the canonical reason an app reaches for this API.

Paint.filterQuality is a no-op for most callers (the default linear/mipmap sampling is what they want), but pixel art is the explicit exception: low-resolution art scaled by non-integer camera factors needs FilterQuality.none (nearest-neighbor) to keep hard pixel edges. Native iOS/Android already honor this; on web/skwasm the hardcoded medium was overriding it and softening every tile.

Implementation

The native iOS/Android path was already correct: lib/ui/painting/canvas.cc:Canvas::drawAtlas translates paint.filterQuality -> DlImageSampling via ImageFilter::SamplingFromIndex and threads it through DisplayListBuilder::DrawAtlas (which carries DlImageSampling in DrawAtlasBaseOp). Both display_list/skia/dl_sk_canvas.cc and Impeller's dl_dispatcher.cc honor it. Skwasm was the only engine that ignored the paint.

This PR threads paint.filterQuality through skwasm's existing FFI in the same shape that canvas_drawImage and canvas_drawImageRect already use:

• skwasm/canvas.cc:canvas_drawAtlas now takes Skwasm::FilterQuality quality and passes SamplingOptionsForQuality(quality) to DrawAtlas instead of the hardcoded medium.
• skwasm_impl/raw/raw_canvas.dart:canvasDrawAtlas FFI gains an int filterQuality slot in the trailing position, matching canvasDrawImage/canvasDrawImageRect.
• skwasm_impl/canvas.dart:drawAtlas and drawRawAtlas extract paint.filterQuality.index and pass it through.
• skwasm/helpers.h adds static_asserts pinning Skwasm::FilterQuality's integer values to dart:ui's FilterQuality.index ordering, so any future drift in either enum becomes a compile error rather than a silently wrong sampling mode.

Two golden tests in lib/web_ui/test/ui/draw_atlas_golden_test.dart lock the behavior in:

• A matrix test over all four FilterQuality values, drawing a 2x2 checker at a non-integer scale chosen so bilinear sampling emits one mathematically perfect 50/50 gray pixel that nearest-neighbour cannot produce — a strong discriminator between the modes.
• A null-colors regression test (see below) covering the second fix in this PR.

A second commit on this branch addresses a pre-existing null-colors deref in canvas_drawAtlas that was flagged in review (the Dart wrapper passes nullptr when the caller's colors list is null; the C++ side was looping colors[i] regardless and was only saved from crashing by WASM reads at address 0 returning zeros). Kept in this PR rather than split out — it sits in the same function as the filterQuality fix and the C++ change is small enough that bundling stays reviewable.

CanvasKit

CanvasKit is now fixed the same way (the earlier "out of scope" reasoning was wrong: CanvasKit's Canvas.drawAtlas JS binding has accepted an optional sampling argument since CanvasKit 0.25.1 (2021), dispatching internally to _drawAtlasCubic/_drawAtlasOptions — no Skia CL needed).

• canvaskit/canvas.dart:_drawAtlas now passes toSkFilterOptions(paint.filterQuality) to SkCanvas.drawAtlas, the same FilterQuality -> sampling mapping drawImage/drawImageRect already use.
• canvaskit_api.dart threads the CkFilterOptions sampling argument through the drawAtlas binding.

All four FilterQuality values now resolve to identical SkSamplingOptions on both renderers (including high -> Mitchell B=C=1/3), so the drawAtlas goldens converge across CanvasKit and skwasm instead of diverging per renderer.

Refs

• Avoid the deprecated SkFilterQuality in the Engine APIs flutter-team-archive/engine#24797 — prior decision to keep Paint.filterQuality as the sampling source for Canvas methods rather than adding explicit per-call sampling parameters
• image edges are always sharp when using Canvas.drawAtlas #152312 — related family of drawAtlas filtering bugs

Pre-launch Checklist

• I read the Contributor Guide and followed the process outlined there for submitting PRs.
• I read the Tree Hygiene wiki page, which explains my responsibilities.
• I read and followed the Flutter Style Guide and the C++, Objective-C, Java style guides.
• I listed at least one issue that this PR fixes in the description above.
• I added new tests to check the change I am making.
• All existing and new tests are passing.