[ShaharMS]

Vision 3 backend and Screen API

Vision keeps its Haxe image API and uses platform backends where available. Screen draws to a browser canvas, a Windows window, or a portable in-memory surface. Bounds are writable and frame waits use the same API across targets. Browser WebGL attachments can perform whole-image inversion, simple grayscale, and black-and-white thresholding on the device. The portable Haxe implementation covers the other image operations.

• Export ScreenBackend and ImageBackend from vision.native and keep platform implementations in their respective packages.
• Keep the HashLink Windows extern beside its C bridge.
• Register Screen and WebGPU coverage in the existing utest runner.
• Remove the root scripts, documentation examples, and Cursor setup. CI builds the bridge and can package a release.
• Implement standalone Screen windows on Linux and macOS.

Windows Screen supports hxcpp and HashLink bytecode. Cross-target CI

Image algorithms and text

• E-means color clustering and posterization
• Harris corner detection as the public entry point, with raw scores available through Harris
• Overlapping-pattern synthesis and ConvChain texture sampling
• Planar panorama stitching for ordered, overlapping photos
• Scale2x pixel-art enlargement
• Browser text drawing with bidirectional shaping, OpenType font loading, and ligatures
• Portable Haxe text drawing for TrueType, CFF1 OpenType, and WOFF 1
• Portable WOFF2 decoding and reconstruction
• Font and Glyph API with individual format readers and HaxeDoc
• Portable CFF2 and full complex-script shaping

Portable image codecs

• Sequential and progressive JPEG decoding and sequential JPEG encoding.
• Still lossless WebP decoding and encoding, including alpha.
• Lossless animated WebP with frame composition, timing, and loop count.
• CMYK and YCCK JPEG decoding and Adobe CMYK encoding.
• OpenType horizontal pair kerning for portable text.
• Lossy WebP decoding and encoding, including lossy animation frames.
• AV1 and HEVC bitstream framing and entropy decoding.
• AV1 lossless block transforms and HEVC intra prediction and 4x4 transforms.
• HEIC and AVIF complete pixel decoding and encoding.