[jcoyne]

Problem

Under the default strict settings, every Rgb8/Rgba8 decode of a JP2 that declares sRGB over four codestream components with no cdef box failed:

unsupported: backend color space cannot be mapped to requested 8-bit pixel format

resolve_alpha_and_color_space saw the component count disagree with the declared colour space and fell back to guessing from the count: four components without alpha became CMYK. CMYK has no 8-bit RGB mapping. Kakadu 7.10.2 wrote files like this from RGBA TIFFs when -jp2_alpha wasn't given. Kakadu 8.4.1 writes the cdef box. We hit it in production on an IIIF image server.

Change

ISO/IEC 15444-1 I.5.3.6: without a channel definition, the components after those of the declared colour space are unspecified. They don't change what the colour space is.

• When a JP2 declares an enumerated or ICC colour space, has no cdef or palette, and has more components than that space (plus alpha), the colour space is kept. The extra components are recorded as Image::ignored_components.
• Decoded output omits them. This covers finish_decoded_image (bitmaps, borrowed and native planes, decode_into), the native RawBitmap decodes (16-bit output), streamed region rows (through a sink adapter), and the native planes sink. supports_direct_device_plane_reuse returns false for these images.
• So sRGB with 4 components decodes to Rgb8, and to Rgba8 with opaque alpha, at full resolution, in regions and in scaled regions.

Unchanged:

• Lenient decoding still infers alpha for exactly one extra component.
• Raw codestreams keep their component-count interpretation (4 components → CMYK).

Recode calls a hidden Image::keep_ignored_components and re-encodes every component, because it reuses the source JP2 metadata.

build_direct_color_region_plan (HTJ2K batch) now treats DirectPlanUnsupported as "not direct" and falls back to the generic decode. Before, it failed the job, as it already did for these images when they were labelled CMYK.

Tests

New file crates/j2k/tests/unspecified_extra_components.rs. The JP2s are built by hand, because the wrapper adds an RGBA cdef for four-component sRGB.

• sRGB with 4 components: whole image, region and sub-region Rgb8 match the first three input channels. Rgba8 has opaque alpha. Scaled regions at 1 and 2 levels work. Rgb16 matches the widened samples.
• Greyscale with 2 components decodes to Gray8.
• Native output for 4 and 5 sRGB components and 2 greyscale components: ignored_components, no alpha, and the expected plane count from decode_native_components, decode_native and the streamed native planes.
• Lenient mode still returns the 4th component as alpha.
• A raw 4-component codestream is still CMYK, with nothing ignored.

Checks run:

• cargo test -p j2k-native -p j2k passes.
• cargo xtask clippy, cargo xtask clippy-strict and cargo fmt --check are clean.
• The production file (640×480, Kakadu v7.10.2) failed at every level before the change and now decodes with decode_region_scaled_pow2_into at levels 0–2.

Notes

• J2kDecoder::info().components still reports the codestream count (4). Bounded row decoding therefore picks Rgba8 for these files and gets opaque alpha.
• Decoding still decodes the ignored components and then drops them. Skipping their tier-1 work would be a separate optimization.