Upstream removed bake/maps.rs, engine/tests/auto_rebake.rs and read_skew
(moved elsewhere); take the deletions. Convert two new chunks_exact uses in
bake/test/setting.rs.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Clippy 1.99 flags chunks_exact and chunks_exact_mut with a constant
chunk size (chunks_exact_to_as_chunks). Switch those call sites to
as_chunks / as_chunks_mut. Behaviour is unchanged: the trailing partial
chunk was already dropped.
Fixes#233
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
`ball 1.obj` and `ball_1.obj` sanitized to the same companion stem, so
exporting both into one folder overwrote each other's .mtl and textures. A
name the sanitizer changes now gets an FNV-1a hash of the original appended;
names that need no change keep their companions as before.
A UV tile with a lower resolution than its atlas block fills the block's
corner, so it has fewer texels per UV unit. Neighbourhood filters (blur,
sharpen, warps, bevel, glow, ...) and height-to-normal slopes used the block's
texel density for every tile, so such a tile blurred twice as wide and showed
twice as steep slopes in UV space.
TextureSet::block_scales gives each block's share; the compositor filters each
block at its tile's radius, texture export and the CPU path tracer scale the
slope by it, and the display packs it into the Extra texture's alpha for the
viewport and GPU path tracer shaders (parallax too).
Found while testing crates/formats/src/mesh_import/obj.rs:77 (model.name passed to raw.polygon); stem fallback in mesh_import/mod.rs `read`
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Found while testing crates/formats/src/mesh_export.rs:24
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Found while testing crates/formats/src/mesh_import/ply.rs:164 (nverts is set at line 140)
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Found while testing crates/formats/src/export/mod.rs:844 (name built at 476-478)
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Found while testing crates/formats/src/project_file/mod.rs:146-147 (also 180-181)
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Keeps both sides of the translation tables, AGENTS rules and the formats module list;
material files in the browser load before the ask-to-save step (they add a layer, not a
new project); Arabic for the eight MaterialCraft strings; menu parity regenerated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
When file.exportTextures is given a colorSpace, colour maps carry it: an ICC profile in PNG,
JPEG, TIFF and WebP, the chromaticities attribute in EXR. TGA, BMP and HDR have no place for one
and warn. Data maps and exports without a chosen space are unchanged.
The profile is a small version 2.1 matrix/curve profile written from the space's primaries
(Bradford-adapted to D50), tested against the published sRGB colorants.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The length limit cuts the longer placeholder at a different place than the tile number,
so a long template got a <UDIM> name that matched none of the files written. A name is
now kept only when putting the tile number back gives the file's own name.
- SkewMap::paint refuses a stroke whose dabs would visit more than 2^26 texels, before any
texel changes; the dab count alone didn't bound the work of a big brush on a big map.
- bake.skew.set checks the image size from its header (decode_image_within) before decoding.
- The 2D view is shown for the skew brush in bake mode only, not in rendering mode.
- Prepared::correct allows its argument count for clippy; the skew test uses its own folder.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
- MaterialX export: a tiled file's <UDIM> name comes from the file plan, so a set called
cloth.1001.body keeps its name; the DirectX normal warning is raised only when a material has a
normal map.
- colorSpace that isn't a string is an error instead of the format's default.
- MaterialX import: an image that isn't a regular file is skipped with a warning (a pipe in the
folder would have blocked the read).
- Auto-rebake: remembers which inputs each map image was baked from, so undoing a rebake brings
back a map that is out of date and rebakes it; not scheduled automatically on the web build,
where a bake would run on the page's thread.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
- core: Fill::link (MaterialLink: package resource, graph, exposed inputs, presets,
image inputs, size); BakedMap::interop_id / for_input shared mesh-map identifiers
- formats: own .mcmat reader (no MaterialCraft dependency) and 16-bit mesh-map PNGs
- engine: material.load takes .mcmat and .mcraft; material.link.set / .preset / .cook /
.saveSource; LinkJob re-cooks through materialcraft-cli at the texture set's
resolution with baked mesh maps as graph inputs; fresh resource names per cook keep
undo and the decoded-image cache consistent; packages count as used resources
- ui: linked-material section in fill properties (presets, sliders, colors, toggles,
options) cooking on a worker thread, one process per layer, stale results redone;
Open in MaterialCraft; File > Import material...; .mcmat drops; library shelf lists
.mcmat and rescans on focus; 8 new strings in every language
- tests: materialcraft_link (import, params, undo, duplicates, mesh maps, save, and a
real materialcraft-cli round trip when MATERIALCRAFT_CLI is set)
- docs: README, features, control protocol, architecture, AGENTS, ROADMAP, parity
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
uvtiling is clamped to 0.001..=1000, the range layer.fill.projection accepts. Image files are capped by resource_cmds::MAX_BYTES instead of a second copy of the number.
The .mtlx tags colour textures with the export's colour space: srgb_texture, lin_rec709, acescg, or lin_ap0_scene for ACES2065-1 (MaterialX 1.39 has no AP0 space; the export warns). A colour map packed into one channel (L) is not converted and now says so. The $colorSpace variable's description lists every value.
Float files (EXR, HDR) hold linear values, so their colour maps are tagged lin_rec709 instead of srgb_texture. The .mtlx goes through formats::write_atomic, and a set whose name sanitises to an earlier set's file name is skipped with a warning instead of overwriting it.