The test draws the UI, and the i18n tests switch the process-wide language to ones the installed fonts cannot draw, which made the Windows CI panic on a Japanese glyph. It now takes TEST_LANGUAGE like the other tests that draw the UI.
Reject orthographic paint points on or behind the eye plane, with regression coverage preserving near-eye and nonfinite fallbacks.
Remove three UI anchor-picker tests. Integrate main's no-op tile-free sharing optimization, copy-on-write regression tests, clearer benchmark fixtures and architecture documentation, and adapter-tolerant sheen and refraction assertions.
Recognize split emissive-color names, route unconsumed dropped files through their existing import path, and pin the dialog's scanned map choices when importing. Add focused regressions for all three behaviors.
Configure AO, thickness, and bent-normal rays before the timed project bakes. Add a regression that checks the settings copied into each bake task and correct the benchmark description.
Carry the reviewed paint visibility repair onto the prior target tree without conflict. Preserve the existing orthographic and grazing visibility fixes.
Preserve the saved image-menu, project-switch cache, and test repairs before integrating main. Remove the disabled-mask anchor UI test at the maintainer request.
Integrate main and adapt the performance suite to current bake settings,
map selection and undo APIs.
Preserve full triplanar image weighting for floating-point resources and
add a regression for lost HDR contributions. Use hardware floor when
SSE4.1 is enabled, retaining the portable fallback.
Split Report and Row into their own files and complete scoped naming,
spacing and comment cleanup.
Validated with 132 release tests, workspace Clippy, wasm and layering
checks, formatting, and retained targeted noise and performance probes.
Use the documented RAM fallback for zero readings. Preserve dotted folder names, recognize split normal aliases, honor later map names, and clear import dialogs after project replacement. Count same-map drop candidates and clean test fixtures on unwind.
Preserve the benchmark at its target-branch location. Improve scoped names, spacing and comments, clarify same_tiles semantics, and add regression coverage proving image clones do not duplicate tile handles.
Check unshifted depth before accepting a grazing allowance and start exact rays just off the surface. Keep the wider depth tolerance in the uncertain gate. Add regressions for both reported cases.
Recover an earlier forced sample from the same contact when the latest touch event is an unforced Move. Stop recovery at a completed contact and add press-frame and contact-boundary regressions.
Preserve signed RGBA16F normal uploads while integrating current main.
Reject invalid baked samples before blending, skip sampling absent baked
normal maps, and cover two-set normal shading with a headless test.
Complete scoped spacing, comment and shader constant cleanup while
preserving shader values.
Preserve main UI and translation changes. Bound archive indexing, directory scans and native reads; fix resolution-token overflow, unique-name replies and dialog behavior.
Avoid generated-map re-decoding and input-buffer clones, release packed sources after their final use, and borrow dialog candidates. Keep generated pixels identical to PNG reloads. Organize new types and preserve the public classifier module.
Validated with 973 release tests, focused UI and allocation probes, panic hunt, clippy, layering, wasm, formatting and diff checks.
Keep file-picker answers tied to their original Project Configuration opening, including after UI layout resets.
Split ProjectConfig and TempDir into dedicated files, complete scoped naming/comment/spacing cleanup, and add delayed-answer regressions and a Project Configuration snapshot.
Validated with the UI release suite, workspace Clippy, wasm checks, formatting, and diff checks.
Recover contact pressure for same-frame releases and clear stale pressure for pressure-free taps. Update the existing cached value instead of allocating for each pressure sample.
Clean up pressure tests, simplify direct input fixtures, and include main's shared-library guidance.
Validated with the UI release suite, workspace Clippy, wasm checks, formatting, and allocation/lifetime probes.
Use parallel view rays for orthographic strokes and share depth-neighbourhood queries. Add direct and full-stroke regressions, correct the silhouette benchmark fixture, and apply scoped naming, spacing and comment cleanup.
Validated with release paint tests, a focused engine regression, workspace Clippy, formatting, layering and full WebAssembly checks.