[borisbat]

Behavior change: linear_to_SRGB now uses the correct linear-segment threshold for dark values. The glTF renderer keeps its existing output by default.

Why. HDR composition needs radiance before display encoding. Tone operators were duplicated across tutorials.

What changes.

• Add optional scene-linear PBR output and shared Reinhard, ACES-fit and PBR Neutral operators.
• Reuse the corrected sRGB encoder and document the transfer boundaries.
• Include operator attribution and its license in SDK installs and bundled tools.

Observable behavior.

• Linear PBR mode retains emissive values above one.
• Encoding linear 0.001 now returns 0.01292.

Where to look. color_mapping.das, gltf_pbr.das, and the CPU/shader/readback regressions.

Validation, claims, ledger

Validation

• Installed-SDK probes pass outside the source tree: color helpers run with dynamic modules disabled, and PBR defaults instantiate using bundled SDK modules.
• Full native test_aot build and sweep passed. The one full preflight stopped on added line lengths; after wrapping, targeted lint and every skipped runtime/utility lane passed. Post-review focused tests also pass; the Windows x86 precision adjustment is described below.
• All 27 repository REVIEW gates pass after adding the license to the SDK server bundle manifest; that consumer-specific gate failed before the entry and passed after.
• All 59 release-profile bundle checks passed. The permanent license check additionally passes with COLOR_MAPPING.LICENSE, fails when it is removed, and passes when restored; installed bytes match the source.
• All reference generators and Sphinx HTML passed with warnings treated as errors. Vulkan bindings were built locally for documentation; both changed Vulkan tutorial shader files also compile.
• Native OpenGL float-target readback covers default mode, switching back, above-white emission, both fog conversions, tint and alpha. Disabling linear output or either fog branch fails the regression.
• The sRGB regression fails on the old threshold. CPU controls distinguish the Neutral toe, chromatic shoulder, RGB lanes and negative-input policy.
• Windows x86 deliberately uses estimated pow. A native 32-bit probe reproduced the three CI precision failures; tests now use measured allowances of 2.5e-5 for their round-trip inputs and 1e-4 for their HDR decode vector on that platform only. Other targets retain their original bounds. The Windows 32-bit CI lane passes with the correction. All 14 negative controls still fail with the wider limits; no assertions were removed.
• GLSL suite: 125/125; SPIR-V suite: 430/430 with external validators. Generated fixtures use GLSL 330, GLSL ES 3.00, and a SPIR-V 1.3 fragment entry point with one runtime color input and four written outputs. The census checks its newly exercised OpUnreachable in both directions.
• All seven names newly listed by the color_mapping generator resolve in daslib/color_mapping.das; checked against those declarations and generated doc/source/stdlib/generated/color_mapping.rst using bin/daslang -documentation doc/reflections/das2rst.das.
• The new color-output-boundaries architecture section was checked against both cited operators (tone_reinhard, tone_aces_fit) and the named OpenGL/Vulkan tutorial consumers.
• Independent Sol audits and two external Sol review rounds completed; final external round found no actionable correctness defect.

Claims - stated, not tested

• Vulkan GPU execution was not run on this host. Shader compilation, SPIR-V generation and spirv-val do not establish device rendering or presentation behavior.

Not done

• Nightly CI is not requested.
• Background tint remains a multiplier in the selected output domain; linear mode does not promise the legacy display-space appearance. The documented behavior is covered by readback.
• Kept concrete RGB wrappers rather than adding generic lifting machinery. The existing example-local aces_tonemap remains unchanged because its negative-input policy differs.
• No new guarantees for extreme floating-point magnitudes beyond the reference operators.

[Copilot]

🔵 Needs a closer look

It changes a public stdlib function's output and adds a renderer output path affecting GLSL/SPIR-V/Vulkan shaders whose rendering correctness cannot be exercised in this automated environment, so it warrants human sign-off.

1 open issue

• Missing rationale for lone OpUnreachable census insert · New

What changed in this PR

This PR extracts the tone-mapping/sRGB helpers that were duplicated across the OpenGL and Vulkan tutorials into a new shared daslib/color_mapping module, adds an optional scene-linear output path to the glTF PBR renderer, and corrects the sRGB linear-segment threshold in linear_to_SRGB. It fits into the stdlib/shader-tutorial surface: the same operators now feed GLSL, SPIR-V and Vulkan consumers, with matching documentation, license attribution, and CPU/shader/readback regression tests.

Changes:

• New daslib/color_mapping module with tone_reinhard / tone_aces_fit / tone_pbr_neutral and color_srgb_encode/decode (+ per-channel forms); tutorials and gltf shaders now call the shared operators instead of local copies.
• Fixes linear_to_SRGB to use the correct 0.0031308 linear-segment threshold (was 0.00031308), with a dedicated transfer regression test.
• Adds GltfPbrRenderer.linearOutput (scene-linear radiance + fog decode) plus license/notice wiring (THIRD_PARTY_NOTICES, CMake install, bundle smoke test) and docs/architecture/tests.

File	Description
daslib/​color_mapping.das	New shared module: tone operators + sRGB transfer functions
daslib/​math_boost.das	Corrects linear_to_SRGB linear-segment threshold
daslib/​color_mapping.LICENSE	Apache-2.0 + ACES attribution for the new operators
daslib/​ARCHITECTURE_SHADER.md	New #color-output-boundaries section cited by the operators
modules/​dasGLTF/​gltf/​gltf_pbr.das	Adds u_linear_output uniform, linear output + fog decode, linearOutput field
modules/​dasGLTF/​tests/​test_render_targets.das	GL readback regression for linear/legacy/fog/tint
modules/​dasGLTF/​README.md	Documents linear output + shared operators
tutorials/​opengl/​{03_sdf,11_hdr,12_gltf}	Replace local tonemap/aces_fit with shared operators
modules/​dasVulkan/​tutorials/​{03_sdf,11_hdr}/​*_shaders.das	Same operator migration for Vulkan
tests/​math/​test_color_mapping.das, test_srgb_transfer.das	CPU reference-value and transfer coverage
tests/​glsl/​test_color_mapping.das, tests/​spirv/​{_color_mapping_shader,test_color_mapping,test_census}.das	GLSL/SPIR-V shader compilation + census coverage
tests/​aot/​CMakeLists.txt	Registers the new SPIR-V shader module for AOT
doc/​* , skills/​daslang/​references/​everything.md, THIRD_PARTY_NOTICES.md, CMakeLists.txt, ci/​smoke_test_bundle.sh, tests/​README.md	Docs digest, toctree, license install, and bundle/test bookkeeping

_{Review effort: Balanced}

[Copilot]

🔵 Needs a closer look

It spans a public-API behavior change (linear_to_SRGB), the glTF renderer, and GPU shader/SPIR-V/GLSL paths whose device-rendering and golden-fixture claims cannot be verified here, so it warrants human review.

3 open issues

• Remove redundant braces from single-statement early exit · New
• Replace placeholder color_mapping docs with full module description · New
• Missing rationale for lone OpUnreachable census insert

_{Review effort: Balanced}

[Copilot]

🔵 Needs a closer look

It changes a shared, widely-used color function's output and spans stdlib, renderer HDR behavior, cross-API shader emission, and license/install wiring, which warrants human sign-off despite the code appearing correct.

1 open issue

• Add module summary so docs digest shows a real description · New

3 resolved since last review

• Replace placeholder color_mapping docs with full module description
• Remove redundant braces from single-statement early exit
• Missing rationale for lone OpUnreachable census insert

_{Review effort: Balanced}

[Copilot]

🔵 Needs a closer look

It changes a shared public color function's behavior, adds a new public stdlib module and a glTF renderer output mode, and spans shader codegen across GLSL/SPIR-V/Vulkan whose device-rendering path the PR itself states was not executed, warranting human final sign-off.

No open issues 🎉

1 resolved since last review

• Add module summary so docs digest shows a real description

_{Review effort: Balanced}