v1.15.7.0

Release 1.15.7.0 (7 Sep 2026) -- compared to 1.15.6.0
---------------------------------------------------------
  - *ShadingSystem*: Query ranked and aggregate post-optimization ShaderGroup compile stats (active_layers, network_depth, texture_ops, noise_ops) through new system `getattribute` keys. [#2147](#2147) (1.15.7.0, 1.14.12.0) (by @lgritz / Larry Gritz)
  - *python*: The OSLQuery Python bindings no longer require the OpenImageIO Python types: a new `OSLQuery.Parameter.type_name` property returns the parameter's type as a string, and the OpenImageIO module is no longer force-imported at startup. This lets Python programs use OSLQuery alongside an OpenImageIO built against a different binding library (pybind11 vs nanobind). [#2142](#2142) (by @lgritz / Larry Gritz)
  - *testrender*: Fix discontinuities in principal directions by switching to a more robust method of building an orthonormal basis from a normal. [#2154](#2154) (by @AlexeySmolenchuk / Alexey Smolenchuk)
  - *fix*: Repair unsafe string formatting of certain errors (a plain error message that happened to contain brace characters could confuse the formatting engine). [#2145](#2145) (1.15.7.0, 1.14.12.0) (by @lgritz / Larry Gritz)
  - *build*: Support LLVM 23. [#2159](#2159) (1.15.7.0, 1.14.12.0) (by @lgritz / Larry Gritz)
  - *build*: Correctly serialize the CUDA build of testrender. [#2157](#2157) (by @jfpanisset / Jean-Francois Panisset) (1.15.7.0, 1.14.12.0)
  - *ci*: Update all GitHub Actions to their latest versions. [#2144](#2144) (by @lgritz / Larry Gritz)
  - *ci*: Fix broken CI caused by a recent OIIO change and a libtiff vs webp conflict. [#2160](#2160) (1.15.7.0, 1.14.12.0) (by @lgritz / Larry Gritz)

v1.14.12.0

Release 1.14.12.0 (7 Sep 2026) -- compared to 1.14.11.0
---------------------------------------------------------
  - *ShadingSystem*: Query ranked and aggregate post-optimization ShaderGroup compile stats (active_layers, network_depth, texture_ops, noise_ops) through new system `getattribute` keys. [#2147](#2147) (@lgritz / Larry Gritz)
  - *jit*: Fix a crash in exception handling after OSL JIT on Ubuntu 24.04 (a libgcc 14.2 bug -- could affect other systems using that version). [#2113](#2113) (by @sergeyvfx / Sergey Sharybin)
  - *RendererServices*: Repair unsafe string formatting of certain errors. [#2145](#2145) (@lgritz / Larry Gritz)
  - *build*: Support LLVM 23. [#2159](#2159) (@lgritz / Larry Gritz)
  - *build*: Correctly serialize the CUDA build of testrender. [#2157](#2157) (by @jfpanisset / Jean-Francois Panisset) (@lgritz / Larry Gritz)
  - *ci*: Various fixes for CI breakage [#2143](#2143) [#2160](#2160), only test against OIIO 3.1 or lower, and backport container fixes for the old ASWF images. (@lgritz / Larry Gritz)

v1.15.6.0

Release 1.15.6.0 (1 Aug 2026) -- compared to 1.15.5.0
---------------------------------------------------------
  - *osltoy*: Improved readability of the cursor-highlighted line in the code editor. [#2067](#2067) (by @silvialpz / Silvia Lopez)
  - *build*: Fix compilation and runtime errors with LLVM 22. [#2124](#2124) (by @sergeyvfx / Sergey Sharybin)
  - *build*: Check for the bad combination of Cuda >= 13.2 with LLVM < 22.1.2, which fails to build. [#2133](#2133) (by @lgritz / Larry Gritz)
  - *build*: ci: turn unity builds off when building OIIO for CI, to unbreak he Windows CI jobs. [#2143](#2143) (by @lgritz / Larry Gritz)
  - *admin*: Clarify security policy on what is not considered a vulnerability. [#2128](#2128) (by @lgritz / Larry Gritz)
  - *admin*: Fix some minor discrepancies in project admin docs. [#2127](#2127) (by @lgritz / Larry Gritz)

v1.15.5.0

Release 1.15.5.0 (1 Jun 2026) -- compared to 1.15.4.0
---------------------------------------------------------
  - *ShadingSystem*: Extend per-ShaderGroup complexity statistics: compile-time metrics are now stored as persistent ShaderGroup fields, exposed via `getattribute(group, "stat:compiled_active_layers")` etc., and `getstats()` emits a new "Shader compilation stats, post-optimized" section showing min/max/median and a ranked leaderboard for each metric (controlled by the existing "stat:rank_groups" option). Also adds `--print-group-stats` flag to `testshade`. [#2120](#2120) (1.15.5.0, 1.14.11.0)
  - *fix*: When a shader parameter is both interactive and connected, the connection now correctly takes precedence (previously the interactive trait incorrectly overrode it, causing writes into a read-only buffer). Fixed in both single-point and batch modes. [#2121](#2121) (by Chris Hellmuth) (1.15.5.0, 1.14.11.0)
  - *ci*: Various CI runner fixes: remove deprecated Windows-2025 runner (routes to Windows-2025+VS2026 which we already test) [#2114](#2114), lock down 2023 container to prevent OCIO build breaks [#2117](#2117). (1.15.5.0, 1.14.11.0)
  - *admin*: Add AI coding assistant configuration (CLAUDE.md, Cursor, Copilot, Codex, OpenCode support). [#2122](#2122) (1.15.5.0, 1.14.11.0)

v1.14.11.0

Release 1.14.11.0 (1 Jun 2026) -- compared to 1.14.10.0
---------------------------------------------------------
  - *ShadingSystem*: Extend per-ShaderGroup complexity statistics: compile-time metrics are now stored as persistent ShaderGroup fields, exposed via `getattribute(group, "stat:compiled_active_layers")` etc., and `getstats()` emits a new "Shader compilation stats, post-optimized" section showing min/max/median and a ranked leaderboard for each metric (controlled by the existing "stat:rank_groups" option). Also adds `--print-group-stats` flag to `testshade`. [#2120](#2120)
  - *fix*: When a shader parameter is both interactive and connected, the connection now correctly takes precedence (previously the interactive trait incorrectly overrode it, causing writes into a read-only buffer). Fixed in both single-point and batch modes. [#2121](#2121) (by Chris Hellmuth)
  - *ci*: Various fixes for unexpected CI runner breakages: fmtlib master→main rename [#2103](#2103), wrong Windows-2025/MSVS2022 runner name [#2108](#2108), lock down 2023 container for OCIO stability [#2117](#2117).
  - *ci*: Remove deprecated Windows-2025 runner (now routes to Windows-2025+VS2026, which we already test). [#2114](#2114)
  - *ci*: Optimize OpenEXR CI build time; fix nightly-on-forks suppression for Windows jobs. [#2097](#2097)
  - *docs*: Add AI contribution policy document; update README, CONTRIBUTING, and pull request template with AI policy references; add AI coding assistant configuration. [#2099](#2099) [#2122](#2122)

v1.15.4.0

Release 1.15.4.0 (May 5, 2026) -- compared to 1.15.3.0
---------------------------------------------------------
  - *fix*: Fix crash in exception handling after OSL JIT on Ubuntu 24.04, caused by a libgcc 14.2 bug; workaround by disabling EH frames registration in the JIT MemoryManager [#2113](#2113) (by Sergey Sharybin)
  - *build*: Port many recent build improvements from OpenImageIO: better dependency reporting, improved ccache detection, TIME_COMMANDS fix, generate compile_commands.json [#2106](#2106)
  - *ci*: Port CI improvements from OIIO: shared steps workflow, upgraded GitHub Actions versions, restored pybind11 bleeding edge test [#2107](#2107)
  - *ci*: Various CI runner breakage fixes: wrong MSVS 2022 runner name [#2108](#2108), fmtlib renamed from master to main [#2103](#2103), skip OpenEXR utility builds and stop running nightly CI on forks [#2097](#2097)
  - *docs*: Fix outdated documentation references (point to ReadTheDocs, remove stale PDF references and old LaTeX build scaffolding) [#2104](#2104)
  - *docs*: Add AI contribution policy document; minor improvements to README, CONTRIBUTING, and PR template [#2099](#2099)
  - *admin*: Rename top-level `doc/` directory to `docs/` to match OpenImageIO and most other ASWF projects [#2105](#2105)
  - *admin*: Update security reporting instructions to prefer GitHub security advisories over email [#2112](#2112)

v1.15.3.0

Release 1.15.3.0 (1 Apr 2026) -- compared to 1.15.2.0
---------------------------------------------------------
  - *bsdl*: Incorporate diffuse MaterialX BSDFs into BSDL library [#2092](#2092) (by Alejandro Conty)
  - *build*: Fix compilation with ld.lld [#2094](#2094) (by Sv. Lockal) (1.15.3.0, 1.14.10.0)
  - *ci*: Add Windows 2025 with MSVS 2026 job variant to CI [#2093](#2093) (1.15.3.0, 1.14.10.0)
  - *ci*: For security, replace some workflow substitutions with safer env substitutions [#2088](#2088) (1.15.3.0, 1.14.10.0)
  - *ci*: Some minor LLVM dependency improvements [#2090](#2090)
  - *ci*: Various fixes for unexpected CI runner breakages: fmt/OIIO main build [#2095](#2095), Mac compiler path [#2096](#2096), Intel compiler certificate update [#2101](#2101), OIIO dependency hash lockdown [#2102](#2102) (1.15.3.0, 1.14.10.0)

v1.14.10.0

Release 1.14.10.0 (1 Apr 2026) -- compared to 1.14.9.0
---------------------------------------------------------
  - *build*: Fix compilation with ld.lld [#2094](#2094) (by Sv. Lockal)
  - *test*: Check in reference images for Mac ARM debug case [#2083](#2083)
  - *ci*: Add Windows 2025 with MSVS 2026 job variant to CI [#2093](#2093)
  - *ci*: For security, replace some workflow substitutions with safer env substitutions [#2088](#2088)
  - *ci*: Various fixes for unexpected CI runner breakages: Windows curl LLVM download [#2089](#2089), Mac homebrew installations [#2091](#2091), Mac compiler path [#2096](#2096), fmt/OIIO main build [#2095](#2095), update certificates to be able to install Intel compilers [#2101](#2101).
  - *ci*: Repair broken CI from OIIO locking down dependency hashes [#2102](#2102)

v1.15.2.0

Release 1.15.2.0 (Mar 10, 2026) -- compared to 1.15.1.0
-------------------------------------------------------------
  - *feat*: Add more shader group diagnostics to `compile_report` (active layers, connection depth, texture and noise op counts) [#2085](#2085)
  - *bsdl improvements*:
    - *bsdl*: Move GeneralizedSchlick lobe from testrender into BSDL library, unifying DielectricRefl/DielectricBoth into a single `DielectricBSDF<Fresnel>` template [#2078](#2078) (by Alejandro Conty)
    - *bsdl*: Move MaterialX sheen BSDF into BSDL as `mtx::SheenLobe`, consolidating sheen implementations from testrender and SPI into one place [#2084](#2084) (by Alejandro Conty)
    - *bsdl*: Clamp BSDL/MTX Fresnel inputs to [0,1] and fix bootstrap Imath link [#2081](#2081) (by Alejandro Conty)
  - *fix*: Make `linearstep` definitely return 1 for x > edge1 [#2064](#2064) (by Roeland Schoukens)
  - *perf/fix*: Change default JIT FMA mode to Strict (no fused multiply-add), improving cross-platform numerical consistency [#2076](#2076)
  - *testshade*: Route errors through renderer's own error handler; convert raw pointers to `unique_ptr` [#2082](#2082)
  - *build*: Support LLVM 22 / clang 22 for building [#2086](#2086)
  - *ci*: Add macOS 26 (ARM) to the CI lineup [#2079](#2079)
  - *ci*: Stop nightly CI workflows on user forks [#2074](#2074)
  - *ci*: Fix Windows curl LLVM download failure [#2089](#2089)
  - *ci*: Fix and speed up Mac CI dependency installations via Homebrew [#2091](#2091)
  - *testing*: Check in ref images for Mac ARM debug case [#2083](#2083)
  - *docs*: Fix typo [#2077](#2077) (by Taraash)
  - *admin*: Fix link to icon used for Slack release announcements [#2070](#2070)

v1.14.9.0

Release 1.14.9.0 (10 Mar 2026) -- compared to 1.14.8.0
-------------------------------------------------------------
  - *feat*: Add more shader group diagnostics to `compile_report`: when `compile_report >= 1`, report number of active layers, maximum connection depth of the shader network, number of texture ops, and number of noise ops. [#2085](#2085)
  - *testshade*: Use the renderer's own error handler and convert raw pointers to `unique_ptr` [#2082](#2082)
  - *build*: Support LLVM 22 and clang 22 [#2086](#2086)
  - *ci*: Several CI improvements (disk space, container names, nightly workflow control, Python version, gif library dedup) [#2074](#2074) [#2061](#2061) [#2055](#2055) [#2066](#2066) [#2049](#2049)
  - *docs*: Fix typo [#2077](#2077)