lusoris · May 30, 2026
@@ -0,0 +1 @@
+- `core/src/read_json_model`: converted to C++23 (`read_json_model.cpp`, ADR-0846 Wave 8). `goto`-based teardown replaced with a scoped RAII guard; `unique_ptr<char[]>` for collection-name buffer; `std::string` for feature-option key copy; `[[nodiscard]]` on capacity helpers. No behaviour change; public C API and ABI unchanged.
@@ -104,8 +104,16 @@ libvmaf/
   is `push_c()` at entry → body → `pop()` before the `ferror`
   check; dropping the `pop()` leaks a `locale_t` on POSIX and
   leaves the calling thread locked to `"C"` on Windows.
+- **JSON model loader is now `read_json_model.cpp` (ADR-0846 / Wave 8).**
+  The active translation unit is [`src/read_json_model.cpp`](src/read_json_model.cpp),
+  compiled as the isolated `read_json_model_cpp23_lib` static lib
+  (`cpp_std=c++23`).  The original [`src/read_json_model.c`](src/read_json_model.c)
+  is still on disk but is **not compiled** — the `libvmaf_sources` list replaced
+  it with a comment.  Do not re-add `read_json_model.c` to `libvmaf_sources`;
+  doing so produces duplicate-symbol link errors for all four `extern "C"` entry
+  points.  The `.c` file will be deleted in a follow-on cleanup PR.
 - **JSON model loader has no fixed feature/knot schema ceiling.**
-  [`src/read_json_model.c`](src/read_json_model.c) grows
+  [`src/read_json_model.cpp`](src/read_json_model.cpp) grows
   `VmafModel.feature` and `score_transform.knots.list` from the JSON
   payload. Do not restore the old `MAX_FEATURE_COUNT` / `MAX_KNOT_COUNT`
   rejection pattern during an upstream sync; external model JSONs with
@@ -457,6 +465,7 @@ the corrected methodology.
   — `enable_avx512=true` with `enable_asm=false` issues a warning (no-op, not an error);
   — `enable_hipcc=true` with `enable_hip=false` issues a warning (no-op, not an error).
   The checks run at configuration time (before `subdir()` calls) to catch misconfigurations early. The principle: every option that depends on another must `error()` on the bad combo, never silently no-op. See [`src/meson.build` lines 100–111, 74–76, 142–144](src/meson.build).
+<<<<<<< HEAD
 
 - **C→C++23 conversion safety invariants** (adversarial review 2026-05-28,
   `docs/research/cpp23-wave-adversarial-review-20260528.md`):
@@ -512,6 +521,8 @@ the corrected methodology.
   The legacy path (`VmafFeatureExtractor`, line ~301) retains the mapping as
   documented debt — tracked as T-LEGACY-RUNNER-ANSNR-BROKEN in `docs/state.md`.
   The checks run at configuration time (before `subdir()` calls) to catch misconfigurations early. The principle: every option that depends on another must `error()` on the bad combo, never silently no-op. See [`src/meson.build` lines 100–111, 74–76, 142–144](src/meson.build).
+=======
+>>>>>>> 24bb5daf89 (docs: post-merge-train sweep — VMAFx + core/ path refs, ADR index, state.md)
 
 ## Performance benchmark invariant (ADR-0752)
 

@@ -551,27 +551,13 @@ libvmaf_include = include_directories(
     is_nvtx_enabled ? '/usr/local/cuda/include' : '',
 )
 
-# GCC/Clang accept `c++23`; MSVC needs `vc++latest`; Intel oneAPI on Windows
-# (icx-cl, `intel-llvm-cl`) and clang-cl reject `c++23` and need plain
-# `c++latest`. Pick the right token per compiler so override_options
-# doesn't fail meson configure.
-_cpp_id = meson.get_compiler('cpp').get_id()
-if _cpp_id == 'msvc'
-    libvmaf_cpu_cpp_std = 'vc++latest'
-elif _cpp_id == 'intel-llvm-cl' or _cpp_id == 'intel-cl' or _cpp_id == 'clang-cl'
-    libvmaf_cpu_cpp_std = 'c++latest'
-else
-    libvmaf_cpu_cpp_std = 'c++23'
-endif
-
 libvmaf_cpu_static_lib = static_library(
     'libvmaf_cpu',
     libvmaf_cpu_sources,
     include_directories : [libvmaf_include],
     # ADR-0755: cpu.cpp uses constinit (C++20) and std::atomic; override
     # cpp_std so this TU compiles under C++23 without affecting other TUs.
-    # ADR-0860: MSVC-vs-GCC/Clang cpp_std token disagreement workaround.
-    override_options : ['cpp_std=' + libvmaf_cpu_cpp_std],
+    override_options : ['cpp_std=c++23'],
 )
 
 platform_specific_cpu_objects = []
@@ -1535,6 +1521,21 @@ metadata_handler_cpp20_lib = static_library(
     install : false,
 )
 
+# ADR-0846: read_json_model.cpp is a C++23 Wave 8 TU. Compiled as an isolated
+# static lib so its cpp_std override (c++23) does not propagate to any other TU
+# in libvmaf.  Mirrors the ADR-0708 pattern for metadata_handler_cpp20_lib.
+# The four public entry points retain their original C signatures and are
+# declared extern "C" inside the .cpp, so all existing C callers link without
+# modification.
+read_json_model_cpp23_lib = static_library(
+    'read_json_model_cpp23',
+    src_dir + 'read_json_model.cpp',
+    include_directories : [vmaf_base_include, libvmaf_include],
+    override_options : ['cpp_std=c++23'],
+    pic : true,
+    install : false,
+)
+
 libvmaf_sources = [
     src_dir + 'libvmaf.c',
     src_dir + 'predict.c',
@@ -1547,7 +1548,7 @@ libvmaf_sources = [
     src_dir + 'dict.c',
     src_dir + 'opt.c',
     src_dir + 'ref.c',
-    src_dir + 'read_json_model.c',
+    # read_json_model.cpp is compiled as read_json_model_cpp23_lib (ADR-0846)
     src_dir + 'pdjson.c',
     src_dir + 'log.c',
     src_dir + 'framesync.c',
@@ -1636,6 +1637,8 @@ libvmaf = library(
         libsvm_static_lib.extract_all_objects(recursive: true),
         # ADR-0708: C++20 pilot TU compiled as isolated static lib.
         metadata_handler_cpp20_lib.extract_all_objects(recursive: true),
+        # ADR-0846: C++23 Wave 8 TU — read_json_model.cpp.
+        read_json_model_cpp23_lib.extract_all_objects(recursive: true),
     ],
     version : vmaf_soname_version,
     soversion : vmaf_soversion,
Original file line number	Diff line number	Diff line change
		@@ -0,0 +1 @@
		- `core/src/read_json_model`: converted to C++23 (`read_json_model.cpp`, ADR-0846 Wave 8). `goto`-based teardown replaced with a scoped RAII guard; `unique_ptr<char[]>` for collection-name buffer; `std::string` for feature-option key copy; `[[nodiscard]]` on capacity helpers. No behaviour change; public C API and ABI unchanged.