fix(build): suppress MLX folding warning after flags

CMakeLists.txt

@@ -831,12 +831,6 @@ if(VLLM_CPP_METAL)
     endif()
     target_compile_definitions(vllm PUBLIC VLLM_CPP_MLX)
     target_sources(vllm PRIVATE src/vt/metal/metal_mlx_provider.mm)
-    # MLX 0.29.x headers use GNU-folding expressions that AppleClang diagnoses.
-    # Suppress that one external-header diagnostic for this TU. A source-level
-    # -Wno-error is ordered before vllm_cpp_set_warnings' target-level -Werror,
-    # so AppleClang promotes it again.
-    set_source_files_properties(src/vt/metal/metal_mlx_provider.mm PROPERTIES
-      COMPILE_OPTIONS "$<$<COMPILE_LANG_AND_ID:OBJCXX,AppleClang>:-Wno-gnu-folding-constant>")
     # SYSTEM so MLX's own headers cannot break our -Werror build.
     target_include_directories(vllm SYSTEM PRIVATE "${MLX_ROOT}/include")
     target_link_libraries(vllm PUBLIC ${MLX_LIBRARY})

docs/BENCHMARKS.md

@@ -6530,6 +6530,11 @@ us by about 1.5 points.
 The default (non-MLX) build is **95.9%** against this corrected baseline, not
 96.4%.
 
+The 2026-08-02 AppleClang warning-suppression correction is **NOT APPLICABLE**
+to benchmark results: it changes only whether the optional MLX provider compiles
+under target-wide `-Werror`. Darwin CI remains the build-verification gate; no
+runtime path, measurement, or binding number changed.
+
 **Everything qualitative in the entry below still holds** — MLX wins prefill, the
 shape gate is the right disposition, the fallback hoist was worth 27.2 vs 17.8 —
 only the headline ratio moves. The gate is still worth about +1.7 points over the

docs/STATUS.md

@@ -1407,6 +1407,9 @@ loop, not parallelism or layout. An optional MLX GEMM provider is available via
 `-DVLLM_CPP_MLX=ON` and currently measures net slower than our own kernels. Full
 per-lever chronology: [docs/BENCHMARKS.md](BENCHMARKS.md) and
 [.agents/specs/metal-dispatch-attribution.md](../.agents/specs/metal-dispatch-attribution.md).
+The MLX-enabled Darwin build remains **build-verification pending**: its narrow
+AppleClang header-warning suppression now lives in the provider translation unit
+so the target-wide `-Werror` cannot re-promote it; Darwin CI is the binding gate.
 
 **CUDA architectures.** The production target is GB10/`sm_121a` (runtime-gated,
 both gate models token-exact + at/above vLLM speed). The arch-additivity

src/vt/metal/metal_mlx_provider.mm

@@ -50,6 +50,13 @@
 #include <cstring>
 #include <vector>
 
+// MLX 0.29.x uses constant-folded variable length arrays in public headers.
+// Keep the target-wide -Werror policy, but ignore that external-header-only
+// AppleClang diagnostic after command-line warning options have been applied.
+#if defined(__clang__)
+#pragma clang diagnostic ignored "-Wgnu-folding-constant"
+#endif
+
 // MLX public headers. Deliberately NOT mlx/backend/metal/*: those pull in
 // metal-cpp and, as noted above, their entry points are not exported anyway.
 #include "mlx/allocator.h"