fix(bench): refuse debug native builds before timing

tests/_native_guard.py rebuilds the cwd-local _interpreter.pyd via plain
`maturin develop` (debug) whenever src/*.rs is newer; the cwd package
shadows the venv release wheel, so a bench run after a Rust edit + pytest
silently timed a DEBUG engine -- engine rows inflated ~5x while
python/duckdb rows stayed sane (measured 2026-07-26).

The extension now exposes BUILD_PROFILE from cfg!(debug_assertions);
bench_serving exits with a rebuild hint unless it reads release, and
the native guard's rebuild message names the profile it builds.

Verified: debug build -> clear refusal (exit 1); release build -> normal
run; mise run gate-specializer green (129 cargo + 753 pytest).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Author: ahrzb

benchmarks/bench_serving.py

@@ -24,7 +24,9 @@
 Run: uv run python -m benchmarks.bench_serving [--json out.json]
 IMPORTANT: rebuild the wheel first (uv run --reinstall-package sql-transform
 python -c pass) — a stale wheel inflates ONLY the engine rows and once
-produced a phantom 7x regression (caught by bisection, 2026-07-26).
+produced a phantom 7x regression (caught by bisection, 2026-07-26). Debug
+builds (the native test guard's `maturin develop` shadowing the release
+wheel) are refused automatically before any timing.
 """
 
 from __future__ import annotations
@@ -140,6 +142,24 @@ def engine_run(engines):
     print(json.dumps(results))
 
 
+def refuse_debug_build():
+    """Abort unless the imported native extension is a release build.
+
+    tests/_native_guard.py rebuilds the cwd-local .pyd via plain `maturin
+    develop` (debug) whenever src/*.rs is newer; that shadows the venv's
+    release wheel and silently inflates engine rows ~5x (measured 2026-07-26).
+    """
+    from sql_transform import _interpreter
+
+    profile = getattr(_interpreter, "BUILD_PROFILE", None)
+    if profile != "release":
+        sys.exit(
+            f"bench_serving: refusing to time a non-release native build "
+            f"(BUILD_PROFILE={profile!r}, loaded from {_interpreter.__file__}).\n"
+            f"Rebuild with: uv run maturin develop --release"
+        )
+
+
 def orchestrate():
     print("parity gate:", flush=True)
     for mod in map(sc.load, scenario_names()):
@@ -192,6 +212,7 @@ def orchestrate():
 
 
 if __name__ == "__main__":
+    refuse_debug_build()
     if "--engine-run" in sys.argv:
         engine_run(os.environ["BENCH_ENGINES"].split(","))
     else:

src/lib.rs

@@ -25,5 +25,15 @@ mod value;
 fn _interpreter(m: &Bound<'_, PyModule>) -> PyResult<()> {
     m.add_class::<datafusion::InferFn>()?;
     m.add_class::<duckdb::DuckDBInferFn>()?;
+    // Lets benchmarks refuse an unoptimized build (a `maturin develop` debug
+    // .pyd shadowing the release wheel once inflated engine rows ~5x).
+    m.add(
+        "BUILD_PROFILE",
+        if cfg!(debug_assertions) {
+            "debug"
+        } else {
+            "release"
+        },
+    )?;
     Ok(())
 }

tests/_native_guard.py

@@ -73,7 +73,8 @@ def ensure_native_built() -> None:
         return
     print(
         "native guard: src/*.rs newer than _interpreter -- rebuilding "
-        "(maturin develop)...",
+        "(maturin develop, DEBUG profile; benchmarks will refuse this build "
+        "-- rerun with --release before timing)...",
         file=sys.stderr,
     )
     # noqa justification: maturin is the venv's own executable and _REPO derives