C内核按库功能拆解，编译结果缓存区构建，编译过程与已有缓存结果对照功能实现

tests/manual/backend_stage_profile.py

@@ -291,7 +291,11 @@ def main():
               'mode':'plain' if args.plain else 'profile','port':args.port,
               'python':sys.version,'platform':sys.platform,
               'frontendFiles':{str(p.relative_to(out/'frontend')):sha256(p.read_bytes()).hexdigest() for p in (out/'frontend').rglob('*') if p.is_file()},
-              'productionKernelSha256':sha256((ROOT/'native/components/kernels.c').read_bytes()).hexdigest()}
+              # Retain the historical entry hash; module/header hashes now
+              # identify the numerical sources behind that diagnostic entry.
+              'productionKernelSha256':sha256((ROOT/'native/components/kernels.c').read_bytes()).hexdigest(),
+              'productionNativeSourceHashes':{p.relative_to(ROOT/'native').as_posix():sha256(p.read_bytes()).hexdigest()
+                    for p in sorted((ROOT/'native').rglob('*')) if p.is_file() and p.suffix in ('.c','.h')}}
     (out/'environment.json').write_text(json.dumps(metadata,indent=2)+'\n')
     import uvicorn
     uvicorn.run(profile.app(api.app) if profile else api.app,host='127.0.0.1',port=args.port)

tests/manual/benchmark_native_build_cache.py

@@ -0,0 +1,382 @@
+"""Prepare or serially measure frozen/current native builds; never run a model.
+
+Example (fresh output directory; omit --run for preparation only)::
+
+    .venv/bin/python tests/manual/benchmark_native_build_cache.py \
+      --baseline-root test/incremental-build-20260912/baseline-source \
+      --output-dir test/incremental-build-20260912/benchmark \
+      --warmups 1 --repeats 3 --run
+
+Each version/round gets an empty application cache, followed by A/full hit,
+A-prime/changed pipe length, A/full hit, four branches/first build, A/full hit.
+Application-cache cold does not mean OS page-cache cold. Versions run in separate
+Python processes, with alternating version order by round. Preparation, imports,
+hashing and recording are outside build timing. NativeBuild.close() is measured
+separately and as part of an adjacent build-and-release wall interval. Compilation
+retains each production builder's concurrency and flags. No solver, frontend,
+strace or previous diagnostic timings are invoked/combined.
+"""
+from __future__ import annotations
+
+import argparse
+from copy import deepcopy
+import csv
+from datetime import datetime, timezone
+from hashlib import sha256
+import json
+import math
+import os
+from pathlib import Path
+import platform
+import shutil
+import statistics
+import subprocess
+import sys
+import time
+
+ROOT = Path(__file__).resolve().parents[2]
+SCRIPT = Path(__file__).resolve()
+SCENARIOS = (
+    ("cold", "a", False), ("same_model", "a", True),
+    ("parameter_change", "a_prime", False), ("return_to_a", "a", True),
+    ("four_branches", "four", False), ("return_from_four", "a", True),
+)
+METRICS = ("buildSeconds", "buildCallWallSeconds", "releaseWallSeconds",
+           "buildAndReleaseWallSeconds", "executableBytes", "cacheBytes",
+           "preprocessSeconds", "compileSeconds", "compileWallSeconds",
+           "linkSeconds", "unitCount", "objectCacheHits", "objectCompilations")
+ENVIRONMENT = ("SIMULATION_NATIVE_CC", "SUNDIALS_ROOT",
+               "SIMULATION_NATIVE_MODEL_CACHE_MB", "SIMULATION_NATIVE_OBJECT_CACHE_MB")
+
+
+def digest(path: Path) -> str:
+    with path.open("rb") as stream:
+        value = sha256()
+        for block in iter(lambda: stream.read(1024 * 1024), b""):
+            value.update(block)
+    return value.hexdigest()
+
+
+def json_digest(value) -> str:
+    return sha256(json.dumps(value, sort_keys=True, separators=(",", ":")).encode()).hexdigest()
+
+
+def write_json(path: Path, value) -> None:
+    path.write_text(json.dumps(value, ensure_ascii=False, indent=2, allow_nan=False) + "\n", encoding="utf-8")
+
+
+def source_files(root: Path) -> dict:
+    paths = list((root / "app").rglob("*.py"))
+    paths += [p for directory in ("native", "schemas") for p in (root / directory).rglob("*") if p.is_file()]
+    return {p.relative_to(root).as_posix(): digest(p) for p in sorted(paths)
+            if "__pycache__" not in p.parts and not p.relative_to(root).as_posix().startswith("app/data/")}
+
+
+def source_record(root: Path) -> dict:
+    if not (root / "app/simulation/native_codegen/build.py").is_file() or not (root / "native").is_dir():
+        raise ValueError(f"Incomplete source root: {root}")
+    frozen = root / "source-manifest.json"
+    if frozen.exists():
+        manifest = json.loads(frozen.read_text())
+        for name, expected in manifest["trackedFiles"].items():
+            if digest(root / name) != expected:
+                raise ValueError(f"Frozen source integrity failure: {root / name}")
+        revision = {"gitHead": manifest["head"], "frozenManifestSha256": digest(frozen), "worktreeStatus": None}
+    else:
+        revision = {"gitHead": subprocess.check_output(["git", "rev-parse", "HEAD"], cwd=root, text=True).strip(),
+                    "worktreeStatus": subprocess.check_output(["git", "status", "--porcelain", "--", "app", "native", "schemas"], cwd=root, text=True)}
+    files = source_files(root)
+    return {"root": str(root), **revision, "sourceFiles": files, "sourceFilesSha256": json_digest(files)}
+
+
+def assert_source(record: dict) -> None:
+    if source_files(Path(record["root"])) != record["sourceFiles"]:
+        raise RuntimeError(f"Source changed after preparation: {record['root']}")
+
+
+def make_inputs(args, output: Path) -> tuple[dict, dict]:
+    directory = output / "inputs"
+    directory.mkdir()
+    paths = {"a": directory / "model-a.json", "a_prime": directory / "model-a-prime.json", "four": directory / "model-four.json"}
+    shutil.copyfile(args.input, paths["a"])
+    shutil.copyfile(args.four_input, paths["four"])
+    original = json.loads(paths["a"].read_text(encoding="utf-8"))
+    variant = deepcopy(original)
+    matches = [(i, node) for i, node in enumerate(variant["nodes"]) if node["id"] == args.variant_node]
+    if len(matches) != 1:
+        raise ValueError(f"Expected exactly one variant node: {args.variant_node}")
+    index, node = matches[0]
+    data = node["data"]
+    if data["modelType"] != "amesim_pnl0001":
+        raise ValueError("The controlled length variant requires an amesim_pnl0001 pipe")
+    old = data["parameters"]["le"]
+    if type(old) not in (int, float) or not math.isfinite(old) or old <= 0:
+        raise ValueError("Pipe le must be a positive numeric SI value")
+    new = old * 1.001
+    if not math.isfinite(new) or new == old:
+        raise ValueError("Pipe length variant did not produce a finite changed value")
+    data["parameters"]["le"] = new
+    write_json(paths["a_prime"], variant)
+    data["parameters"]["le"] = old
+    if variant != original:
+        raise RuntimeError("Variant changed fields beyond the selected pipe length")
+    inputs = {key: {"path": str(path), "sha256": digest(path)} for key, path in paths.items()}
+    change = {"nodeId": node["id"], "modelType": data["modelType"], "jsonPointer": f"/nodes/{index}/data/parameters/le",
+              "before": old, "after": new, "storedUnit": "m", "factor": 1.001,
+              "note": "Stored numeric values are SI; editor parameterUnits do not rescale them."}
+    return inputs, change
+
+
+def generate(config: dict, version: str, *, save: bool):
+    source = config["versions"][version]
+    assert_source(source)
+    root = Path(source["root"])
+    # Every worker starts a fresh interpreter; neither version imports the other.
+    sys.path.insert(0, str(root))
+    from app.main import compile_system_xml_network
+    from app.simulation.native_codegen.compiler import compile_native_program
+    from app.simulation.native_codegen.input import load_input
+    from app.simulation.native_codegen import build as builder
+    for name, module in tuple(sys.modules.items()):
+        location = getattr(module, "__file__", None)
+        if (name == "app" or name.startswith("app.")) and location and not Path(location).resolve().is_relative_to(root):
+            raise RuntimeError(f"Mixed source imports: {name}: {location}")
+    if builder.NATIVE.resolve() != root / "native":
+        raise RuntimeError("Builder native source root differs from the selected version")
+    generated = Path(config["outputDir"]) / version / "generated"
+    if save:
+        generated.mkdir(parents=True)
+    programs, records = {}, {}
+    for key, item in config["inputs"].items():
+        path = Path(item["path"])
+        if digest(path) != item["sha256"]:
+            raise RuntimeError(f"Input changed after preparation: {path}")
+        started = time.perf_counter()
+        xml, document = load_input(path)
+        program = compile_native_program(compile_system_xml_network(document))
+        preparation = time.perf_counter() - started
+        contract = program.manifest()
+        record = {"inputSha256": item["sha256"], "xmlSha256": sha256(xml).hexdigest(),
+                  "sourceSha256": sha256(program.source.encode()).hexdigest(), "headerSha256": sha256(program.header.encode()).hexdigest(),
+                  "contractSha256": json_digest(contract), "stateCount": len(program.state_keys),
+                  "variableCount": len(program.variables)}
+        if save:
+            directory = generated / key
+            directory.mkdir()
+            (directory / "input.xml").write_bytes(xml)
+            (directory / "model.c").write_text(program.source, encoding="utf-8")
+            (directory / "model.h").write_text(program.header, encoding="utf-8")
+            write_json(directory / "contract.json", contract)
+            write_json(directory / "preparation.json", {**record, "generationSeconds": preparation})
+        programs[key], records[key] = program, record
+    if programs["a"].header != programs["a_prime"].header or programs["a"].state_keys != programs["a_prime"].state_keys:
+        raise RuntimeError("Length variant changed model header/state layout")
+    if programs["a"].source == programs["a_prime"].source:
+        raise RuntimeError("Length variant did not change generated C")
+    if not save and records != config["generated"][version]:
+        raise RuntimeError("Regenerated model differs from prepared artifacts")
+    assert_source(source)
+    return builder, programs, records
+
+
+def cache_size(cache: Path) -> dict:
+    files = [p for p in cache.rglob("*") if p.is_file()]
+    return {"logicalFileBytes": sum(p.stat().st_size for p in files), "fileCount": len(files),
+            "modelsBytes": sum(p.stat().st_size for p in files if p.relative_to(cache).parts[0] == "models"),
+            "objectsBytes": sum(p.stat().st_size for p in files if p.relative_to(cache).parts[0] == "objects")}
+
+
+def worker(args) -> None:
+    config = json.loads(args.worker_config.read_text())
+    if digest(SCRIPT) != config["scriptSha256"]:
+        raise RuntimeError("Benchmark script changed after preparation")
+    if {key: os.environ.get(key) for key in ENVIRONMENT} != config["environment"]:
+        raise RuntimeError("Build environment changed after preparation")
+    builder, programs, records = generate(config, args.version, save=args.worker == "prepare")
+    if args.worker == "prepare":
+        write_json(Path(config["outputDir"]) / args.version / "prepared-models.json", records)
+        return
+    directory = Path(config["outputDir"]) / args.version / args.label
+    directory.mkdir()
+    cache = directory / "cache"
+    if cache.exists():
+        raise RuntimeError("Cold build requires a previously nonexistent cache directory")
+    rows = []
+    for scenario, model, expected_hit in SCENARIOS:
+        case = directory / scenario
+        case.mkdir()
+        before = time.perf_counter()
+        built = builder.build_native(programs[model], cache_dir=cache)
+        returned = time.perf_counter()
+        release_start = time.perf_counter()
+        if hasattr(built, "close"):
+            built.close()
+        released = time.perf_counter()
+        manifest = built.manifest
+        copied_manifest = case / "build-manifest.json"
+        write_json(copied_manifest, manifest)
+        details = getattr(built, "details", {})
+        size = cache_size(cache)
+        row = {"version": args.version, "run": args.label, "warmup": args.label.startswith("warmup-"),
+               "scenario": scenario, "model": model, "buildSeconds": built.seconds,
+               "buildCallWallSeconds": returned - before, "releaseWallSeconds": released - release_start,
+               "buildAndReleaseWallSeconds": released - before, "cacheHit": built.cache_hit,
+               "buildKey": manifest["buildKey"], "details": details,
+               "executable": str(built.executable), "executableSha256": digest(built.executable),
+               "executableBytes": built.executable.stat().st_size, "cacheBytes": size["logicalFileBytes"], "cacheSize": size,
+               "sourceFilesSha256": config["versions"][args.version]["sourceFilesSha256"],
+               "generated": records[model], "compiler": manifest["compiler"], "compilerFlags": manifest["compilerFlags"],
+               "dependencyHashes": manifest["dependencyHashes"], "sourceHashes": manifest["sourceHashes"],
+               "nativeHeaderHashes": manifest.get("nativeHeaderHashes"),
+               "manifest": str(copied_manifest.relative_to(Path(config["outputDir"]))), "manifestSha256": digest(copied_manifest)}
+        row["checks"] = {"expectedCacheHit": built.cache_hit == expected_hit,
+                         "parameterChangeReusesOtherObjects": (details.get("objectCompilations") == 1 and
+                             details.get("objectCacheHits") == details.get("unitCount", 0) - 1)
+                         if scenario == "parameter_change" and "objectCompilations" in details else None}
+        write_json(case / "measurement.json", row)
+        rows.append(row)
+        if any(value is False for value in row["checks"].values()):
+            raise RuntimeError(f"Unexpected cache behavior in {args.version}/{args.label}/{scenario}; measurement preserved")
+    assert_source(config["versions"][args.version])
+    write_json(directory / "measurements.json", {"complete": True, "solverExecuted": False, "rows": rows})
+
+
+def invoke(config: dict, mode: str, version: str, label: str | None = None) -> dict:
+    output = Path(config["outputDir"])
+    command = [sys.executable, str(SCRIPT), "--worker", mode, "--worker-config", str(output / "prepared.json"), "--version", version]
+    if label:
+        command += ["--label", label]
+    log = output / "logs" / (version + "-" + (label or mode))
+    start = time.perf_counter()
+    with log.with_suffix(".stdout.log").open("wb") as stdout, log.with_suffix(".stderr.log").open("wb") as stderr:
+        process = subprocess.run(command, cwd=config["versions"][version]["root"], stdout=stdout, stderr=stderr, timeout=config["workerTimeout"])
+    observed = {"command": command, "processWallSeconds": time.perf_counter() - start, "exitCode": process.returncode,
+                "version": version, "run": label, "scope": "Whole worker including import/generation/checks/recording; not build latency."}
+    write_json(log.with_suffix(".process.json"), observed)
+    if process.returncode:
+        raise RuntimeError(f"Worker failed ({process.returncode}); see {log.with_suffix('.stderr.log')}")
+    return observed
+
+
+def stat(values: list) -> dict:
+    return {"n": len(values), "min": min(values), "median": statistics.median(values), "max": max(values)} if values else {"n": 0, "min": None, "median": None, "max": None}
+
+
+def metric(row: dict, name: str):
+    return row[name] if name in row else row["details"].get(name)
+
+
+def summarize(config: dict, rows: list, processes: list) -> dict:
+    expected = 2 * (config["warmups"] + config["repeats"]) * len(SCENARIOS)
+    if len(rows) != expected:
+        raise RuntimeError(f"Incomplete benchmark: expected {expected} rows, found {len(rows)}")
+    compilers = {(r["compiler"], tuple(r["compilerFlags"])) for r in rows}
+    libraries = {json_digest({k: v for k, v in r["dependencyHashes"].items() if "/" not in k}) for r in rows}
+    if len(compilers) != 1 or len(libraries) != 1:
+        raise RuntimeError("Compiler/flags or linked library bytes differ across benchmark runs")
+    statistics_by_version = {}
+    for version in config["versions"]:
+        statistics_by_version[version] = {}
+        for scenario, _, _ in SCENARIOS:
+            selected = [r for r in rows if r["version"] == version and r["scenario"] == scenario and not r["warmup"]]
+            if len(selected) != config["repeats"]:
+                raise RuntimeError(f"Incomplete formal group: {version}/{scenario}")
+            statistics_by_version[version][scenario] = {key: stat([metric(r, key) for r in selected if metric(r, key) is not None]) for key in METRICS}
+    changes = {}
+    for scenario, _, _ in SCENARIOS:
+        changes[scenario] = {}
+        for key in ("buildSeconds", "buildCallWallSeconds", "buildAndReleaseWallSeconds"):
+            old = statistics_by_version["baseline"][scenario][key]["median"]
+            new = statistics_by_version["candidate"][scenario][key]["median"]
+            changes[scenario][key] = {"baselineMedian": old, "candidateMedian": new,
+                "reductionPercent": 100 * (1 - new / old) if old else None,
+                "speedup": old / new if new else None, "method": "ratio of group medians"}
+    return {"complete": True, "solverExecuted": False, "prepared": "prepared.json", "scriptSha256": config["scriptSha256"],
+            "contract": config["contract"], "checks": {"compilerAndFlagsIdentical": True, "linkedLibrariesIdentical": True,
+                "generatedModelsIdenticalAcrossVersions": True, "variantHeaderAndStateLayoutUnchanged": True},
+            "statistics": statistics_by_version, "changes": changes, "rows": rows, "workers": processes}
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
+    parser.add_argument("--baseline-root", type=Path, default=ROOT / "test/incremental-build-20260912/baseline-source")
+    parser.add_argument("--candidate-root", type=Path, default=ROOT)
+    parser.add_argument("--input", type=Path, default=ROOT / "tests/data/test-mql-8-corrected.json")
+    parser.add_argument("--four-input", type=Path, default=ROOT / "tests/data/test-mql-4-corrected.json")
+    parser.add_argument("--variant-node", default="amesim_pnl0001_1")
+    parser.add_argument("--output-dir", type=Path)
+    parser.add_argument("--warmups", type=int, default=1)
+    parser.add_argument("--repeats", type=int, default=3)
+    parser.add_argument("--worker-timeout", type=float, default=300)
+    parser.add_argument("--run", action="store_true")
+    parser.add_argument("--worker", choices=("prepare", "round"), help=argparse.SUPPRESS)
+    parser.add_argument("--worker-config", type=Path, help=argparse.SUPPRESS)
+    parser.add_argument("--version", choices=("baseline", "candidate"), help=argparse.SUPPRESS)
+    parser.add_argument("--label", help=argparse.SUPPRESS)
+    args = parser.parse_args()
+    if args.worker:
+        if not args.worker_config or not args.version or (args.worker == "round" and not args.label):
+            parser.error("Internal worker arguments are incomplete")
+        worker(args)
+        return
+    if not args.output_dir or args.warmups < 1 or args.repeats < 1 or not math.isfinite(args.worker_timeout) or args.worker_timeout <= 0:
+        parser.error("A fresh --output-dir, warmups/repeats >= 1 and a positive worker timeout are required")
+    output = args.output_dir.resolve()
+    if not output.is_relative_to(ROOT / "test") or output.exists():
+        parser.error("Choose a previously nonexistent directory under the ignored test/ tree")
+    versions = {"baseline": source_record(args.baseline_root.resolve()), "candidate": source_record(args.candidate_root.resolve())}
+    output.mkdir(parents=True)
+    (output / "logs").mkdir()
+    try:
+        inputs, change = make_inputs(args, output)
+        config = {"createdAtUtc": datetime.now(timezone.utc).isoformat(), "outputDir": str(output), "versions": versions,
+                  "script": str(SCRIPT), "scriptSha256": digest(SCRIPT), "python": sys.version, "pythonExecutable": sys.executable,
+                  "platform": platform.platform(), "cpuCount": os.cpu_count(), "environment": {key: os.environ.get(key) for key in ENVIRONMENT},
+                  "inputs": inputs, "variant": change, "warmups": args.warmups, "repeats": args.repeats,
+                  "workerTimeout": args.worker_timeout, "runRequested": args.run,
+                  "contract": {"cold": "Independent empty application cache per version/round; OS caches are not flushed.",
+                    "sequence": [name for name, _, _ in SCENARIOS], "solverExecuted": False,
+                    "fourBranchCase": "Switch from the full eight-branch project to the corrected four-branch project, then restore A; no cached files are manually deleted.",
+                    "wallTiming": "Builder call and adjacent release are measured separately; generation, hashing, imports and recording are outside these spans.",
+                    "overlap": "buildSeconds is inside buildCallWallSeconds, which is inside buildAndReleaseWallSeconds. Detail spans are nested; compileSeconds sums concurrent TU walls and is not additive to compileWallSeconds.",
+                    "unavailable": "Old builder has no object/stage details; absent measurements are null, never inferred from differences.",
+                    "comparison": "Formal group-median ratios; warmups excluded. No old strace timings or solver performance claims."}}
+        write_json(output / "prepared.json", config)
+        for version in versions:
+            invoke(config, "prepare", version)
+        config["generated"] = {version: json.loads((output / version / "prepared-models.json").read_text()) for version in versions}
+        if config["generated"]["baseline"] != config["generated"]["candidate"]:
+            raise RuntimeError("Generated model/XML/header/contract differs between source versions")
+        write_json(output / "prepared.json", config)
+        if not args.run:
+            print(json.dumps({"preparedOnly": True, "outputDir": str(output), "compiled": False, "solverExecuted": False}))
+            return
+        rows, processes = [], []
+        for index in range(args.warmups + args.repeats):
+            label = f"warmup-{index + 1}" if index < args.warmups else f"run-{index - args.warmups + 1}"
+            order = ("baseline", "candidate") if index % 2 == 0 else ("candidate", "baseline")
+            for version in order:
+                processes.append(invoke(config, "round", version, label))
+                rows.extend(json.loads((output / version / label / "measurements.json").read_text())["rows"])
+        for record in versions.values():
+            assert_source(record)
+        summary = summarize(config, rows, processes)
+        write_json(output / "summary.json", summary)
+        with (output / "timings.csv").open("w", newline="", encoding="utf-8") as stream:
+            writer = csv.DictWriter(stream, fieldnames=["version", "run", "warmup", "scenario", "metric", "value", "n", "min", "median", "max"])
+            writer.writeheader()
+            for row in rows:
+                for key in METRICS:
+                    writer.writerow({**{k: row[k] for k in ("version", "run", "warmup", "scenario")}, "metric": key, "value": metric(row, key)})
+            for version, groups in summary["statistics"].items():
+                for scenario, values in groups.items():
+                    for key, value in values.items():
+                        writer.writerow({"version": version, "run": "formal-statistics", "warmup": False, "scenario": scenario, "metric": key, **value})
+        print(json.dumps({"complete": True, "summary": str(output / "summary.json"), "solverExecuted": False}))
+    except Exception as exc:
+        write_json(output / "failure.json", {"complete": False, "error": str(exc), "type": type(exc).__name__, "solverExecuted": False})
+        raise
+
+
+if __name__ == "__main__":
+    main()

tests/manual/benchmark_native_pipe_solver.py

@@ -33,6 +33,76 @@ NEW_CALL = '''double base=area*p*cm/sqrt(T),K=pow(4*base/den,2)*d/length;
     return sign*native_pipe_resistance(K,rr,den/4,NULL)*den/4;'''
 
 
+def pipe_source_path(native: Path) -> Path:
+    """Locate executable pipe formulas, never a modular aggregation entry."""
+    module = native / 'components/modules/pipe.c'
+    legacy = native / 'components/kernels.c'
+    if module.is_file():
+        target = module
+    elif (native / 'components/modules').exists():
+        raise ValueError(f'Incomplete modular native snapshot: {module} is missing')
+    else:
+        target = legacy
+    if not target.is_file():
+        raise ValueError(f'Native snapshot has no audited pipe source: {native}')
+    source = target.read_text()
+    if source.count('double native_pipe_resistance(') != 1 or source.count('double native_pipe_flow(') != 1:
+        raise ValueError(f'Unsupported pipe source layout: {target}; refusing an unmodified or uninstrumented run')
+    return target
+
+
+def working_native_paths() -> list[str]:
+    return sorted('native/' + path.relative_to(ROOT / 'native').as_posix()
+                  for path in (ROOT / 'native').rglob('*') if path.is_file())
+
+
+def build_snapshot(program, native: Path, cache: Path, *, required_sources=()):
+    """Build exact snapshot sources and verify the modified files were compiled.
+
+    Historical monolithic trees retain their original set of C translation
+    units. This temporary selector override is confined to the standalone
+    manual process; production source selection is always restored afterwards.
+    An incompatible old model/runtime ABI fails compilation explicitly.
+    """
+    native = native.resolve()
+    pipe = pipe_source_path(native)
+    required = (pipe, *required_sources)
+    old_native = builder.NATIVE
+    old_selector = getattr(builder, '_runtime_sources', None)
+    legacy = pipe.name == 'kernels.c'
+    try:
+        builder.NATIVE = native
+        if legacy and old_selector is not None:
+            builder._runtime_sources = lambda _: sorted(native.rglob('*.c'))
+        elif not legacy and old_selector is None:
+            raise ValueError('A modular snapshot requires the incremental native source selector; use its matching checkout')
+        selected = (builder._runtime_sources(program) if old_selector is not None
+                    else sorted(native.rglob('*.c')))
+        if any(path.resolve() not in {source.resolve() for source in selected} for path in required):
+            raise ValueError('The model/build selector does not compile the modified pipe/RHS source; refusing a misleading run')
+        try:
+            build = builder.build_native(program, cache_dir=cache)
+        except RuntimeError as exc:
+            raise RuntimeError(
+                f'Cannot build this native snapshot ({native}). Its source layout or model/runtime ABI may be incompatible; '
+                'use matching compiler/runtime revisions. No benchmark was executed.'
+            ) from exc
+        recorded = build.manifest.get('sourceHashes', {})
+        for path in required:
+            suffix = 'native/' + path.resolve().relative_to(native).as_posix()
+            matches = [digest for name, digest in recorded.items()
+                       if name == suffix or name.replace('\\', '/').endswith('/' + suffix)]
+            if matches != [sha256(path.read_bytes()).hexdigest()]:
+                if hasattr(build, 'close'):
+                    build.close()
+                raise ValueError(f'Build manifest does not prove the modified source was compiled: {suffix}')
+        return build
+    finally:
+        builder.NATIVE = old_native
+        if old_selector is not None:
+            builder._runtime_sources = old_selector
+
+
 def main():
     parser=argparse.ArgumentParser(description=__doc__)
     parser.add_argument('input',type=Path)
@@ -42,7 +112,8 @@ def main():
     args=parser.parse_args()
     if args.runs<1:parser.error('--runs must be positive')
     out=args.output_dir.resolve();out.mkdir(parents=True,exist_ok=True)
-    if (out/'summary.json').exists():parser.error('Choose a fresh output directory')
+    if (out/'summary.json').exists() or any((out/name).exists() for name in ('fixed-point','previous-newton','guarded-newton')):
+        parser.error('Choose a fresh output directory; existing variant sources must not be mixed')
     revision=subprocess.check_output(['git','rev-parse',args.baseline_ref],cwd=ROOT,text=True).strip()
     xml,doc=load_input(args.input)
     program=compile_native_program(compile_system_xml_network(doc))
@@ -55,16 +126,16 @@ def main():
     try:
         for variant in ('fixed-point','previous-newton','guarded-newton'):
             directory=out/variant
-            for name in paths:
+            variant_paths = working_native_paths() if variant == 'guarded-newton' else paths
+            for name in variant_paths:
                 if variant=='guarded-newton':data=(ROOT/name).read_bytes()
                 else:data=subprocess.check_output(['git','show',f'{revision}:{name}'],cwd=ROOT)
                 target=directory/name;target.parent.mkdir(parents=True,exist_ok=True);target.write_bytes(data)
             if variant=='fixed-point':
-                target=directory/'native/components/kernels.c';source=target.read_text()
+                target=pipe_source_path(directory/'native');source=target.read_text()
                 if source.count(NEW_CALL)!=1:raise ValueError('Baseline pipe flow layout does not match the audited fixed-point substitution')
                 target.write_text(source.replace(NEW_CALL,OLD_ITERATION))
-            builder.NATIVE=directory/'native'
-            variants[variant]=builder.build_native(program,cache_dir=out/'cache')
+            variants[variant]=build_snapshot(program,directory/'native',out/'cache')
     finally:builder.NATIVE=original_native
     rows=[]
     for index in range(args.runs+1):
@@ -80,7 +151,9 @@ def main():
     summary={'baselineRef':revision,'input':str(args.input.resolve()),
              'inputSha256':sha256(args.input.read_bytes()).hexdigest(),'xmlSha256':sha256(xml).hexdigest(),
              'settings':vars(config),'sampleStep':doc.simulation.sample_step,'rows':rows,
-             'variants':{name:{'sourceSha256':sha256((out/name/'native/components/kernels.c').read_bytes()).hexdigest(),
+             'variants':{name:{'sourceSha256':sha256(pipe_source_path(out/name/'native').read_bytes()).hexdigest(),
+                 'pipeSource':str(pipe_source_path(out/name/'native').relative_to(out/name)),
+                 'nativeSourceHashes':build.manifest['sourceHashes'],
                  'buildKey':build.manifest['buildKey'],
                  'medianSolveSeconds':statistics.median(row['solveSeconds'] for row in rows if row['variant']==name and row['run']>0),
                  'medianProcessSeconds':statistics.median(row['processWallSeconds'] for row in rows if row['variant']==name and row['run']>0)}

tests/manual/native_compute_profile.py

@@ -220,9 +220,15 @@ def prepare(args: argparse.Namespace) -> dict:
     # Reject numerical/runtime drift; a sparse timing-only cached main is allowed
     # because control and our current writer share the numeric model contract.
     differences = []
-    for name, expected in manifest["sourceHashes"].items():
+    recorded_sources = dict(manifest["sourceHashes"])
+    if manifest.get("cacheVersion", 1) >= 2:
+        headers = manifest.get("nativeHeaderHashes")
+        if not isinstance(headers, dict) or not headers:
+            raise RuntimeError("Cached build lacks native header hashes; rebuild before profiling")
+        recorded_sources.update(headers)
+    for name, expected in recorded_sources.items():
         relative = name.split("native/", 1)[-1]
-        current = ROOT / "native" / relative
+        current = cache / relative if relative == "model.c" else ROOT / "native" / relative
         if digest(current) != expected:
             differences.append(relative)
     if any(name != "runtime/main.c" for name in differences):
@@ -243,7 +249,16 @@ def prepare(args: argparse.Namespace) -> dict:
     for name in ("model.c", "model.h", "manifest.json"):
         shutil.copy2(cache / name, output / name)
     instrument(native)
-    command = [compiler, *manifest["compilerFlags"], "-I", str(output), "-I", str(native / "include"), "-I", str(sundials / "include"), str(output / "model.c"), *map(str, sorted(native.rglob("*.c"))), "-Wl,--start-group", *map(str, libraries), "-Wl,--end-group", "-lm", "-o", str(output / "profiled-model")]
+    # Match the cached translation units. Compiling the compatibility entry
+    # together with its new modules would define every component twice.
+    sources = [native / name.split("native/", 1)[-1]
+               for name in manifest["sourceHashes"]
+               if name.endswith(".c") and name != "model.c"]
+    if not sources or (native / "components/kernels.c" in sources and
+                       any(path.parent == native / "components/modules" for path in sources)):
+        raise RuntimeError("Cached translation units are missing or mix amalgamation and modules")
+    sources.append(native / "runtime/compute_profile.c")
+    command = [compiler, *manifest["compilerFlags"], "-I", str(output), "-I", str(native / "include"), "-I", str(sundials / "include"), str(output / "model.c"), *map(str, sources), "-Wl,--start-group", *map(str, libraries), "-Wl,--end-group", "-lm", "-o", str(output / "profiled-model")]
     prepared = {"controlExecutable": str(cache / "model"), "profiledExecutable": str(output / "profiled-model"), "buildCommand": command, "runtimeArguments": numerical_arguments, "verifyJacobian": "--verify-jacobian" in numerical_arguments, "algorithm": "production-automatic", "cacheDir": str(cache), "cachedMainDifference": differences, "stateCount": len(manifest["stateKeys"]), "modelSha256": digest(output / "model.c"), "compiler": compiler_version, "warmups": args.warmups, "repeats": args.repeats, "scope": "Inclusive times overlap. Exclusive scope times partition integration including instrumentation overhead. No component or libc sub-cost inference.", "sourceHashes": {str(p.relative_to(output)): digest(p) for p in sorted(native.rglob("*")) if p.is_file()}}
     write_json(output / "prepared.json", prepared)
     return prepared

tests/manual/profile_pipe_iterations.py

@@ -24,7 +24,9 @@ from app.simulation.backends import simulation_config
 from app.simulation.native_codegen import build as builder
 from app.simulation.native_codegen.compiler import compile_native_program
 from app.simulation.native_codegen.input import load_input
+from tests.manual.benchmark_native_pipe_solver import build_snapshot, pipe_source_path, working_native_paths
 
+SNAPSHOT_HELPER = ROOT / 'tests/manual/benchmark_native_pipe_solver.py'
 VARIANTS = ('guarded-newton', 'previous-newton', 'fixed-point')
 FIELDS = '''cache_requests cache_hits cache_misses flow_calls zero_pressure_calls
 pnl00r_analytic_calls resistance_calls scalar_analytic_calls iterative_calls
@@ -214,7 +216,7 @@ def instrument(source: str, variant: str) -> str:
     limit = '(kind==0?64:16)' if fixed else '80' if variant == 'previous-newton' else '128'
     prefix = PROFILE_PREFIX.replace('@VARIANT@', variant).replace('@FIXED@', str(int(fixed)))
     prefix = prefix.replace('@LIMIT@', limit).replace('@FIELDS@', ' '.join(f'X({name})' for name in FIELDS))
-    source = replace_once(source, '#include <stddef.h>', '#include <stddef.h>\n' + prefix)
+    source = replace_once(source, '#include <math.h>', '#include <math.h>\n' + prefix)
     # Exhaustion is marked at the actual loop fall-through, not inferred from
     # visiting the last allowed iteration (which can still converge).
     start = source.index('double native_pipe_resistance(')
@@ -259,6 +261,19 @@ def instrument(source: str, variant: str) -> str:
     return source
 
 
+
+def instrument_common(source: str) -> str:
+    """Count both ordinary and canonical Jacobian RHS evaluations exactly once."""
+    calls = [('model_eval', '    return model_eval(t,y,dy,w);')]
+    if 'int native_jacobian_rhs(' in source:
+        calls.append(('model_eval_jacobian', '    return model_eval_jacobian(t,y,dy,w);'))
+    for function, fragment in calls:
+        source = replace_once(source, fragment,
+            '    extern void pipe_profile_rhs_enter(void),pipe_profile_rhs_leave(void);\n'
+            f'    pipe_profile_rhs_enter();int ok={function}(t,y,dy,w);pipe_profile_rhs_leave();return ok;')
+    return source
+
+
 def git(*arguments: str) -> str:
     return subprocess.check_output(['git', *arguments], cwd=ROOT, text=True).strip()
 
@@ -268,11 +283,14 @@ def prepare(args, out: Path) -> dict:
         metadata = json.loads((out / 'prepared.json').read_text())
         if metadata['input_sha256'] != sha256(args.input.read_bytes()).hexdigest():
             raise ValueError('Prepared model no longer matches the input file')
-        if metadata['script_sha256'] != sha256(Path(__file__).read_bytes()).hexdigest():
+        if (metadata['script_sha256'] != sha256(Path(__file__).read_bytes()).hexdigest()
+                or metadata.get('snapshot_helper_sha256') != sha256(SNAPSHOT_HELPER.read_bytes()).hexdigest()):
             raise ValueError('Diagnostic helper changed; choose a fresh output directory')
         return metadata
+    if any((out / variant).exists() for variant in VARIANTS):
+        raise ValueError('Partially prepared variant directories exist; choose a fresh output directory')
     previous = git('rev-parse', args.previous_ref)
-    current = git('rev-parse', args.current_ref)
+    current = 'working-tree' if args.current_ref == 'working-tree' else git('rev-parse', args.current_ref)
     xml, doc = load_input(args.input)
     config = simulation_config(doc.simulation)
     if config.rtol != 1e-8 or config.t_start != 0 or config.t_stop != 10:
@@ -281,37 +299,40 @@ def prepare(args, out: Path) -> dict:
     out.mkdir(parents=True, exist_ok=True)
     (out / 'input.xml').write_bytes(xml)
     (out / 'input.json').write_bytes(args.input.read_bytes())
-    paths = git('ls-tree', '-r', '--name-only', current, 'native').splitlines()
     metadata = dict(input=str(args.input.resolve()), input_sha256=sha256(args.input.read_bytes()).hexdigest(),
         xml_sha256=sha256(xml).hexdigest(), script_sha256=sha256(Path(__file__).read_bytes()).hexdigest(),
+        snapshot_helper_sha256=sha256(SNAPSHOT_HELPER.read_bytes()).hexdigest(),
         previous_revision=previous, current_revision=current, settings=vars(config),
         sample_step=doc.simulation.sample_step, variants={},
         measurement_note='Instrumented times are diagnostic overhead and MUST NOT be used as production benchmark results.',
-        counter_scope='rhs counts native_rhs/model_eval calls including rejected trials and Jacobian differences; non_rhs includes initialization/output/probe evaluations.',
+        counter_scope='rhs counts native_rhs/model_eval and, where present, native_jacobian_rhs/model_eval_jacobian calls including rejected trials and Jacobian differences; non_rhs includes initialization/output/probe evaluations.',
         replay_scope='All guarded RHS resistance calls, including scalar analytic low-Re cases; no sampling or deduplication. Cache hits, zero pressure difference and direct PNL00R analytic calls are excluded and counted separately.',
         capture_format='Native-endian IEEE-754 binary64 records: base, diameter, length, relative roughness, den=pi*d*mu, kind (six doubles, 48 bytes). Replay on the same host.',
         residual_test='At actual returned q, independently factored long-double f(Re) evaluates abs(Re^2*f(Re)/K-1)<=1e-9. Separate from each algorithm stopping rule.')
     original_native = builder.NATIVE
+    prepared_builds = []  # Pin earlier variants while later variants are built.
     try:
         for variant in VARIANTS:
             directory = out / variant
             revision = current if variant == 'guarded-newton' else previous
+            paths = (working_native_paths() if revision == 'working-tree'
+                     else git('ls-tree', '-r', '--name-only', revision, 'native').splitlines())
             for name in paths:
                 target = directory / name
                 target.parent.mkdir(parents=True, exist_ok=True)
-                target.write_bytes(subprocess.check_output(['git', 'show', f'{revision}:{name}'], cwd=ROOT))
-            kernel = directory / 'native/components/kernels.c'
+                data = ((ROOT / name).read_bytes() if revision == 'working-tree'
+                        else subprocess.check_output(['git', 'show', f'{revision}:{name}'], cwd=ROOT))
+                target.write_bytes(data)
+            kernel = pipe_source_path(directory / 'native')
             original_hash = sha256(kernel.read_bytes()).hexdigest()
             kernel.write_text(instrument(kernel.read_text(), variant))
             common = directory / 'native/runtime/common.c'
-            common.write_text(replace_once(common.read_text(), '    return model_eval(t,y,dy,w);',
-                '''    extern void pipe_profile_rhs_enter(void),pipe_profile_rhs_leave(void);
-    pipe_profile_rhs_enter();int ok=model_eval(t,y,dy,w);pipe_profile_rhs_leave();return ok;'''))
+            common.write_text(instrument_common(common.read_text()))
             main = directory / 'native/runtime/main.c'
             main.write_text(replace_once(main.read_text(), 'int main(int argc, char **argv) {',
                 'int main(int argc, char **argv) {\n    extern void pipe_profile_install(void);pipe_profile_install();'))
-            builder.NATIVE = directory / 'native'
-            build = builder.build_native(program, cache_dir=out / 'cache')
+            build = build_snapshot(program, directory / 'native', out / 'cache', required_sources=(common, main))
+            prepared_builds.append(build)
             replay_source = directory / 'replay.c'
             replay_source.write_text(REPLAY_MAIN)
             replay = directory / ('replay.exe' if os.name == 'nt' else 'replay')
@@ -323,7 +344,8 @@ def prepare(args, out: Path) -> dict:
             if compiled.returncode:
                 raise RuntimeError(f'Replay compilation failed: {compiled.stderr}')
             metadata['variants'][variant] = dict(executable=str(build.executable), replay=str(replay),
-                build_key=build.manifest['buildKey'], original_kernel_sha256=original_hash,
+                build_key=build.manifest['buildKey'], pipe_source=str(kernel.relative_to(directory)),
+                compiled_source_hashes=build.manifest['sourceHashes'], original_kernel_sha256=original_hash,
                 instrumented_kernel_sha256=sha256(kernel.read_bytes()).hexdigest())
             print(f'Prepared {variant}', flush=True)
     finally:
@@ -423,7 +445,8 @@ def main():
     parser.add_argument('--input', type=Path, default=ROOT / 'tests/data/test-mql-8-corrected.json')
     parser.add_argument('--output-dir', type=Path, required=True)
     parser.add_argument('--previous-ref', default='5d5a2e1')
-    parser.add_argument('--current-ref', default='808c484')
+    parser.add_argument('--current-ref', default='808c484',
+                        help='Audited guarded-solver revision, or working-tree for the current modular sources')
     parser.add_argument('--timeout', type=float, default=120)
     parser.add_argument('--prepare-only', action='store_true')
     args = parser.parse_args()