Hmbown/CodeWhale · error · RuntimeContractError

measurement emitted invalid JSON: {error}

Error message

measurement emitted invalid JSON: {error}

What it means

After a successful measurement, run_measurement() parses the child's entire stdout as JSON to obtain the receipt. If json.loads raises, the raw stdout contained something non-JSON - the receipt printer's output was polluted or truncated. The JSONDecodeError detail (line/column) is embedded in the message to point at the offending byte.

Source

Thrown at scripts/check-runtime-contract-budget.py:411

    env["CARGO_NET_OFFLINE"] = "true"
    proc = subprocess.run(
        [sys.executable, str(MEASURE_SCRIPT)],
        cwd=REPO_ROOT,
        env=env,
        capture_output=True,
        text=True,
        check=False,
    )
    sys.stderr.write(proc.stderr)
    if proc.returncode != 0:
        sys.stdout.write(proc.stdout)
        raise RuntimeContractError(
            f"runtime-contract measurement failed with exit code {proc.returncode}"
        )
    try:
        receipt = json.loads(proc.stdout)
    except json.JSONDecodeError as error:
        raise RuntimeContractError(f"measurement emitted invalid JSON: {error}") from error
    if not isinstance(receipt, dict):
        raise RuntimeContractError("measurement top level must be an object")
    validate_receipt(receipt)
    return receipt


def update_command(receipt_path: Path | None, budget_path: Path) -> str:
    parts = ["python3", "scripts/check-runtime-contract-budget.py"]
    if receipt_path is not None:
        parts.extend(["--receipt", str(receipt_path)])
    if budget_path != BUDGET_PATH:
        parts.extend(["--budget", str(budget_path)])
    parts.append("--update")
    return shlex.join(parts)


FRAGMENT_MODULE = REPO_ROOT / "crates" / "core" / "src" / "fragments.rs"
FRAGMENT_MAX_TOKENS_CEILING = 10_000

View on GitHub (pinned to 8880682c63)

Solutions

1. Reproduce and capture: python3 scripts/measure-runtime-contract.py > out.json and inspect around the reported line/column
2. Route every diagnostic in the measure path to stderr: print(..., file=sys.stderr) in Python, eprintln! in Rust
3. Pipe the stdout through python3 -m json.tool as a quick validity check before wiring changes back into the gate

Example fix

# before (measure path)
print(f"measuring {name}")  # pollutes stdout, breaks json.loads

# after
import sys
print(f"measuring {name}", file=sys.stderr)

import json, subprocess, sys


def measurement_stdout_is_json() -> bool:
    proc = subprocess.run(
        [sys.executable, "scripts/measure-runtime-contract.py"],
        capture_output=True,
        text=True,
    )
    if proc.returncode != 0:
        return False
    try:
        json.loads(proc.stdout)
        return True
    except json.JSONDecodeError:
        return False

try:
    receipt = mod.run_measurement()
except mod.RuntimeContractError as err:
    if "invalid JSON" in str(err):
        import json
        proc = subprocess.run(
            [sys.executable, "scripts/measure-runtime-contract.py"],
            capture_output=True,
            text=True,
        )
        try:
            json.loads(proc.stdout)
        except json.JSONDecodeError as e:
            print("offending stdout near:", repr(proc.stdout[max(0, e.pos - 80): e.pos + 80]))
    raise SystemExit(2)

When it happens

Trigger: The --nocapture test or the measure script printing diagnostics to stdout (println!/print) instead of stderr; a panic message interleaved into stdout; partial output when the child dies mid-write while still exiting 0; progress bars or cargo notes leaking into the captured pipe.

Common situations: A developer adds a debug println! in the metric test and the gate suddenly fails locally; a dependency upgrades and starts writing to stdout; running with a wrapper (rustfilt, backtrace pretty-printers) that injects text.

Understand the failure class

• Parsing and encoding errors: unexpected token, malformed input — why parsers reject input and how to find the real culprit.

Related errors

• measurement emitted invalid JSON: {error}
• baseline provenance build profile/sample count changed
• baseline receipt {field} does not match baseline_observation
• baseline receipt {field} does not match baseline provenance
• measurement command failed

AI-assisted analysis of Hmbown/CodeWhale@8880682c63 (2026-08-16). Data as JSON: /api/errors/d5811f5eafd4ea60. Report an issue: GitHub.