horizon-airdef/tools/test_command_exec_parity.py

#!/usr/bin/env python3
"""Compare Horizon's command execution with vanilla's real command implementations.

Java (VanillaCommandExecOracle) starts a stand-in integrated server: a real
MinecraftServer with a generated world, Commands.CommandSelection.INTEGRATED
and one single-player owner named Player. Rust runs the same script through
the Horizon client headless (HORIZON_COMMAND_SCRIPT). Both run each command as
the player through Commands.performPrefixedCommand and record every system or
chat message the player receives (text and resolved style: colour, bold,
italic, underlined, strikethrough, obfuscated) together with the results the
command reported to the player's CommandResultCallback. World, entity and item
changes are observed through probe commands whose feedback reports them
(execute if, clear with a zero count, data get, scoreboard queries...).
Messages that depend on time or a random UUID are compared after masking.
"""
import argparse
import json
import os
from pathlib import Path
import platform
import re
import shutil
import subprocess
import sys

try:
    from . import lab, test_rust_parity as base
    from . import command_exec_corpus as corpus_module
except ImportError:
    import lab
    import test_rust_parity as base
    import command_exec_corpus as corpus_module

SEED = 1


def main():
    parser = argparse.ArgumentParser(description=__doc__)
    parser.add_argument('--prepare', action='store_true')
    parser.add_argument('--limit', type=int, default=40, help='print at most this many mismatches')
    parser.add_argument('--allow-gaps', action='store_true', help='report mismatches without failing')
    parser.add_argument('--only', help='run only corpus groups whose name contains this')
    args = parser.parse_args()
    build = lab.LOCAL / 'command-exec-parity'
    build.mkdir(parents=True, exist_ok=True)
    report_path = build / 'report.json'
    report_path.unlink(missing_ok=True)
    if args.prepare:
        lab.prepare()
    lock, jars, libraries = base.verify_artifacts()
    oracle = lab.ROOT / 'mods/horizon/src/test/java/dev/horizonlab/parity/VanillaCommandExecOracle.java'
    classes = build / 'classes'
    classes.mkdir(exist_ok=True)
    dependency_path = os.pathsep.join(map(str, libraries))

    def run(command, name, **kw):
        with (build / (name + '.log')).open('w', encoding='utf-8') as log:
            result = subprocess.run(list(map(str, command)), stdout=log, stderr=subprocess.STDOUT, **kw)
        if result.returncode:
            tail = (build / (name + '.log')).read_text(errors='replace')[-3000:]
            sys.exit(f'{name} failed ({result.returncode}):\n{tail}')

    groups = corpus_module.groups()
    if args.only:
        groups = [g for g in groups if args.only in g[0] or g[0] == 'setup']
    script = []
    for _, lines in groups:
        script.extend(lines)
    script_path = build / 'script.jsonl'
    script_path.write_text(''.join(json.dumps(op, ensure_ascii=False) + '\n' for op in script), encoding='utf-8')

    run(['javac', '--release', lock['minecraft']['java_major'], '-proc:none', '-cp', str(jars['server']) + os.pathsep + dependency_path, '-d', classes, oracle], 'compile-java')
    vanilla_dir = build / 'vanilla'
    shutil.rmtree(vanilla_dir, ignore_errors=True)
    vanilla_dir.mkdir()
    run(['java', '-Xmx3G', '-Duser.timezone=UTC', '-cp', os.pathsep.join([str(classes), str(jars['server']), dependency_path]),
         'dev.horizonlab.parity.VanillaCommandExecOracle', vanilla_dir / 'universe', script_path, build / 'vanilla.jsonl', SEED], 'java', cwd=vanilla_dir)

    run(['cargo', 'build', '--release', '-p', 'horizon-client', '--bin', 'horizon-client'], 'compile-rust', cwd=lab.ROOT)
    metadata = json.loads(subprocess.check_output(['cargo', 'metadata', '--no-deps', '--format-version', '1'], cwd=lab.ROOT, text=True))
    binary = Path(metadata['target_directory']) / 'release' / ('horizon-client.exe' if os.name == 'nt' else 'horizon-client')
    rust_dir = build / 'rust'
    shutil.rmtree(rust_dir, ignore_errors=True)
    rust_dir.mkdir()
    env = dict(os.environ, HORIZON_NO_AUDIO='1', HORIZON_COMMAND_SCRIPT=str(script_path), HORIZON_COMMAND_OUTPUT=str(build / 'rust.jsonl'))
    run([binary, '--resources', lab.LOCAL / 'horizon-client/26.3', '--world', 'harness', '--seed', SEED, '--radius', 4], 'rust', cwd=rust_dir, env=env)

    vanilla = [json.loads(l) for l in (build / 'vanilla.jsonl').open(encoding='utf-8')]
    rust = [json.loads(l) for l in (build / 'rust.jsonl').open(encoding='utf-8')]
    if len(vanilla) != len(rust):
        sys.exit(f'record count differs: vanilla {len(vanilla)} rust {len(rust)}')
    commands_by_group = []
    index = 0
    for name, lines in groups:
        n = sum(1 for op in lines if (op['op'] == 'cmd' and op.get('record', True)) or (op['op'] == 'tick' and op.get('record')))
        commands_by_group.append((name, index, index + n))
        index += n
    mismatches = []
    for v, r in zip(vanilla, rust):
        assert v['cmd'] == r['cmd'], (v['cmd'], r['cmd'])
        kind = compare(v, r)
        if kind:
            mismatches.append({'kind': kind, 'cmd': v['cmd'], 'vanilla': v, 'rust': r})
    (build / 'mismatches.jsonl').write_text(''.join(json.dumps(m, ensure_ascii=False) + '\n' for m in mismatches), encoding='utf-8')
    summary = []
    bad = {m['cmd'] for m in mismatches}
    for name, lo, hi in commands_by_group:
        failed = sum(1 for rec in vanilla[lo:hi] if rec['cmd'] in bad)
        summary.append({'group': name, 'commands': hi - lo, 'mismatches': failed})
        print(f'{name}: {hi - lo - failed}/{hi - lo} commands match', flush=True)
    for m in mismatches[:args.limit]:
        print(m['kind'], repr(m['cmd']))
        print('  vanilla', render(m['vanilla'])[:400])
        print('  rust   ', render(m['rust'])[:400])
    total = len(vanilla)
    print(f'total: {total - len(mismatches)}/{total} commands match', flush=True)
    report = {'schema': 1, 'minecraft': lock['minecraft']['version'], 'platform': platform.platform(), 'seed': SEED,
              'commands': total, 'mismatches': len(mismatches), 'groups': summary,
              'scope': 'Feedback and failure messages (text and resolved style), chat lines, and CommandResultCallback results of commands run by the single-player owner of an integrated server; world, entity and item effects through probe commands.'}
    lab.write_json(report_path, report)
    if mismatches and not args.allow_gaps:
        sys.exit(f'{len(mismatches)} command execution mismatches; see {build}')
    print(f'Command execution parity report: {report_path}')


MASKS = [
    (re.compile(r'debug-trace-\d{4}-\d\d-\d\d_\d\d\.\d\d\.\d\d\.txt'), 'debug-trace-<time>.txt'),
    (re.compile(r'after [0-9.]+ second\(s\) and \d+ tick\(s\) \([0-9.,NaInfity]+ tick\(s\) per second\)'), 'after <time>'),
    (re.compile(r'Average time per tick: [0-9.]+ms'), 'Average time per tick: <ms>'),
    (re.compile(r'P50: [0-9.]+ms P95: [0-9.]+ms P99: [0-9.]+ms'), 'P50: <ms> P95: <ms> P99: <ms>'),
    (re.compile(r'debug/client-[0-9_.-]+\.(jfr|json)'), 'debug/client-<time>'),
]


def masked(record):
    text = json.dumps(record['messages'], ensure_ascii=False)
    for pattern, replacement in MASKS:
        text = pattern.sub(replacement, text)
    return text


def compare(v, r):
    if masked(v) != masked(r):
        return 'messages'
    if v['results'] != r['results']:
        if v['cmd'].startswith('/debug stop'):
            return None
        return 'results'
    return None


def render(record):
    lines = [''.join(span[0] for span in m) for m in record['messages']]
    return json.dumps({'messages': lines, 'styles': [[span[1] for span in m] for m in record['messages']], 'results': record['results']}, ensure_ascii=False)


if __name__ == '__main__':
    main()