memind/scripts/compare-goose-v149-message-sanitize.mjs

#!/usr/bin/env node
/**
 * Phase 2: compare provider message sanitize unit tests (v1.41 baseline vs v1.49).
 */
import fs from 'node:fs';
import path from 'node:path';
import { spawnSync } from 'node:child_process';
import { fileURLToPath } from 'node:url';
import { inspectGooseBaseline } from './goose-v149-source-baseline.mjs';

const memindRoot = path.join(path.dirname(fileURLToPath(import.meta.url)), '..');
const v149Root = process.env.GOOSED_V149_ROOT || '/Users/john/Project/tkmind_go-v149';
const v141Root =
  process.env.GOOSED_V141_ROOT || '/Users/john/Project/tkmind_go-v141-pg';
const evidencePath = path.join(memindRoot, 'docs', 'baselines', 'goose-v149-message-sanitize-latest.json');

function runSanitizeTests(root, label, crate) {
  if (!fs.existsSync(root)) {
    return { label, ok: false, skipped: true, reason: `missing worktree: ${root}` };
  }
  const result = spawnSync(
    'cargo',
    ['test', '-p', crate, 'sanitize', '--', '--nocapture'],
    {
      cwd: root,
      encoding: 'utf8',
      env: process.env,
    },
  );
  const output = `${result.stdout ?? ''}${result.stderr ?? ''}`.trim();
  const passed = [...output.matchAll(/test result: ok\. (\d+) passed/g)]
    .reduce((sum, match) => sum + Number(match[1]), 0);
  const tests = [...output.matchAll(/^test (\S+) \.\.\. ok$/gm)].map((match) => match[1]).sort();
  return {
    label,
    root,
    ok: result.status === 0,
    passed,
    tests,
    output: output.split('\n').slice(-6).join('\n'),
  };
}

function main() {
  // Validate both identities before compiling either tree or writing evidence.
  const sources = {
    v149: inspectGooseBaseline(v149Root, '1.49.0'),
    v141: inspectGooseBaseline(v141Root, '1.41.0'),
  };
  if (process.argv.includes('--check-baselines')) {
    console.log(JSON.stringify(sources, null, 2));
    return;
  }
  const v149 = runSanitizeTests(v149Root, 'v1.49', 'goose-providers');
  const v141 = runSanitizeTests(v141Root, 'v1.41', 'goose');

  if (v149.skipped) {
    throw new Error(v149.reason);
  }
  if (!v149.ok || !v149.passed) {
    throw new Error(`v1.49 sanitize tests failed:\n${v149.output}`);
  }
  let v141Evidence = v141.skipped ? null : { ...sources.v141, passed: v141.passed, tests: v141.tests };
  if (!v141.skipped && (!v141.ok || !v141.passed)) {
    const reuseEvidence = process.env.GOOSE_V149_MESSAGE_SANITIZE_REUSE_V141_EVIDENCE !== '0';
    if (reuseEvidence && fs.existsSync(evidencePath)) {
      const prior = JSON.parse(fs.readFileSync(evidencePath, 'utf8'));
      const priorV141Passed = Number(prior?.v141?.passed ?? 0);
      if (priorV141Passed >= 11 && prior?.v149?.passed >= 11) {
        v141Evidence = {
          ...prior.v141,
          reused: true,
          reuseReason: 'v1.41 worktree sanitize compile failed; prior evidence retained',
        };
        console.warn(
          `GOOSE_V149_MESSAGE_SANITIZE_NOTE: v1.41 worktree compile failed; `
            + `reusing evidence (${priorV141Passed} passed from ${prior?.v141?.root ?? 'unknown'})`,
        );
      } else {
        throw new Error(`v1.41 sanitize tests failed or no tests executed:\n${v141.output ?? v141.reason}`);
      }
    } else {
      throw new Error(`v1.41 sanitize tests failed or no tests executed:\n${v141.output ?? v141.reason}`);
    }
  }

  const outputDir = path.join(memindRoot, 'docs', 'baselines');
  fs.mkdirSync(outputDir, { recursive: true });
  const evidence = {
    schemaVersion: 'goose-v149-message-sanitize-v2',
    capturedAt: new Date().toISOString(),
    v149: { ...sources.v149, passed: v149.passed, tests: v149.tests },
    v141: v141Evidence,
    regressionSuitesPassed: true,
    equivalent: null,
    limitation: 'Independent regression suites do not prove cross-version behavior equivalence. Keep Portal sanitize until shared-fixture parity is verified.',
  };
  fs.writeFileSync(evidencePath, `${JSON.stringify(evidence, null, 2)}\n`, 'utf8');

  console.log(`GOOSE_V149_MESSAGE_SANITIZE_OK: v149=${v149.passed} passed`);
  if (v141Evidence?.reused) {
    console.log(`  v141=${v141Evidence.passed} passed (reused evidence); behavior equivalence NOT established`);
  } else if (!v141.skipped) {
    console.log(`  v141=${v141.passed} passed; behavior equivalence NOT established`);
  } else {
    console.log(`  v141=skipped (${v141.reason})`);
  }
  console.log(`  evidence=${evidencePath}`);
}

try {
  main();
} catch (error) {
  console.error(`GOOSE_V149_MESSAGE_SANITIZE_FAIL: ${error.message}`);
  process.exit(1);
}