Improve pytest parsing for beginner-friendly evaluation feedback

executor.js

@@ -2,6 +2,7 @@ const { exec, spawn } = require('child_process');
 const path = require('path');
 const fs = require('fs').promises;
 const { v4: uuidv4 } = require('uuid');
+const { parsePytestOutput } = require('./pytest-parser');
 
 /**
  * Ensures the datasets repo is cloned or updated in the unique directory.
@@ -237,145 +238,6 @@ ${testCode}
   }
 }
 
-/**
- * Parses pytest output to extract structured test results.
- * @param {string} stdout - The stdout from pytest.
- * @param {string} [stderr=''] - The stderr from pytest.
- * @param {number|null} [exitCode=null] - The pytest process exit code.
- * @returns {object} - The test summary.
- */
-function buildRawOutput(stdout = '', stderr = '') {
-  if (stdout && stderr) {
-    return `${stdout}\n${stderr}`;
-  }
-  return stdout || stderr || '';
-}
-
-function extractPytestSummary(output) {
-  const lines = output
-    .split(/\r?\n/)
-    .map((line) => line.trim())
-    .filter(Boolean);
-
-  const summaryLine = [...lines].reverse().find((line) => (
-    /^=+/.test(line) &&
-    /=+$/.test(line) &&
-    (/\bin [\d.]+s\b/.test(line) || /\bno tests ran\b/.test(line))
-  ));
-
-  if (!summaryLine) {
-    return '';
-  }
-
-  return summaryLine.replace(/^=+\s*/, '').replace(/\s*=+$/, '');
-}
-
-function extractPytestCount(summary, labelPattern) {
-  const match = summary.match(new RegExp(`(\\d+) ${labelPattern}\\b`));
-  return match ? parseInt(match[1], 10) : 0;
-}
-
-function extractPytestShortSummaryTarget(output, prefix) {
-  const line = output
-    .split(/\r?\n/)
-    .map((entry) => entry.trim())
-    .find((entry) => entry.startsWith(`${prefix} `));
-
-  return line ? line.slice(prefix.length + 1).trim() : '';
-}
-
-function extractPytestErrorMessage(output, stderr = '') {
-  const combined = buildRawOutput(output, stderr);
-  const patterns = [
-    /^E\s+([A-Za-z_.]+(?:Error|Exception): .+)$/m,
-    /^([A-Za-z_.]+(?:Error|Exception): .+)$/m,
-    /^(ImportError while importing test module .+)$/m,
-    /^E\s+(.+)$/m,
-  ];
-
-  for (const pattern of patterns) {
-    const match = combined.match(pattern);
-    if (match) {
-      return match[1].trim();
-    }
-  }
-
-  return 'Pytest error during collection or execution';
-}
-
-function parsePytestOutput(stdout = '', stderr = '', exitCode = null) {
-  const summary = extractPytestSummary(stdout);
-  const rawout = buildRawOutput(stdout, stderr);
-  const passed_tests = extractPytestCount(summary, 'passed');
-  const failed_tests = extractPytestCount(summary, 'failed');
-  const errors = extractPytestCount(summary, 'error(?:s)?');
-  const no_tests_collected = exitCode === 5 || /\bno tests ran\b/.test(summary);
-  const failures = [];
-
-  const failureBlocks = stdout.split(/={10,} FAILURES ={10,}/)[1]?.split(/={10,}/)[0] || '';
-  const matches = [...failureBlocks.matchAll(
-    /_{5,}\s*(.*?)\s*_{5,}[\s\S]*?>\s*assert\s+(.*?)\s*?\nE\s+assert\s+(.*?)\s*?(?:\nE\s+\+\s+where\s+(.*?)\s+=)?/g
-  )];
-
-  matches.forEach((match, index) => {
-    const test_case = match[1]?.trim() || `Test ${index + 1}`;
-    const assertionLine = match[2]?.trim();
-    const failedExpr = match[3]?.trim();
-    const evaluated = match[4]?.trim() || '';
-
-    failures.push({
-      test_case,
-      expected: failedExpr.split('==')[1]?.trim() || '',
-      received: evaluated || failedExpr.split('==')[0]?.trim(),
-      error_message: `Assertion failed: ${assertionLine}`,
-      rawout,
-    });
-  });
-
-  if (failed_tests > 0 && failures.length === 0) {
-    failures.push({
-      test_case: extractPytestShortSummaryTarget(stdout, 'FAILED') || 'pytest assertion failure',
-      expected: '',
-      received: '',
-      error_message: 'Pytest reported one or more failed assertions',
-      rawout,
-    });
-  }
-
-  let runtime_error = '';
-
-  if (errors > 0) {
-    runtime_error = extractPytestErrorMessage(stdout, stderr);
-    failures.push({
-      test_case: extractPytestShortSummaryTarget(stdout, 'ERROR') || 'pytest collection/execution',
-      expected: '',
-      received: '',
-      error_message: runtime_error,
-      rawout,
-    });
-  } else if (no_tests_collected) {
-    runtime_error = 'Pytest did not collect any tests';
-    failures.push({
-      test_case: 'pytest collection',
-      expected: 'at least 1 collected test',
-      received: '0 collected tests',
-      error_message: runtime_error,
-      rawout,
-    });
-  }
-
-  return {
-    tests_run: passed_tests + failed_tests,
-    passed: passed_tests,
-    failed: failed_tests,
-    errors,
-    no_tests_collected,
-    exit_code: exitCode,
-    failure_details: failures,
-    runtime_error,
-  };
-}
-
 function parseCppTestOutput(output, stdout = '', stderr = '') {
   output = output.toString();
   let total_tests = 0;

package.json

@@ -3,7 +3,7 @@
   "version": "1.0.0",
   "main": "index.js",
   "scripts": {
-    "test": "echo \"Error: no test specified\" && exit 1"
+    "test": "node --test"
   },
   "keywords": [],
   "author": "",