From 9c773c493c29155a138e64d8d6676bed2b807dc3 Mon Sep 17 00:00:00 2001 From: multipleof4 Date: Mon, 13 Oct 2025 05:56:09 -0700 Subject: [PATCH] Refactor: Generate indented text block for README --- scripts/run-benchmark.js | 48 ++++++++++++++++++++++------------------ 1 file changed, 27 insertions(+), 21 deletions(-) diff --git a/scripts/run-benchmark.js b/scripts/run-benchmark.js index cecac0c..1f371e8 100644 --- a/scripts/run-benchmark.js +++ b/scripts/run-benchmark.js @@ -40,43 +40,49 @@ const main = async () => { const readme = await fs.readFile(README_PATH, 'utf-8'); const models = readme.match(/\n([\s\S]+?)\n/)[1].trim().split('\n'); const percentage = parseInt(readme.match(/RUN_PERCENTAGE:\s*(\d+)/)?.[1] ?? '100', 10); - const allTestDirs = (await fs.readdir(TESTS_DIR, { withFileTypes: true })) .filter(d => d.isDirectory()).map(d => d.name).sort(); - const testsToRun = allTestDirs.slice(0, Math.ceil(allTestDirs.length * (percentage / 100))); - let resultsTable = '| Model | ' + allTestDirs.join(' | ') + ' |\n'; - resultsTable += '|' + ' --- |'.repeat(allTestDirs.length + 1) + '\n'; - + const resultsData = []; for (const model of models) { - resultsTable += `| ${model} |`; + const row = { Model: model }; for (const dir of allTestDirs) { if (!testsToRun.includes(dir)) { - resultsTable += ' ⚪ Not Run |'; + row[dir] = '⚪ Not Run'; continue; } - const { prompt, harness } = require(path.join(TESTS_DIR, dir, 'test.js')); console.log(`Running ${dir} for ${model}...`); const llmCode = await getLlmCode(prompt, model); - - if (!llmCode) { - resultsTable += ' ❌ API Error |'; - continue; + if (llmCode) { + const outDir = path.join(TESTS_DIR, dir, 'outputs'); + await fs.mkdir(outDir, { recursive: true }); + const fname = `${model.replace(/[\/:]/g, '_')}_${new Date().toISOString().replace(/:/g, '-')}.js`; + await fs.writeFile(path.join(outDir, fname), llmCode); + row[dir] = (await runTest(`${llmCode}\n${harness}`)) ? '✅ Pass' : '❌ Fail'; + } else { + row[dir] = '❌ API Error'; } - - const outDir = path.join(TESTS_DIR, dir, 'outputs'); - await fs.mkdir(outDir, { recursive: true }); - const fname = `${model.replace(/[\/:]/g, '_')}_${new Date().toISOString().replace(/:/g, '-')}.js`; - await fs.writeFile(path.join(outDir, fname), llmCode); - - const passed = await runTest(`${llmCode}\n${harness}`); - resultsTable += ` ${passed ? '✅ Pass' : '❌ Fail'} |`; } - resultsTable += '\n'; + resultsData.push(row); } + const headers = ['Model', ...allTestDirs]; + const colWidths = Object.fromEntries(headers.map(h => + [h, Math.max(h.length, ...resultsData.map(r => r[h]?.length ?? 0))] + )); + + const formatRow = (data, padChar = ' ') => + headers.map(h => (data[h] ?? '').padEnd(colWidths[h], padChar)).join(' | '); + + const headerObj = Object.fromEntries(headers.map(h => [h,h])); + const resultsTable = [ + formatRow(headerObj), + formatRow({}, '-'), + ...resultsData.map(row => formatRow(row)) + ].map(line => ' ' + line).join('\n'); + const newReadme = readme.replace( /[\s\S]*/, `\n${resultsTable}\n`