// 索引截断顺序回归测试(issue #10 实测反馈)。 // // 核心事实(本测试锁定):`git ls-files --cached --others` 的输出**不是全局字典序** —— // 未跟踪文件先按 readdir 顺序输出,`node_modules/` 这类目录可能占满前 N 行,把 `head` // 的配额吃光,导致 `AGENTS.md`、`src/**` 等真实文件被整段切掉。 // 因此排除必须发生在 `head` **之前**(远端 grep -vE),而不是数据到本地之后。 // // 三层: // A. 命令结构断言(grep 必须在 head 之前、正则来自单一事实来源、完整路径段匹配) // B. **真实端到端**:临时 git 仓库(已跟踪 AGENTS.md/src + 未跟踪大量 node_modules) // → 用真实代码生成的命令跑真实管道 → 断言旧行为会切掉、新行为不会 // C. 解析器双保险(客户端 excludedSegment 仍生效) import { readFileSync, writeFileSync, mkdirSync, rmSync, mkdtempSync, existsSync } from 'node:fs' import { execFileSync } from 'node:child_process' import { tmpdir } from 'node:os' import { join } from 'node:path' /** 可用的 POSIX shell(优先 Git for Windows 自带 bash;WSL 的 bash 视图不同,跳过)。 */ function findPosixShell() { if (process.platform !== 'win32') return existsSync('/bin/bash') ? '/bin/bash' : undefined const candidates = [ process.env.DSH_TEST_BASH, 'E:\\Program Files\\Git\\bin\\bash.exe', 'C:\\Program Files\\Git\\bin\\bash.exe', 'E:\\Program Files\\Git\\usr\\bin\\bash.exe', 'C:\\Program Files\\Git\\usr\\bin\\bash.exe', ].filter(Boolean) return candidates.find((p) => existsSync(p)) } const root = new URL('..', import.meta.url).pathname.replace(/^\/([A-Za-z]:)/, '$1') const src = readFileSync(root + '/lib/index.js', 'utf8') function grab(name) { const start = src.indexOf(name) if (start < 0) throw new Error(name + ' not found') const declStart = src.lastIndexOf('\n', start) let i = src.indexOf('{', start) let depth = 0 for (; i < src.length; i++) { if (src[i] === '{') depth++ else if (src[i] === '}') { depth--; if (depth === 0) break } } return src.slice(declStart + 1, i + 1) } const pick = (re) => { const m = src.match(re); if (!m) throw new Error('missing ' + re); return m[0] } const api = new Function([ pick(/const REF_MAX_RESULTS = \d+;/), pick(/const REF_MAX_ENTRIES = \d+;/), pick(/const REF_INDEX_FIND_MAXDEPTH = \d+;/), pick(/const REF_INDEX_GIT_FULL_BUDGET_SEC = \d+;/), pick(/const REF_INDEX_GIT_CACHED_BUDGET_SEC = \d+;/), pick(/const REF_INDEX_FIND_BUDGET_SEC = \d+;/), pick(/const REF_INDEX_BUILD_TIMEOUT_MS = \d+;/), pick(/const REF_EXCLUDED_DIRS = new Set\(\[[\s\S]*?\]\);/), grab('function refExcludeRegex'), grab('function refIndexCommand'), grab('function refParseIndexOutput'), grab('function shellQuote'), grab('function shellQuotePath'), ].join('\n') + '\nreturn { refIndexCommand, refExcludeRegex, refParseIndexOutput, REF_MAX_ENTRIES, REF_EXCLUDED_DIRS };')() let pass = 0, fail = 0 const check = (cond, label) => { if (cond) { pass++; console.log(' ✓ ' + label) } else { fail++; console.log(' ✗ FAIL: ' + label) } } console.log('A · 命令结构(排除必须在 head 之前)') { const cmd = api.refIndexCommand('~/proj') const gitLine = cmd.split('\n').find((l) => l.includes('git ls-files')) const iGrep = gitLine.indexOf('grep -vE') const iHead = gitLine.indexOf('head -n') // sed 已移到输出行(printf '%s\n' "$G" | sed 's/^/f\t/'),故只在 git 行内校验「排除在截断之前」 check(iGrep > 0 && iHead > 0 && iGrep < iHead, 'git 分支:grep -vE 在 head 之前(先排除后截断)') check(cmd.split('\n').some((l) => l.includes("| sed 's/^/f\t/'")), 'git 结果由 printf|sed 加帧输出') const re = api.refExcludeRegex() check(re === "(^|/)(\\.git|node_modules|dist|build|out|coverage|target|\\.next|\\.nuxt|\\.turbo|\\.venv|__pycache__|\\.pytest_cache|\\.mypy_cache|\\.gradle)(/|$)", '排除正则由 REF_EXCLUDED_DIRS 生成(单一事实来源)') for (const d of api.REF_EXCLUDED_DIRS) check(re.includes(d.replace(/\./g, '\\.')), `正则包含 ${d}`) check(new RegExp(re).test('node_modules/x.js') && new RegExp(re).test('a/node_modules/b'), '匹配任意层级的排除目录') check(!new RegExp(re).test('distribution/x.js'), '不误伤前缀相同的普通目录(distribution)') check(!new RegExp(re).test('src/node_modules_helper.js'), '不误伤名字前缀相同的文件') const findLine = cmd.split('\n').find((l) => l.includes('find .')) check(findLine.includes('-prune') && findLine.indexOf('-prune') < findLine.indexOf('head -n'), 'find 分支同样先剪枝后截断') } console.log('B · 真实端到端(临时 git 仓库 + 真实管道)') { const shell = findPosixShell() if (!shell) { console.log(' ⚠ 未找到 POSIX shell(Git for Windows bash / /bin/bash),跳过端到端段(其余断言仍有效)') } else { const dir = mkdtempSync(join(tmpdir(), 'rssh-idx-')) const git = (args, opts) => execFileSync('git', args, { cwd: dir, stdio: ['ignore', 'pipe', 'pipe'], encoding: 'utf8', ...opts }) try { git(['init', '-q']) git(['config', 'user.email', 't@t']) git(['config', 'user.name', 't']) // 已跟踪:AGENTS.md + src/** writeFileSync(join(dir, 'AGENTS.md'), 'x') mkdirSync(join(dir, 'src')) for (let i = 0; i < 40; i++) writeFileSync(join(dir, 'src', `m${i}.jl`), 'x') git(['add', '-A']) git(['commit', '-qm', 'init']) // 未跟踪:大量 node_modules(不写 .gitignore,模拟报告者的仓库) for (let i = 0; i < 400; i++) { const d = join(dir, 'node_modules', `pkg${i}`) mkdirSync(d, { recursive: true }) writeFileSync(join(d, 'index.js'), 'x') } const all = git(['ls-files', '--cached', '--others', '--exclude-standard']).split('\n').filter(Boolean) const agentsLine = all.indexOf('AGENTS.md') + 1 check(all[0].startsWith('node_modules/'), '真实输出:未跟踪的 node_modules 排在最前(非全局字典序)') check(agentsLine > all.length * 0.5, `真实输出:AGENTS.md 落到第 ${agentsLine}/${all.length} 行(字典序假设不成立)`) // 用**真实代码生成**的命令,但把上限压到 200 以便在夹具规模下复现截断 const realCmd = api.refIndexCommand(dir) const withCap = (cap) => realCmd.replace(/head -n \d+/g, 'head -n ' + cap) const oldCmd = withCap(200).replace(" | grep -vE '" + api.refExcludeRegex() + "'", '') // 旧行为:无 grep const runSh = (script) => execFileSync(shell, ['-c', script], { encoding: 'utf8', stdio: ['ignore', 'pipe', 'pipe'] }) const oldOut = runSh(oldCmd).split('\n').filter((l) => l.includes('\t')) const newOut = runSh(withCap(200)).split('\n').filter((l) => l.includes('\t')) const hasAgents = (lines) => lines.some((l) => l.endsWith('\tAGENTS.md')) const srcCount = (lines) => lines.filter((l) => /\tsrc\//.test(l)).length const nmCount = (lines) => lines.filter((l) => /\tnode_modules\//.test(l)).length check(!hasAgents(oldOut) && nmCount(oldOut) > 0, `旧行为(先截断):AGENTS.md 被切掉,前 ${oldOut.length} 行几乎全是 node_modules(${nmCount(oldOut)} 条)`) check(hasAgents(newOut), '新行为(先排除):AGENTS.md 存活') check(srcCount(newOut) === 40, `新行为:src/ 全部 ${srcCount(newOut)} 条保留`) check(nmCount(newOut) === 0, '新行为:node_modules 零残留') check(newOut.length < oldOut.length, `新行为输出更小(${newOut.length} < ${oldOut.length} 行)`) // 解析器仍能吃下新输出,且客户端排除作为双保险 const parsed = api.refParseIndexOutput(newOut.join('\n')) check(parsed.some((x) => x.path === 'AGENTS.md' && x.kind === 'file'), '解析器:AGENTS.md 进入索引') check(parsed.some((x) => x.path === 'src' && x.kind === 'directory'), '解析器:合成出 src 目录') check(!parsed.some((x) => x.path.startsWith('node_modules')), '解析器:node_modules 不在索引中') } finally { rmSync(dir, { recursive: true, force: true }) } } } console.log('C · 解析器双保险(客户端排除仍生效)') { const e = api.refParseIndexOutput(['f\tnode_modules/a.js', 'f\tsrc/b.js', 'f\tdist/c.js', 'f\tAGENTS.md'].join('\n')) check(!e.some((x) => x.path.startsWith('node_modules')) && !e.some((x) => x.path === 'dist'), '即便远端漏掉排除,客户端也会过滤') check(e.some((x) => x.path === 'AGENTS.md'), '真实文件不受影响') } console.log('D · 截断告警接线(建议 1)') { check(/rawLines > REF_MAX_ENTRIES/.test(src), '按上限 +1 行取样并检测截断') check(/refTruncationWarned\.add\(key\)/.test(src), '每个 workspace 只告警一次(避免刷日志)') check(/文件引用索引已达上限/.test(src) && /@ 补全可能漏掉部分文件/.test(src), '告警文案说明影响与缓解方式') check(/const refTruncationWarned = new Set\(\)/.test(src), '告警状态集合已声明') } console.log(`\n结果: ${pass} 通过, ${fail} 失败`) process.exit(fail ? 1 : 0)