@bevel-software/platform-core-backend 0.32.0 → 0.34.0

package/src/modules/workspace/__tests__/ripgrep-search.test.ts

@@ -0,0 +1,570 @@
+import { execFile } from 'node:child_process';
+import { chmod, mkdir, mkdtemp, readFile, realpath, rm, symlink, writeFile } from 'node:fs/promises';
+import { tmpdir } from 'node:os';
+import { delimiter, join } from 'node:path';
+import { promisify } from 'node:util';
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest';
+import {
+  compareSearchPaths,
+  RipgrepPatternMatcher,
+  type FileMatches,
+  type IPatternSearch,
+  type RipgrepPatternMatcherOptions,
+  type TreeSearchOptions,
+} from '../index.js';
+import { slowRipgrep } from '../../../__tests__/slow-ripgrep.js';
+
+/**
+ * The ripgrep runner through its contract, against the real `rg` on PATH:
+ * what a walk covers and in which order, how its JSON is read, how a pattern
+ * the engine refuses is answered, and that every run ends — at the budget, on
+ * an early stop, when the caller goes away, or on `close()`.
+ */
+let dir = '';
+let matchers: RipgrepPatternMatcher[] = [];
+
+function matcher(opts?: RipgrepPatternMatcherOptions): RipgrepPatternMatcher {
+  const m = new RipgrepPatternMatcher(opts);
+  matchers.push(m);
+  return m;
+}
+
+const search = (pattern: string, opts?: RipgrepPatternMatcherOptions, ignoreCase = false): Promise<IPatternSearch> =>
+  matcher(opts).search(pattern, { ignoreCase });
+
+/**
+ * Every listed file searched, as grep's text pass does: the listing first,
+ * then the files it named, less those with an `exclude` extension or named in
+ * `skip`.
+ */
+async function all(
+  s: IPatternSearch,
+  exclude: string[] = [],
+  opts: Partial<Pick<TreeSearchOptions, 'maxLinesPerFile'>> & { root?: string; skip?: string[] } = {},
+): Promise<FileMatches[]> {
+  const { root = 'kb', skip = [], maxLinesPerFile = 1_000 } = opts;
+  const scope = { cwd: root === '.' ? join(dir, 'kb') : dir, root };
+  const only = (await s.list(scope)).filter((p) => !exclude.some((ext) => p.toLowerCase().endsWith(ext)) && !skip.includes(p));
+  const out: FileMatches[] = [];
+  for await (const f of s.files(scope, { only, maxLinesPerFile })) out.push(f);
+  return out;
+}
+
+async function put(rel: string, content: string | Buffer): Promise<void> {
+  const p = join(dir, rel);
+  await mkdir(join(p, '..'), { recursive: true });
+  await writeFile(p, content);
+}
+
+/**
+ * The real `rg`, put first on PATH behind a script that records where each
+ * tree search starts (what follows its `--`); `last()` answers the latest.
+ */
+async function recordWalks(): Promise<{ last(): Promise<string[]>; runs(): Promise<string[][]>; remove(): Promise<void> }> {
+  const real = (await promisify(execFile)('sh', ['-c', 'command -v rg'])).stdout.trim();
+  const shim = await mkdtemp(join(tmpdir(), 'rec-rg-'));
+  const log = join(shim, 'walks');
+  await writeFile(
+    join(shim, 'rg'),
+    [
+      '#!/bin/sh',
+      'seen=; roots=; prev=; depth=',
+      'for a in "$@"; do if [ -n "$seen" ]; then roots="$roots$a$depth|"; elif [ "$a" = "--" ]; then seen=1; elif [ "$prev" = "--max-depth" ]; then depth="@$a"; fi; prev="$a"; done',
+      `if [ -n "$seen" ]; then echo "$roots" >> '${log}'; fi`,
+      `exec '${real}' "$@"`,
+      '',
+    ].join('\n'),
+  );
+  await chmod(join(shim, 'rg'), 0o755);
+  const path = process.env.PATH;
+  process.env.PATH = `${shim}${delimiter}${path ?? ''}`;
+  return {
+    async last() {
+      const lines = (await readFile(log, 'utf8')).split('\n').filter(Boolean);
+      return lines[lines.length - 1]!.split('|').filter(Boolean);
+    },
+    // Every run since the last call, each its roots (`@1`: its own files only).
+    async runs() {
+      const lines = (await readFile(log, 'utf8').catch(() => '')).split('\n').filter(Boolean);
+      await writeFile(log, '');
+      return lines.map((line) => line.split('|').filter(Boolean));
+    },
+    async remove() {
+      process.env.PATH = path;
+      await rm(shim, { recursive: true, force: true });
+    },
+  };
+}
+
+beforeEach(async () => {
+  dir = await realpath(await mkdtemp(join(tmpdir(), 'rg-search-')));
+});
+afterEach(async () => {
+  await Promise.all(matchers.map((m) => m.close()));
+  matchers = [];
+  await rm(dir, { recursive: true, force: true });
+});
+
+describe('RipgrepPatternMatcher', () => {
+  it('walks in path order, segment by segment, and reads each file\'s lines as ripgrep printed them', async () => {
+    await put('kb/b.md', 'x\nneedle b\r\n');
+    await put('kb/a/z.md', 'needle z\n');
+    await put('kb/a.md', 'needle a\n');
+    await put('kb/with space/ü.md', 'needle ü\n');
+    const files = await all(await search('needle'));
+    expect(files).toEqual([
+      { path: 'a/z.md', lines: [{ line: 1, text: 'needle z' }] },
+      { path: 'a.md', lines: [{ line: 1, text: 'needle a' }] },
+      // A CRLF line keeps its `\r`, as `split('\n')` of the file would.
+      { path: 'b.md', lines: [{ line: 2, text: 'needle b\r' }] },
+      { path: 'with space/ü.md', lines: [{ line: 1, text: 'needle ü' }] },
+    ]);
+    const paths = files.map((f) => f.path);
+    expect([...paths].sort(compareSearchPaths)).toEqual(paths);
+  });
+
+  it('leaves out binary files, lines that are not UTF-8, excluded extensions and .git in any case', async () => {
+    await put('kb/bin.dat', Buffer.concat([Buffer.from('needle\n'), Buffer.from([0])]));
+    // NUL far past the first buffer: what ripgrep matched before it is dropped too.
+    await put('kb/late-bin.dat', Buffer.concat([Buffer.from('needle\n'), Buffer.alloc(200_000, 'x'), Buffer.from([0])]));
+    await put('kb/latin.txt', Buffer.concat([Buffer.from('needle \xff\n', 'latin1'), Buffer.from('needle ok\n')]));
+    await put('kb/doc.DOCX', 'needle\n');
+    await put('kb/.git/x', 'needle\n');
+    await put('kb/sub/.Git/x', 'needle\n');
+    const files = await all(await search('needle'), ['.docx']);
+    expect(files).toEqual([{ path: 'latin.txt', lines: [{ line: 2, text: 'needle ok' }] }]);
+  });
+
+  it('reads a match on one very long line whole, however many chunks its record spans', async () => {
+    const long = `needle ${'é'.repeat(3_000_000)} end`;
+    await put('kb/minified.json', `${long}\nneedle short\n`);
+    const files = await all(await search('needle'));
+    expect(files).toHaveLength(1);
+    expect(files[0]!.lines).toHaveLength(2);
+    expect(files[0]!.lines[0]!.text === long).toBe(true);
+    expect(files[0]!.lines[1]).toEqual({ line: 2, text: 'needle short' });
+  });
+
+  it('lists every file by the same walk, in the same order, outside .git and never through a link', async () => {
+    await put('kb/b.pdf', 'x');
+    await put('kb/a/One.PDF', 'x');
+    await put('kb/c.md', 'x');
+    await put('kb/.git/d.pdf', 'x');
+    await put('kb/.hidden/e.eml', 'x');
+    await symlink(join(dir, 'kb/c.md'), join(dir, 'kb/link.md'));
+    const s = await search('x');
+    expect(await s.list({ cwd: dir, root: 'kb' })).toEqual(['.hidden/e.eml', 'a/One.PDF', 'b.pdf', 'c.md']);
+  });
+
+  it('searches exactly the files it is given, whatever their names, from any root', async () => {
+    const names = ['secret.md', 'we ird[1]/#c*.md', '!d.md', 'e.md ', '{f,g}.md', 'back\\slash.md', 'q?.md'];
+    for (const n of names) await put(`kb/${n}`, 'needle\n');
+    await put('kb/open.md', 'needle\n');
+    await put('kb/qx.md', 'needle\n');
+    for (const root of ['kb', '.']) {
+      // Each special name is given alone: it must name itself and nothing else.
+      expect((await all(await search('needle'), [], { root, skip: names })).map((f) => f.path), root).toEqual(['open.md', 'qx.md']);
+      const scope = { cwd: root === '.' ? join(dir, 'kb') : dir, root };
+      for (const n of names) {
+        const files: string[] = [];
+        for await (const f of (await search('needle')).files(scope, { only: [n], maxLinesPerFile: 10 })) files.push(f.path);
+        expect(files, `${root}: ${n}`).toEqual([n]);
+      }
+    }
+  });
+
+  it('never follows a given name that is a link, refuses a name it cannot spell, and runs nothing for no file', async () => {
+    await put('kb/target.md', 'needle\n');
+    await symlink(join(dir, 'kb/target.md'), join(dir, 'kb/link.md'));
+    const s = await search('needle');
+    const scope = { cwd: dir, root: 'kb' };
+    const found = async (only: string[]): Promise<string[]> => {
+      const out: string[] = [];
+      for await (const f of s.files(scope, { only, maxLinesPerFile: 10 })) out.push(f.path);
+      return out;
+    };
+    expect(await found(['link.md'])).toEqual([]);
+    expect(await found(['link.md', 'target.md'])).toEqual(['target.md']);
+    // No whitelist at all would have ripgrep search everything.
+    expect(await found([])).toEqual([]);
+    await expect(found(['a\nb.md'])).rejects.toThrow(/line break/);
+  });
+
+  it('keeps at most maxLinesPerFile lines of a file and still answers the next file', async () => {
+    await put('kb/a.md', 'needle\n'.repeat(10));
+    await put('kb/b.md', 'x\nneedle b\n');
+    const files = await all(await search('needle'), [], { maxLinesPerFile: 3 });
+    expect(files).toEqual([
+      { path: 'a.md', lines: [1, 2, 3].map((line) => ({ line, text: 'needle' })) },
+      { path: 'b.md', lines: [{ line: 2, text: 'needle b' }] },
+    ]);
+  });
+
+  it('reads a file no further than its cap, yet drops it when a NUL lies past the cap', async () => {
+    // Past ripgrep's first buffer, so the walk that stops at the cap never reaches the NUL.
+    const lines = 'needle\n'.repeat(500_000);
+    await put('kb/a.md', Buffer.concat([Buffer.from(lines), Buffer.from([0]), Buffer.from('needle\n')]));
+    await put('kb/b.md', lines);
+    await put('kb/c.md', 'x\nneedle c\n');
+    const files = await all(await search('needle'), [], { maxLinesPerFile: 2 });
+    expect(files).toEqual([
+      { path: 'b.md', lines: [1, 2].map((line) => ({ line, text: 'needle' })) },
+      { path: 'c.md', lines: [{ line: 2, text: 'needle c' }] },
+    ]);
+  });
+
+  it('counts matching lines that are not UTF-8 toward the cap, and reads the file on for the lines it may answer', async () => {
+    const latin = (n: number): Buffer => Buffer.from('needle \xff\n'.repeat(n), 'latin1');
+    await put('kb/latin.txt', Buffer.concat([latin(1), Buffer.from('needle ok\n')]));
+    // Many lines that are not UTF-8 before the first one that is: the cap is raised until it is reached.
+    await put('kb/many.txt', Buffer.concat([latin(1_000), Buffer.from('needle many\n'), Buffer.from('needle more\n')]));
+    // The raised cap reads on past the NUL the first run stopped short of.
+    await put('kb/nul.txt', Buffer.concat([latin(1), Buffer.alloc(200_000, 'x'), Buffer.from([0]), Buffer.from('\nneedle nul\n')]));
+    await put('kb/z.txt', 'needle z\n');
+    expect(await all(await search('needle'), [], { maxLinesPerFile: 1 })).toEqual([
+      { path: 'latin.txt', lines: [{ line: 2, text: 'needle ok' }] },
+      { path: 'many.txt', lines: [{ line: 1_001, text: 'needle many' }] },
+      { path: 'z.txt', lines: [{ line: 1, text: 'needle z' }] },
+    ]);
+  });
+
+  it('walks only the folders holding the files it is given, a folder\'s own files without its subfolders, in the order one walk of the root would', async () => {
+    for (const p of ['A/x/1.md', 'A/y/0.md', 'B/2.md', 'C/y/3.md', 'C/z/9.md', 'T/a/1.md', 'T/a/2.md', 'T/b/3.md', 'top.md', 'U/u.md']) {
+      await put(`kb/${p}`, 'needle\n');
+    }
+    const walked = await recordWalks();
+    try {
+      const found = async (root: string, only: string[]): Promise<{ paths: string[]; runs: string[][] }> => {
+        const scope = { cwd: root === '.' ? join(dir, 'kb') : dir, root };
+        const s = await search('needle');
+        await walked.runs();
+        const paths: string[] = [];
+        for await (const f of s.files(scope, { only, maxLinesPerFile: 10 })) paths.push(f.path);
+        return { paths, runs: await walked.runs() };
+      };
+      expect(await found('kb', ['T/a/1.md', 'T/a/2.md'])).toEqual({ paths: ['T/a/1.md', 'T/a/2.md'], runs: [['kb/T/a@1']] });
+      // The first and last folder are narrowed, every one between walked whole.
+      expect(await found('kb', ['A/x/1.md', 'B/2.md', 'C/y/3.md'])).toEqual({
+        paths: ['A/x/1.md', 'B/2.md', 'C/y/3.md'],
+        runs: [['kb/A/x@1'], ['kb/B'], ['kb/C/y@1']],
+      });
+      expect(await found('.', ['T/a/2.md', 'T/b/3.md'])).toEqual({ paths: ['T/a/2.md', 'T/b/3.md'], runs: [['./T/a@1', './T/b@1']] });
+      // A file directly in a folder is reached by walking that folder's own
+      // files, not its subfolders: the root's file walks the root one deep.
+      expect(await found('kb', ['T/b/3.md', 'top.md'])).toEqual({ paths: ['T/b/3.md', 'top.md'], runs: [['kb/T/b@1', 'kb@1']] });
+      expect(await found('kb', ['top.md'])).toEqual({ paths: ['top.md'], runs: [['kb@1']] });
+    } finally {
+      await walked.remove();
+    }
+  });
+
+  it('answers a folder\'s own files on both sides of a subfolder once each, in order', async () => {
+    for (const p of ['a.md', 'M/m.md', 'M/deep/d.md', 'z.md']) await put(`kb/${p}`, 'needle\n');
+    const walked = await recordWalks();
+    try {
+      const s = await search('needle');
+      await walked.runs();
+      const paths: string[] = [];
+      for await (const f of s.files({ cwd: dir, root: 'kb' }, { only: ['a.md', 'M/deep/d.md', 'M/m.md', 'z.md'], maxLinesPerFile: 10 })) paths.push(f.path);
+      expect(paths).toEqual(['a.md', 'M/deep/d.md', 'M/m.md', 'z.md']);
+      expect(await walked.runs()).toEqual([['kb@1'], ['kb/M'], ['kb@1']]);
+    } finally {
+      await walked.remove();
+    }
+  });
+
+  it('walks a folder whole rather than run once per alternation when its files and subfolders alternate past the cap', async () => {
+    const only: string[] = [];
+    for (let i = 0; i < 12; i++) {
+      const name = `p${String(i).padStart(2, '0')}`;
+      only.push(`${name}.md`, `${name}/x.md`);
+      await put(`kb/${name}.md`, 'needle\n');
+      await put(`kb/${name}/x.md`, 'needle\n');
+    }
+    only.sort(compareSearchPaths);
+    const walked = await recordWalks();
+    try {
+      const s = await search('needle');
+      await walked.runs();
+      const paths: string[] = [];
+      for await (const f of s.files({ cwd: dir, root: 'kb' }, { only, maxLinesPerFile: 10 })) paths.push(f.path);
+      expect(paths).toEqual(only);
+      expect(await walked.runs()).toEqual([['kb']]);
+    } finally {
+      await walked.remove();
+    }
+  });
+
+  it('walks from the root when a folder it would start at has become a link, so the link is not followed', async () => {
+    await put('kb/T/a/1.md', 'x\n');
+    await put('outside/1.md', 'needle\n');
+    const s = await search('needle');
+    const scope = { cwd: dir, root: 'kb' };
+    const only = await s.list(scope);
+    expect(only).toEqual(['T/a/1.md']);
+    // Listed as a folder, then swapped for a link out before the search runs.
+    await rm(join(dir, 'kb/T/a'), { recursive: true });
+    await symlink(join(dir, 'outside'), join(dir, 'kb/T/a'));
+    const walked = await recordWalks();
+    try {
+      const paths: string[] = [];
+      for await (const f of s.files(scope, { only, maxLinesPerFile: 10 })) paths.push(f.path);
+      expect(paths).toEqual([]);
+      expect(await walked.last()).toEqual(['kb']);
+    } finally {
+      await walked.remove();
+    }
+  });
+
+  it('walks nothing when the root itself has become a link since the scope was made', async () => {
+    await put('kb/T/1.md', 'needle\n');
+    await put('outside/T/1.md', 'needle outside\n');
+    const s = await search('needle');
+    const scope = { cwd: dir, root: 'kb/T' };
+    const only = await s.list(scope);
+    expect(only).toEqual(['1.md']);
+    // The scope's root, checked when it was made, swapped for a link out.
+    await rm(join(dir, 'kb/T'), { recursive: true });
+    await symlink(join(dir, 'outside/T'), join(dir, 'kb/T'));
+    const paths: string[] = [];
+    for await (const f of s.files(scope, { only, maxLinesPerFile: 10 })) paths.push(f.path);
+    expect(paths).toEqual([]);
+    expect(await s.list(scope)).toEqual([]);
+  });
+
+  it('places each matching line in the text it came from', async () => {
+    const s = await search('^n');
+    expect(await s.matchingLines(['no\nyes\nnah', '', 'x\n', 'n'])).toEqual([[0, 2], [], [], [0]]);
+    expect(await s.matchingLines([])).toEqual([]);
+    // An empty last line is a line, as `split('\n')` has it.
+    expect(await (await search('^$')).matchingLines(['a\n'])).toEqual([[1]]);
+  });
+
+  it('honours ignoreCase', async () => {
+    expect(await (await search('NEEDLE')).matchingLines(['needle'])).toEqual([[]]);
+    expect(await (await search('NEEDLE', {}, true)).matchingLines(['needle'])).toEqual([[0]]);
+  });
+
+  it('refuses what the engine refuses, quoting its reason without its advice to pass flags', async () => {
+    await expect(search('foo(?=bar)')).rejects.toMatchObject({ status: 400, details: { code: 'invalid_pattern' } });
+    const err = (await search('(a)\\1').catch((e: unknown) => e)) as Error;
+    expect(err.message).toMatch(/^Invalid regex: regex parse error:/);
+    expect(err.message).toContain('backreferences are not supported');
+    expect(err.message).not.toContain('PCRE2');
+    expect(err.message).toContain("grep runs ripgrep's default regex engine");
+    await expect(search('a\u0000b')).rejects.toMatchObject({ status: 400 });
+    await expect(search('x'.repeat(1_001))).rejects.toMatchObject({ status: 400, details: { code: 'pattern_too_long' } });
+  });
+
+  it('reads no ripgrep configuration file', async () => {
+    const config = join(dir, 'ripgreprc');
+    await writeFile(config, '--ignore-case\n--hidden\n');
+    const before = process.env.RIPGREP_CONFIG_PATH;
+    process.env.RIPGREP_CONFIG_PATH = config;
+    try {
+      expect(await (await search('NEEDLE')).matchingLines(['needle'])).toEqual([[]]);
+    } finally {
+      if (before === undefined) delete process.env.RIPGREP_CONFIG_PATH;
+      else process.env.RIPGREP_CONFIG_PATH = before;
+    }
+  });
+
+  it('reads a leading dash as the pattern, never as a flag', async () => {
+    await put('kb/f.md', 'use --files here\n');
+    const files = await all(await search('--files'));
+    expect(files).toEqual([{ path: 'f.md', lines: [{ line: 1, text: 'use --files here' }] }]);
+  });
+
+  it('stops a walk at the budget with the 400, killing the run', async () => {
+    const slow = await slowRipgrep();
+    try {
+      const s = await search('needle', { budgetMs: 300 });
+      const started = Date.now();
+      await expect(all(s)).rejects.toMatchObject({ status: 400, details: { code: 'pattern_too_expensive' } });
+      expect(Date.now() - started).toBeLessThan(2_000);
+      expect(await slow.alive()).toEqual([]);
+      // The budget is the search's: spent, every later run is refused at once.
+      await expect(s.matchingLines(['needle'])).rejects.toMatchObject({ details: { code: 'pattern_too_expensive' } });
+    } finally {
+      await slow.remove();
+    }
+  });
+
+  it('spends the budget on the work it is charged with, and answers the 400 at the budget when that work outlasts it', async () => {
+    const s = await search('needle', { budgetMs: 600 });
+    const sleep = (ms: number) => () => new Promise<string>((r) => setTimeout(() => r('done'), ms));
+    expect(await s.charged(sleep(200))).toBe('done');
+    const started = Date.now();
+    await expect(s.charged(sleep(5_000))).rejects.toMatchObject({ status: 400, details: { code: 'pattern_too_expensive' } });
+    expect(Date.now() - started).toBeLessThan(1_000);
+    // Spent: neither more work nor another run is started.
+    await expect(s.charged(sleep(1))).rejects.toMatchObject({ details: { code: 'pattern_too_expensive' } });
+    await expect(s.matchingLines(['needle'])).rejects.toMatchObject({ details: { code: 'pattern_too_expensive' } });
+    // A failure of the work itself is the caller's to hear.
+    const t = await search('needle');
+    await expect(t.charged(() => Promise.reject(new Error('verdict failed')))).rejects.toThrow('verdict failed');
+  });
+
+  it('spends the whole budget when its timer fires, however little time the clock measured', async () => {
+    // A clock that measures no time at all: only the timer's firing can spend the budget.
+    const clock = vi.spyOn(Date, 'now').mockReturnValue(Date.now());
+    const slow = await slowRipgrep();
+    try {
+      const sleep = (ms: number) => () => new Promise<string>((r) => setTimeout(() => r('done'), ms));
+      const s = await search('needle', { budgetMs: 300 });
+      await expect(s.charged(sleep(5_000))).rejects.toMatchObject({ details: { code: 'pattern_too_expensive' } });
+      await expect(s.charged(sleep(1))).rejects.toMatchObject({ details: { code: 'pattern_too_expensive' } });
+      const t = await search('needle', { budgetMs: 300 });
+      await expect(all(t)).rejects.toMatchObject({ details: { code: 'pattern_too_expensive' } });
+      await expect(t.charged(sleep(1))).rejects.toMatchObject({ details: { code: 'pattern_too_expensive' } });
+      const refused = performance.now();
+      await expect(t.matchingLines(['needle'])).rejects.toMatchObject({ details: { code: 'pattern_too_expensive' } });
+      expect(performance.now() - refused).toBeLessThan(100);
+    } finally {
+      clock.mockRestore();
+      await slow.remove();
+    }
+  });
+
+  it('answers charged work as abandoned when the caller goes away', async () => {
+    const gone = new AbortController();
+    const s = await matcher().search('needle', { ignoreCase: false, signal: gone.signal });
+    const waiting = s.charged(() => new Promise((r) => setTimeout(r, 5_000)));
+    gone.abort();
+    await expect(waiting).rejects.toMatchObject({ status: 499 });
+  });
+
+  it('ends the run when the caller stops reading, and gives its slot back', async () => {
+    for (let i = 0; i < 50; i++) await put(`kb/f${String(i).padStart(2, '0')}.md`, 'needle\n');
+    const m = matcher({ maxRunning: 1, maxWaiting: 0 });
+    const s = await m.search('needle', { ignoreCase: false });
+    const scope = { cwd: dir, root: 'kb' };
+    for await (const f of s.files(scope, { only: await s.list(scope), maxLinesPerFile: 1 })) {
+      expect(f.path).toBe('f00.md');
+      break;
+    }
+    // With one slot and no queue, a run still holding it would make this a 503.
+    expect(await (await m.search('needle', { ignoreCase: false })).matchingLines(['needle'])).toEqual([[0]]);
+  });
+
+  it('spends one budget across every run of a search, so runs that each finish can still exhaust it', async () => {
+    // Each walk takes 1 s and finishes; the third has 500 ms of a 2.5 s budget left.
+    const slow = await slowRipgrep({ walkMs: 1_000 });
+    try {
+      const s = await search('needle', { budgetMs: 2_500 });
+      expect(await all(s)).toEqual([]);
+      expect(await s.list({ cwd: dir, root: 'kb' })).toEqual([]);
+      const started = Date.now();
+      await expect(all(s)).rejects.toMatchObject({ status: 400, details: { code: 'pattern_too_expensive' } });
+      expect(Date.now() - started).toBeLessThan(1_000);
+      expect(await slow.alive()).toEqual([]);
+    } finally {
+      await slow.remove();
+    }
+  });
+
+  it('kills every run when the caller goes away, answers it abandoned, and gives the slot back', async () => {
+    const slow = await slowRipgrep();
+    try {
+      const m = matcher({ maxRunning: 1, maxWaiting: 0 });
+      const gone = new AbortController();
+      const s = await m.search('needle', { ignoreCase: false, signal: gone.signal });
+      const running = all(s);
+      while ((await slow.alive()).length === 0) await new Promise((r) => setTimeout(r, 20));
+      const started = Date.now();
+      gone.abort();
+      await expect(running).rejects.toMatchObject({ status: 499, details: { code: 'search_abandoned' } });
+      expect(Date.now() - started).toBeLessThan(1_000);
+      expect(await slow.alive()).toEqual([]);
+      // A later run of the abandoned search is refused before it starts.
+      await expect(s.matchingLines(['needle'])).rejects.toMatchObject({ details: { code: 'search_abandoned' } });
+      expect(await (await m.search('needle', { ignoreCase: false })).matchingLines(['needle'])).toEqual([[0]]);
+    } finally {
+      await slow.remove();
+    }
+  });
+
+  it('a caller gone while its run waits for a slot leaves the queue, answered abandoned, and the queue is free again', async () => {
+    const slow = await slowRipgrep();
+    try {
+      const m = matcher({ maxRunning: 1, maxWaiting: 1 });
+      const holder = await m.search('needle', { ignoreCase: false });
+      const running = all(holder);
+      while ((await slow.alive()).length === 0) await new Promise((r) => setTimeout(r, 20));
+      // The pattern check of a new search is a run: with the one slot held it queues.
+      const gone = new AbortController();
+      const queued = m.search('needle', { ignoreCase: false, signal: gone.signal });
+      await new Promise((r) => setTimeout(r, 50));
+      gone.abort();
+      await expect(queued).rejects.toMatchObject({ status: 499, details: { code: 'search_abandoned' } });
+      // Its place in the queue went with it: the next search waits rather than being refused as one too many.
+      const next = m.search('needle', { ignoreCase: false });
+      const outcome = await Promise.race([next.then(() => 'ran', () => 'refused'), new Promise((r) => setTimeout(() => r('waiting'), 200))]);
+      expect(outcome).toBe('waiting');
+      await m.close();
+      await expect(next).rejects.toMatchObject({ status: 503 });
+      await expect(running).rejects.toMatchObject({ status: 503 });
+      expect(await slow.alive()).toEqual([]);
+    } finally {
+      await slow.remove();
+    }
+  });
+
+  it('charges the wait for a slot to the budget, and answers the 400 from the queue when the budget passes there', async () => {
+    // Every search of a matcher has the same budget, so a queued search can
+    // outwait its own only once it has spent part of it: one walk of 500 ms
+    // against a budget of 800, then a wait behind a fresh search's 500 ms
+    // walk, which outlasts the 300 ms the first has left.
+    const slow = await slowRipgrep({ walkMs: 500 });
+    try {
+      await put('kb/a.md', 'needle\n');
+      const m = matcher({ maxRunning: 1, maxWaiting: 1, budgetMs: 800 });
+      const scope = { cwd: dir, root: 'kb' };
+      const s = await m.search('needle', { ignoreCase: false });
+      expect(await s.list(scope)).toEqual([]);
+      const holder = await m.search('needle', { ignoreCase: false });
+      const holding = holder.list(scope);
+      while ((await slow.alive()).length === 0) await new Promise((r) => setTimeout(r, 20));
+      const started = Date.now();
+      const queued = s.list(scope);
+      // Refused from the queue, before the holder's walk ends — so it never ran.
+      const first = await Promise.race([
+        queued.then(
+          () => 'ran',
+          () => 'refused',
+        ),
+        holding.then(() => 'the holder finished first'),
+      ]);
+      expect(first).toBe('refused');
+      await expect(queued).rejects.toMatchObject({ status: 400, details: { code: 'pattern_too_expensive' } });
+      expect(Date.now() - started).toBeLessThan(1_000);
+      // The wait spent the budget: a later run is refused at once.
+      await expect(s.matchingLines(['needle'])).rejects.toMatchObject({ details: { code: 'pattern_too_expensive' } });
+      expect(await holding).toEqual([]);
+    } finally {
+      await slow.remove();
+    }
+  });
+
+  it('refuses a run past the queue with a 503 and ends every run on close', async () => {
+    const slow = await slowRipgrep();
+    try {
+      const m = matcher({ maxRunning: 1, maxWaiting: 0 });
+      const s = await m.search('needle', { ignoreCase: false });
+      const running = all(s);
+      while ((await slow.alive()).length === 0) await new Promise((r) => setTimeout(r, 20));
+      await expect(m.search('needle', { ignoreCase: false })).rejects.toMatchObject({ status: 503, details: { code: 'search_busy' } });
+      await m.close();
+      await expect(running).rejects.toMatchObject({ status: 503 });
+      expect(await slow.alive()).toEqual([]);
+    } finally {
+      await slow.remove();
+    }
+  });
+
+  it('refuses sizes it cannot work with', () => {
+    expect(() => new RipgrepPatternMatcher({ maxRunning: 0 })).toThrow(/maxRunning must be an integer from 1 /);
+    expect(() => new RipgrepPatternMatcher({ budgetMs: 0 })).toThrow(/budgetMs/);
+    expect(() => new RipgrepPatternMatcher({ maxWaiting: -1 })).toThrow(/maxWaiting/);
+  });
+});

package/src/modules/workspace/__tests__/ripgrep-version.test.ts

@@ -0,0 +1,29 @@
+import { mkdtemp, rm } from 'node:fs/promises';
+import { tmpdir } from 'node:os';
+import { join } from 'node:path';
+import { afterEach, describe, expect, it } from 'vitest';
+import { assertRipgrep } from '../index.js';
+
+/** The boot check, against the real PATH and against one where `rg` is nowhere. */
+describe('assertRipgrep', () => {
+  const path = process.env.PATH;
+  let empty = '';
+  afterEach(async () => {
+    process.env.PATH = path;
+    if (empty) await rm(empty, { recursive: true, force: true });
+    empty = '';
+  });
+
+  it('passes where rg is on the PATH', async () => {
+    await expect(assertRipgrep()).resolves.toBeUndefined();
+  });
+
+  it('refuses to boot without rg, naming the binary and the Dockerfile that installs it', async () => {
+    empty = await mkdtemp(join(tmpdir(), 'no-rg-'));
+    process.env.PATH = empty;
+    const err = (await assertRipgrep().catch((e: unknown) => e)) as Error;
+    expect(err).toBeInstanceOf(Error);
+    expect(err.message).toMatch(/^ripgrep \(`rg`\) is required on the server's PATH but could not be run: .*ENOENT/);
+    expect(err.message).toContain('The Hexis Dockerfile installs it');
+  });
+});