Function collectIncludedFiles

src/extraction/index.ts:410–477

* Actively discover the source files an `include` whitelist forces in. `git * ls-files` never lists gitignored files, so a filtered filesystem walk of just * the opted-in subtrees (`includeStaticRoots`) is the only way to find them. * Returns project-root-relative, normalized source-file paths.

(
  rootDir: string,
  include: Ignore,
  exclude: Ignore | null,
  roots: string[],
  overrides: Record<string, Language>,
)

 * `.gitignore` is deliberately NOT consulted: overriding it is the whole point.
 */
function collectIncludedFiles(
  rootDir: string,
  include: Ignore,
  exclude: Ignore | null,
  roots: string[],
  overrides: Record<string, Language>,
): Set<string> {
  const out = new Set<string>();
  const defaults = defaultsOnlyIgnore();
  const visited = new Set<string>();

  const consider = (abs: string, rel: string, isDir: boolean): void => {
    if (isDir) {
      if (defaults.ignores(rel + '/')) return; // never node_modules/dist/… via include
      // An explicit `exclude` always wins over `include`; prune the whole subtree
      // here so a large excluded dir (a committed frontend's own vendored deps,
      // build output, …) is never walked — the per-file guard below still catches
      // anything a directory pattern doesn't, so this is a pure efficiency win.
      if (exclude && exclude.ignores(rel + '/')) return;
      walk(abs);
    } else {
      if (defaults.ignores(rel)) return;
      if (!include.ignores(rel)) return;
      if (exclude && exclude.ignores(rel)) return;
      if (!isSourceFile(rel, overrides)) return;
      out.add(rel);
    }
  };

  function walk(absDir: string): void {
    let realDir: string;
    try {
      realDir = fs.realpathSync(absDir);
    } catch {
      return;
    }
    if (visited.has(realDir)) return; // symlink-cycle guard
    visited.add(realDir);

    let entries: fs.Dirent[];
    try {
      entries = fs.readdirSync(absDir, { withFileTypes: true });
    } catch {
      return;
    }
    for (const entry of entries) {
      if (entry.name === '.git' || isCodeGraphDataDir(entry.name)) continue;
      const abs = path.join(absDir, entry.name);
      const rel = normalizePath(path.relative(rootDir, abs));
      if (!rel || rel.startsWith('..')) continue;
      if (entry.isSymbolicLink()) {
        try {
          const st = fs.statSync(fs.realpathSync(abs));
          consider(abs, rel, st.isDirectory());
        } catch {
          // broken symlink — skip
        }
        continue;