Kills all surviving processes that reference dirPath (checked via cwd, cmdline, and open file descriptors), then retries fs.rm() until it succeeds or a 5-second deadline expires. Re-scans /proc on each failed rm() attempt to catch processes that appear after the initial scan (e.g. late-forked Chrome helpers).

Deliberately does NOT wait for killed processes to fully leave the process table (no killAndAwait). kill(pid, 0) succeeds for zombie processes (FDs already released) and for D-state processes (SIGKILL queued but not yet delivered), so polling it can stall indefinitely on a loaded CI machine while rm() could already succeed. Instead, rm() itself is the authoritative check: it fails while FDs are held and succeeds once they are released, regardless of how far the process-table cleanup has progressed.

Linux-only: falls back to a single best-effort rm() on other platforms.

import * as Chrome from './index.ts';

// Defined, not invoked: a real call SIGKILLs FD holders and rm()s the directory.
async function reapUserDataDir(userDataDir: string) {
  await Chrome.cleanupDir(userDataDir); // resolves once gone, or after the 5s deadline
}