## Scenario Design a backend API that recursively searches a directory tree for entries whose file name exactly matches a requested name. Required behavior: - If the starting path is `null` or does not identify a directory, return the error `Current search location is not a directory`. - If the directory is empty or no matching file exists anywhere below it, return `File not found: name`, substituting the requested name. - Otherwise return every matching regular-file path in deterministic lexicographic order. Explain validation, traversal, symbolic-link policy, permission errors, time and space complexity, and how the API avoids exposing arbitrary server files. ### Constraints & Assumptions - The requested root must resolve beneath a configured allowlisted base directory. - Do not follow symbolic links in the baseline design. - A permission error is returned separately from not-found; it must not be silently interpreted as an empty subtree. - The tree may be deep enough to make recursive call stacks unsafe. ### Clarifying Questions to Ask - Is matching exact and case-sensitive? Yes in the baseline. - Should directories with the requested name count? No, regular files only. - Are all matches returned or just the first? All. - How should concurrent file changes be handled? Best-effort snapshot semantics with documented race behavior. ```hint Treat paths as untrusted input Resolve and normalize the root, then verify it remains beneath the configured base before traversing. ``` ### What a Strong Answer Covers - Exact validation and error distinctions required by the prompt. - Iterative DFS or BFS, deterministic output sorting, and cycle avoidance through the no-symlink rule. - Permission and time-of-check/time-of-use behavior. - Root allowlisting, authorization, result limits, cancellation, and audit logging for a service endpoint. ### Follow-up Questions 1. How would you safely support symbolic links without cycles or base-directory escape? 2. When would an index be preferable to walking the filesystem per request? 3. How would you stream results while preserving deterministic order?