# Find Duplicate Files with a Hand-Written Directory Traversal Implement a duplicate-file finder starting from a root directory, but do not use a recursive walking helper such as `Files.walk` or `os.walk`. You may use only low-level operations to list one directory, inspect one directory entry, and stream one file's bytes. Return groups of paths whose file contents are byte-for-byte identical. Include only groups containing at least two files. Explain how you keep memory bounded for large files and how you protect correctness from hash collisions. ### Constraints & Assumptions - Directory depth may exceed the safe recursion depth. - Files may be empty, very large, or unreadable. - Directory enumeration order is unspecified. - By default, do not follow symbolic links; state how the design changes if links must be followed. - The result order must be deterministic for testing. ### Clarifying Questions to Ask - Should hard links to the same underlying file appear as duplicates or as one object? - Should permission and transient I/O errors fail the scan or be returned separately? - Is a cryptographic digest sufficient, or is byte-for-byte verification required? - Can files change while the scan is running? ### What a Strong Answer Covers - An explicit stack or queue for manual traversal - Filtering by file size before hashing - Streaming digests and final equality verification - Cycle, symlink, error, and concurrent-modification policies - Deterministic output and clear complexity bounds ### Follow-up Questions - How would you reuse work across repeated scans? - What race occurs if a file changes between metadata lookup and hashing? - How could the scanner exploit parallel I/O without overwhelming storage? - Would you delete duplicate files automatically based on this result?