OA — File Systems

Recently observed CodeSignal OA in the file-system family — implement a simplified file system with paths, directory traversal, and (in the final level) an LLM-related extension. Public details are thin; the third level in particular is reported as bug-prone in the supplied helpers. Requirements CodeSignal, 90 minutes, 4 progressively-unlocking levels. Build a path-based file system: create, read, write, list, delete. Hierarchical traversal of /a/b/c style paths. Level 3 introduces an LLM-flavored extension in one rotation (candidates flag bugs in a supplied helper) and a multi-user-quota model in another — see the canonical variant below. I/O contract is harness-string-based: commands arrive as [verb, arg1, arg2, …] lists; every operation returns a string, with "true" / "false" standing in for booleans and "" for null/failure. Notes Candidates report this OA being noticeably more typing-intensive than the in-memory DB variant; the bar appears to be implementation speed. As of mid-2026 the OA increasingly ships a single generic prompt — "a toy simulation of an app that won't require anything except the standard library of your chosen language" — with the actual problem randomized at runtime; one such prompt resolved to the cloud-storage variant below. Prep all the OA families rather than betting on the title. The single most common stumble across rotations is forgetting that COPY_FILE (or its rename) preserves the original owner — not the caller. The hidden tests around mixed-owner copies and post-update-capacity eviction routinely trip candidates who reach Level 3 with shaky bookkeeping. Canonical 4-level shape — Cloud Storage rotation The most-leaked rotation models an in-memory cloud-storage service. All commands are dispatched as string-array queries and return strings; size fields are decimal integers serialised as strings. # Level 1 — basic file management ADD_FILE <name> <size> -> "true" | "false" # "false" if a file already exists at <name>. COPY_FILE <nameFrom> <nameTo> -> "true" | "false" # "false" if source is missing, source is a directory, OR destination exists. GET_FILE_SIZE <name> -> "<size>" | "" # "" if the file is missing. # Level 2 — search FIND_FILE <prefix> <suffix> -> "<n1>(<s1>), <n2>(<s2>), …" | "" # Files whose path starts with <prefix> AND ends with <suffix>. # Sort: size desc, then name asc (alphabetical). Format literally # "<name>(<size>)" with no space inside the parens, comma-space between entries. # "" when no file matches. # Level 3 — users and storage limits ADD_USER <userId> <capacity> -> "true" | "false" # "false" if userId already exists. ADD_FILE_BY <userId> <name> <size> -> "<remaining>" | "" # Like ADD_FILE but owned by <userId>. "" if user is missing, file path # already taken, OR adding <size> would push the user's usage > capacity. # On success returns the user's remaining capacity (string). # Bare ADD_FILE calls are owned by "admin" with unlimited capacity. # COPY_FILE preserves the source file's owner — not the caller. UPDATE_CAPACITY <userId> <capacity> -> "<deleted_count>" | "" # Sets a new capacity. If current usage > new capacity, evict the user's files # largest-first; ties broken by name asc, until usage <= capacity. # Returns the number of files deleted. "" if userId is unknown. # Level 4 — compression COMPRESS_FILE <userId> <name> -> "<remaining>" | "" # Renames <name> -> <name>.COMPRESSED and halves its size (integer division). # "" if file is missing, not owned by <userId>, or already ends in .COMPRESSED. # COPY_FILE preserves the .COMPRESSED suffix on copies. DECOMPRESS_FILE <userId> <name> -> "<remaining>" | "" # Strips the .COMPRESSED suffix and doubles the size back. # "" if the resulting (full-size) file would exceed the user's capacity, # if a file at the un-suffixed name already exists, or if owner does not match. Implementation notes for this variant: Compressed size = original_size // 2 (the half is exact for evens; the OA's test inputs are constructed so all sizes are even). On decompression, restore to 2 × compressed_size. UPDATE_CAPACITY is the eviction trigger most candidates botch: it's NOT "reject future adds" — it actively deletes the largest files until the user fits under the new cap. Compressed and non-compressed files are evicted by their current size. For FIND_FILE, the result string is literal — hidden tests compare it as a string, so any extra space or trailing comma fails. A dict[path] -> FileEntry(size, owner, compressed: bool) + dict[user] -> {capacity, used} is enough; no tree of dicts needed because all paths are flat keys and directories are implicit. Preparation Build a tree-of-dicts file system from scratch in under 45 minutes, supporting mkdir, touch, read, write, ls, rm (recursive), and mv. For the Cloud Storage rotation specifically, drill the dispatch loop pattern: for cmd, *args in queries: results.append(getattr(self, handlers[cmd])(*args)). Spend muscle-memory on parsing string args back to ints rather than reinventing the harness each round. Have a clean helper for parsing and validating paths ("/", trailing slash, dot-segments). Pre-write the FIND_FILE formatter exactly: ", ".join(f"{n}({s})" for n, s in sorted_matches). Hidden tests compare the literal string. Reserve the last 20 minutes for level 4 — keep code modular enough that the extension lands cleanly.