Avoid getcwd when the path is already absolute

[Brett-Best]

Avoid getcwd when the path is already absolute

What

NSString.absolutePathRepresentation(rootDirectory:) has an early return for already-absolute paths, but its default argument runs a getcwd syscall on every call regardless — even when that early return means the value is never used.

-    public func absolutePathRepresentation(rootDirectory: String = FileManager.default.currentDirectoryPath) -> String {
+    public func absolutePathRepresentation(rootDirectory: String? = nil) -> String {
         if isAbsolutePath { return expandingTildeInPath }
+        let rootDirectory = rootDirectory ?? FileManager.default.currentDirectoryPath

This makes rootDirectory optional and resolves the working directory lazily, only on the relative-path branch that actually uses it.

Why it's not as cheap as it looks

Swift evaluates a default argument at the call site, before the callee body runs. So absolutePathRepresentation() on an absolute path still computes FileManager.default.currentDirectoryPath and then discards it.

FileManager.default is a singleton, but .currentDirectoryPath is not a cached value — it does real work on every access. In swift-foundation it allocates a buffer, issues a getcwd(2) syscall, and constructs a fresh String:

https://github.com/swiftlang/swift-foundation/blob/96d4094612d02bdc6f474f73efa0695196e786e9/Sources/FoundationEssentials/FileManager/FileManager+Directories.swift#L512-L529

(On Darwin the ObjC Foundation implementation is at least as expensive.)

Impact

File(pathDeferringReading:) calls this once per file, always with an absolute path. Tools that enumerate large projects therefore make one pointless getcwd per file during file discovery.

Profiling SwiftLint linting a ~7,000-file project (Instruments Time Profiler, release build): currentDirectoryPath accounts for ~0.8 s of CPU (≈2.7 % of the run), and it sits on the serial pre-lint file-discovery path, so it is a proportionally larger share of warm/incremental runs where little else executes. An isolated before/after A/B of the same lint measured ~3 % off both cold and warm wall-clock (the wall figure is smaller than the CPU figure because the directory walk is parallelized).

Correctness

Behavior-preserving by construction: the resolved working directory is only ever used on the relative-path branch, which is unchanged; the absolute-path branch never read it. Verified end-to-end by linting a large project with all rules enabled — identical violation output (same count, files, lines, columns) before and after.