Make filename return Nothing for paths with no filename component (resolves #4)

[carpentry-agent]

Resolves #4.

filename returns (Maybe String), but because split never yields an empty array its Array.last was always a Just — so Nothing was unreachable and the Maybe was dishonest.

The issue offered two fixes: (a) define cases where filename should return Nothing, or (b) drop the Maybe altogether. This PR implements (a): filename now returns Nothing when the path has no filename component — i.e. when the last split component is the empty string. Otherwise it returns Just of the last component.

input	result
"a/b.txt"	Just "b.txt"
"foo"	Just "foo"
"a/b/c"	Just "c"
""	Nothing
"/"	Nothing
"foo/"	Nothing

Why (a) over (b): you listed it first and gave the empty-string example, it keeps the Maybe meaningful (a directory path or bare root genuinely has no filename), and it's the smaller API break — callers already had to handle Maybe, and the only ones that relied on the old "always Just, empty string for empty path" behavior are now steered toward the honest Nothing. If you'd rather go with (b) and drop the Maybe entirely, happy to redirect.

basename is implemented independently (via split/Array.prefix) and is unaffected; it was the only other internal user of the split machinery. The docstring is reversed to document the new Nothing cases, and focused tests cover each row above. No CHANGELOG.md in this repo, so nothing to add there.

---

Opened by the carpentry-org heartbeat agent (Claude). Veit has not reviewed this yet.

[carpentry-reviewer]

Build & Tests

Checked out claude/filename-nothing-empty at 65e8307 — master (90a7235) + one commit, clean topology. Two files changed: path.carp (the fix) and test/path.carp (5 new asserts).

• carp -x test/path.carp → 37 passed, 0 failed.
• carp-fmt --check and angler: clean (rc 0). CI: both jobs green.

Findings

• Implementation is correct and idiomatic. filename is now (Maybe.filter (Array.last &(split p)) &(fn [f] (not (String.empty? f)))). I traced the types against core: Array.last returns a copied (Maybe String), and since split never yields an empty array it's always Just; Maybe.filter's predicate is called as (~f &x), so f binds a &String, and String.empty? : (Fn [&String] Bool) matches exactly. Sound, no ref/ownership issue.
• Probed beyond the suite. Empirically: split yields [""] for "", ["" ""] for "/", ["foo" "" ""] for "foo//" → filename is Nothing for each. Just cases: "a/b/c"→c, "/foo"→foo, ".hidden"→.hidden (hidden files handled), "."→. (matches os.path.basename(".")), "a/b/"→Nothing. All sensible.
• basename is independent (uses Array.prefix) and unaffected — confirmed; it's the only other split user, and there's no other caller of filename in the lib.
• Docstring is reversed to match and reads accurately. No CHANGELOG.md in this repo (confirmed), and README.md has no filename docs — nothing else to update.

Direction note (your call)

Issue #4 offered (a) define Nothing cases or (b) drop the Maybe. This PR implements (a) — which you listed first and illustrated with the empty-string example, so it's well-grounded. Two things worth naming explicitly:

• Merging this is the choice of (a) over (b). The code correctly implements (a); if you'd rather drop the Maybe entirely (b), this would need redirecting.
• It's a user-facing behavior change: filename "" was Just "" (per the just-merged #3) and is now Nothing. That's exactly the intent of #4 and fine pre-0.1 — just flagging it's observable for any caller relying on the old always-Just contract.

Verdict: merge

Correct, minimal, well-tested implementation of option (a) from issue #4, with the empty-string case you named handled exactly. Builds, 37/37 pass, fmt/lint clean, no bugs found. The only open item is the (a)-vs-(b) direction, which is yours — merging commits to (a).