Add str/prn for Diff, refactor unified, add unified-with-context

[carpentry-agent]

Summary

• str/prn for the Diff type: the Diff sumtype now implements both interfaces, so diffs can be printed and formatted (e.g. (str &(Eq [1 2 3])) → "(Eq [1 2 3])")
• Refactored unified: eliminated O(n²) string accumulation (now collects hunk strings in an array and concatenates once at the end) and deduplicated the hunk-start logic that was copy-pasted across the Deletion and Insertion branches (now a single when-do block before the match-ref)
• unified-with-context: new public function that takes the number of context lines as a parameter; unified now delegates to it with 3. Renamed the internal emit-hunk (which accumulated into the result string) to format-hunk (which returns a single hunk string)
• Tests: 8 new tests covering str on all three variants, unified-with-context with 0/1/3 context lines, and equivalence with unified

All 62 tests pass. carp-fmt and angler clean.

---

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

[carpentry-reviewer]

Build & Tests

CI: pass (ubuntu-latest, macos-latest) — all 62 tests pass (54 existing + 8 new).
Local build: no Carp compiler available on this machine — relied on CI.

Findings

Reviewed all changed code in diff.carp (lines 14-266) and all 8 new tests in tests/diff.carp (lines 313-381).

str/prn (lines 14-26): correct

Both use match-ref (borrowing), dispatch to the inner type's str/prn, and wrap with the variant name. Standard Carp pattern for sumtype printing. implements declarations are present.

format-hunk rename (lines 144-159): correct

Renamed from emit-hunk, now returns a single hunk string instead of concatenating onto the accumulator. Clean separation of concerns.

change? helper (line 142): correct

Returns false for Eq, true for Insertion/Deletion. Used to deduplicate the hunk-start logic.

unified-with-context refactoring (lines 165-261): correct

Verified the refactoring preserves behavior by tracing through several cases:

1. Hunk-start deduplication: the when-do (and (not in-hunk) (change? elem)) block (lines 181-189) runs before the match-ref, correctly replacing the identical unless-do in-hunk blocks that were duplicated in the old Deletion and Insertion branches. Since change? returns false for Eq, the when-do never fires for context lines. ✓

2. O(n²) elimination: result accumulation changed from (String.concat &[output header ... body ...]) (copying the entire string on each hunk emission) to collecting into a hunks array and concatenating once at the end (line 261). Genuine asymptotic improvement for multi-hunk diffs. ✓

3. ctx=0: traced through manually — trailing = min(0, n) = 0 and leading = min(0, ...) = 0 correctly suppress all context lines. The test at line 332 confirms. ✓

4. ctx=1 multi-hunk: traced through — two hunks with 3 intervening Eq lines (gap > 2*ctx) correctly split into separate hunks. The test at line 359 confirms. ✓

5. Edge case — change at start of diff: lead-lines is empty initially, so hunk-lines starts empty, hunk-old-count = 0. Context is only appended after the change. ✓

6. Edge case — change at end of diff: the post-loop when in-hunk (line 254) correctly emits the final hunk. ✓

unified delegation (line 266): correct

(defn unified [d] (unified-with-context d 3)) — matches the original hardcoded context of 3. The test at line 350 explicitly checks unified and unified-with-context 3 produce identical output.

Minor notes (non-blocking)

• No test for prn output (only str is tested). For integer arrays they produce the same output, so this is cosmetic. Would be meaningful for (Diff String) where prn quotes strings.
• No CHANGELOG in this repo.

Verdict: merge

Clean, well-motivated refactoring. The O(n²) fix is genuine. Hunk-start deduplication reduces 15 lines of copy-paste to a single when-do block. unified-with-context is a useful addition. All 62 tests pass. No bugs found.