[cjfields]

Refs #23, which stays open until upstream C resolves it. This change is behind --no-c-compat; C mode is unchanged.

Problem

In score-only mode, C prints a score even when there's no alignment. CIGAR mode prints -1 - for the same pairs. What score-only prints depends on the case:

• Unreachable with a match reward: the classic score evaluated at the unset end position (k = INT_MAX, offset = OFFSET_NULL). v = offset - k is a signed overflow, which wraps so that v + h == 1. Example: wfa.utest.seq pair 262 with affine -3,5,3,2 and X-drop 20,1 prints -113. (The example in Known issue (C parity): score-only score for unreachable alignments evaluated at an unset end position #23, pair 101, is actually reachable; the pair number was off. I'll correct the issue.)
• Unreachable with match == 0: the negated score where the search gave up.
• Failed BiWFA score search: INT32_MIN, left by cigar_clear.

In Rust, the library already returns these results with an unreachable status, and its docs say the score is meaningless then. Only align_benchmark converts them, reproducing C. So the fix is in the CLI.

Fix

With --no-c-compat, score-only prints -1, as CIGAR mode does, when:

• the alignment ended unreachable with no Z-drop cell; or
• BiWFA's score search failed.

Z-drop results keep their score, because they come from a real cell.

Results

• Differential suite, C mode: unchanged; only the 20 known Known issue (C bug): Z-drop in med/low backtraces from uninitialized cells (high vs med differ) #22 cases differ.
• Differential suite, corrected mode: only score-only cases change (banded-static, banded-adaptive, X-drop, Z-drop, BiWFA). No CIGAR output changes.
• Consistency: on the sampled configurations, every -1 corresponds to a CIGAR-mode -1 -, except the cases listed below.

Left as is (FIXED_BUGS 14h)

• BiWFA score-only has no partial result. When its search Z-drops it prints -1 (C: INT32_MIN), even though CIGAR mode returns the partial alignment. Example: utest with affine -3,5,3,2 and Z-drop 100,5, 7/305 pairs. Reporting a partial score there would need new BiWFA score-only logic.
• BiWFA's score search can succeed where its CIGAR search fails in a sub-problem, because the cutoffs are independent.
• A Z-drop whose CIGAR max-trims to empty keeps its score.
• With a match reward, -1 can also be a real score (utest pair 281, 10 × 10 bp), as it can in C's CIGAR mode.

Tests

unreachable_score_only_corrected_prints_minus_one (tests/cli_output.rs, tests/data/issue23_unreachable.seq) covers:

• unidirectional: C -113 vs corrected -1;
• BiWFA: C INT32_MIN vs corrected -1;
• banded with match == 0.

All tests pass; clippy is clean.

🤖 Generated with Claude Code