Address review findings on sub-k seed labels (#691) and map-only mode (#692)

Merged
AD
12h ago
+27-7

Follow-up to the cubic review comments on #691 and #692, which were merged before the comments were processed.

#692, --align-map-only with --map / --count-kmers / --query-presence. Those flags take the k-mer mapping path, so --align-map-only was silently ignored. The combination is now rejected in Config, with an error naming the flags.

#691, coordinates in the sub-k seed regression test. SubkSeedLabels now also runs the BASIC case with a coordinate annotation. For every alignment label it checks three things: there is one coordinate set per label, every coordinate is non-negative, and each coordinate points at the aligned sequence in that label. That includes the query that matches only a source dummy k-mer.

#691, negative coordinates for source dummies (no code change). The $ā€¦$ source-dummy coordinates are c - num_dummy, which is negative at a sequence start. They are stored as int64_t (the subtraction wraps in unsigned arithmetic, and the conversion back to int64_t gives the intended negative value). They do not reach the output:

  • A seed on a source dummy spells no $, so its offset is at least num_dummy.
  • filter_seeds adds the offset back, so the seed's coordinate is at least c.
  • Extension compares only differences of coordinates.

Checks:

  • The new coordinate assertions pass.
  • On a 24-genome BASIC graph with a column_coord annotation, I aligned 252 reads that are bare 20/25/30 bp contig-start prefixes on both strands. No reported coordinate wrapped, at seed length 19 (246 aligned) or 15 (252 aligned).

#692, two rows per header with --fwd-and-reverse (kept). With that flag the FASTA parser yields each record and its reverse complement as two queries with the same header. Plain align prints a row for each, and map-only now does the same. Dropping one in map-only mode would make the two modes disagree on the row contract.

#691, test for the >16-continuation cap (not added). Exceeding the cap needs a source dummy with at least 3 $ and more than 16 distinct continuations. A test for that would mostly assert on graph construction, so I did not add one.

šŸ¤– Generated with Claude Code

Summary by cubic

Addresses review comments on the merged PRs #691 and #692.

Behavior change

  • --align-map-only is now rejected when combined with --map, --count-kmers, or --query-presence, since those flags take the k-mer mapping path and previously caused --align-map-only to be silently ignored.

Bug fixes

  • Extends the SubkSeedLabels test to run the BASIC case with a coordinate annotation and asserts that every reported coordinate is non-negative and points at the aligned sequence within its label.