clintval / unmux / 29472065806

97%

push

github

feat: split a multi-read-group BAM/SAM/CRAM by its @RG header (#2)

* feat(grammar): add @RG group source and ReadGroupKey

* style(grammar): rustfmt + lock lowercase @rg-as-file behavior

* feat(input): expose input @RG read groups and @SQ presence

* feat(readgroup): resolve @RG groups into tag sets, samples, and an RG-id map

* fix(readgroup): make two-tier routing token structurally unique

The `.`-joined label(tier1, tier2) token could collide for distinct
(SM, LB) pairs (e.g. ("sA", "l1.2") and ("sA.l1", "2") both format to
"sA.l1.2"), causing TagSet::resolve to silently misroute records
between two real samples. Fold the entry index into the token so it
is structurally unique; Sample.sample/sub_sample still carry the real
SM/LB.

* refactor(demux): open reader before compilation and resolve @RG groups

* feat(matcher): carry RG-id map on compiled @RG groups

* feat(demux): route records by RG:Z for @RG groups

* feat(demux): fan out @RG groups via resolved samples

* feat(output): emit constituent original @RG lines and preserve RG:Z

Claude-Session: https://claude.ai/code/session_016BCzogPq387cHgUdJXVqhT

* fix(output): emit constituent @RG lines in deterministic id order

build_constituent_rgs buckets read groups by iterating a HashMap, so when
multiple RGs merge into one target the emitted @RG line order varied
run-to-run. Sort the deduped constituent entries by id at emission time in
read_group_header so the output is deterministic regardless of upstream
HashMap iteration order.

* feat(demux): warn on aligned @RG input and tier-dropping merges

* feat(demux): clear error when @RG has no read groups (incl. FASTX)

* docs(cli): document the @RG group source in --help

* test: cucumber scenarios for @RG-header splitting

* fix(demux): only read the input @RG header when an @RG group is present

Reading the input @RG header on every run made a non-@RG invocation over two or more aligned inputs that share a read-group id with conflicting SM/LB ... (continued)

172 of 177 new or added lines in 6 files covered. (97.18%)

3421 of 3531 relevant lines covered (96.88%)

73.24 hits per line