adds support for bam input, with tag preservation

Makefile

@@ -5,7 +5,7 @@ LDFLAGS=
 INCLUDES=
 LOBJS=		kommon.o kalloc.o bwt.o l2bit.o options.o seed.o map-algo.o lchain.o align.o pe.o cs.o format.o \
 			ksw2_extz2_sse.o ksw2_extd2_sse.o ksw2_ll_sse.o
-AOBJS=		kthread.o libsais.o libsais64.o index.o bseq.o map-main.o fastmap.o
+AOBJS=		kthread.o libsais.o libsais64.o index.o bseq.o bam.o map-main.o fastmap.o
 MALLOC_O=	mimalloc.o
 PROG=		minibwa
 LIBS=		-lpthread -lz -lm
@@ -65,7 +65,8 @@ depend:
 
 QSufSort.o: QSufSort.h
 align.o: mbpriv.h minibwa.h l2bit.h bwt.h kommon.h bseq.h kalloc.h ksw2.h
-bseq.o: bseq.h kommon.h kseq.h
+bam.o: bam.h bseq.h kommon.h
+bseq.o: bseq.h kommon.h bam.h kseq.h
 bwt.o: kommon.h kalloc.h bwt.h
 bwtgen.o: QSufSort.h
 cs.o: mbpriv.h minibwa.h l2bit.h bwt.h kommon.h bseq.h kalloc.h

NEWS.md

@@ -1,3 +1,34 @@
+Unreleased
+----------
+
+Notable changes:
+
+ * New feature: BAM input. Minibwa reads BAM wherever it reads FASTA/FASTQ,
+   detecting the format from the file contents. Both unaligned and aligned BAM
+   work; for aligned input the old alignments are discarded and secondary and
+   supplementary records are skipped. Paired-end mode is turned on
+   automatically when the first record is flagged as paired.
+
+ * Bugfix: in `--meth` mode the bisulfite conversion was chosen from a read's
+   position within its fragment, so an unpaired read 2 -- which a BAM can carry
+   but a pair of FASTQ files cannot express -- was converted as if it were
+   read 1 and usually failed to map. With a BAM input the SAM FLAG now decides,
+   and the segment a read belongs to is carried through to the output FLAG, so
+   an unpaired read 2 is still reported as read 2 rather than as single-end.
+
+ * New feature: paired-end input is now validated. A coordinate-sorted BAM is
+   rejected in paired-end mode instead of being mapped as single-end, and two
+   query files whose reads are out of order are reported rather than silently
+   mispaired.
+
+ * New feature: with a BAM input, `@RG`, `@PG` and `@CO` header records and
+   per-record auxiliary tags such as `RX` are carried over to the output.
+   `@HD` and `@SQ` are replaced, as contigs come from the index, and tags
+   describing an alignment are dropped because minibwa recomputes them. Use
+   `--bam-tags=no` to copy no tags.
+
+
+
 Release 0.7-r421 (6 August, 2026)
 ---------------------------------
 

README.md

@@ -93,6 +93,39 @@ Note in the default adaptive mode, `-g`/`-w`/`-W`/`-N`/`-m`/`-s` only changes
 the short-read setting; the long-read setting is fixed. This mode is disabled
 with `--adap=no` or when `-x sr` or `-x lr` is specified.
 
+#### BAM input support
+
+Minibwa reads BAM as well as FASTA/FASTQ; the format is detected from
+the file contents, so no option is needed. Both unaligned BAM (uBAM) and
+aligned BAM work — for aligned input the existing alignments are discarded and
+the reads are re-mapped. 
+```sh
+minibwa map -t8 ref.fa reads.ubam > aln.sam  # read pairs interleaved in one uBAM
+minibwa map -t8 ref.fa r1.bam r2.bam         # read 1 and read 2 in separate BAMs
+minibwa map -t8 ref.fa - < reads.ubam        # BAM on stdin
+samtools collate -@4 -Ou in.bam | minibwa map -t8 ref.fa -  # collate first
+```
+If the first record of a BAM is flagged as paired, paired-end mode is turned on
+automatically; `--pe=no` overrides this. Records must be collated by name, as
+`samtools collate` or `samtools sort -n` produces: mates are found among
+adjacent records, so minibwa rejects a coordinate-sorted BAM in paired-end mode.Secondary and supplementary input records are skipped so that a read is mapped only once.
+
+When read 1 and read 2 are given as two files, mates are paired by position.
+Minibwa checks that the names agree and aborts if the files are out of sync.
+
+The primary advantage of bam input mode is that header (e.g. RG, SN, BC) and read tags (e.g. RX, BC, MI) are carried over from the input avoiding the need for a later step to stich these attributes back in.
+BAM: `@HD` and `@SQ` are replaced (contigs come from the index), while `@RG`,
+`@PG` and `@CO` are preserved, with `@RG` deduplicated by ID across inputs and
+minibwa's own `@PG` chained via `PP:`. Tags that
+describe an alignment (`NM`, `MD`, `AS`, `SA`, `MC`, `MQ`, `XA`, ...) are
+dropped, as minibwa recomputes them. Use `--bam-tags=no` to copy no tags at
+all. `-R` overrides any `RG` the records already carry, and replaces the input
+`@RG` header lines.
+
+BAM decompression is single-threaded and could become a bottleneck at high
+thread counts. Piping in uncompressed BAM, as `samtools collate -Ou` above
+does, avoids most of that cost.
+
 #### Mapping with legacy bwa-mem CLI
 
 Minibwa also provides legacy bwa-mem command-line interface (CLI) via the `mem` subcommand.