easy-search SAM alignment extends off the end of the reference

[schorlton-bugseq]

Thanks for your hard work on MMSeqs!

I ran the following using v17.b804f installed via bioconda:

mmseqs easy-search --search-type 2 --format-mode 1 --greedy-best-hits 1 query.fna hiv.fna result.sam tmp

Query:

>test
CAAAGACTGCTGACACAGACTGCTGACATGAAGATTGCTGACAGGGACTTTCCGCGGGACTTTCCAGGGGGGTGTGGTTTGGGCGGAGTTGGGGAGTGGCTAACCCTCAGATGCTGCATATAAGCAGCGGCTTCTCGCTTGTACTGGGTCTCTCTGATT

Reference: NC_001802.1

I got the following alignment:

@HD	VN:1.4	SO:queryname
@SQ	SN:NC_001802.1	LN:9181
test	16	NC_001802.1	9094	254	111M	*	0	0	AGAGAGACCCAGTACAAGCGAGAAGCCGCTGCTTATATGCAGCATCTGAGGGTTAGCCACTCCCCAACTCCGCCCAAACCACACCCCCCTGGAAAGTCCCGCGGAAAGTCC	*	AS:i:118	NM:i:13

However, I think there is a bug as the start coordinate is 9094 and alignment length 111 (from CIGAR) = end of alignment 9195, which is longer than the reference (9181bp). What I think is going on is the start coordinate is flipped with the end coordinate, as when I TBLASTX these sequences, the alignment spans ~8994 to ~9098:

OR

and when I output in SAM via blastn, I get the following:

@HD     VN:1.2  SO:coordinate   GO:reference
@SQ     SN:Query_1      LN:9181
@PG     ID:0    VN:2.16.0+      CL:blastn.REAL -query hiv.fna -db query.fna -outfmt 17 -task blastn    PN:blastn
BL_ORD_ID:0     0       Query_1 8956    255     18H37M2D104M    *       0       0       *    *AS:i:168        EV:f:1.36602e-40        NM:i:2  PI:f:85.11      BS:f:152.769

Not sure if related to #845. I tried reviewing the source but couldn't spot the issue and it looks like MMSeqs tries to do the correct thing (as per SAM spec) here,

MMseqs2/src/util/convertalignments.cpp

Lines 733 to 734 in 8ef870f

	int start = std::min(res.qStartPos, res.qEndPos);
	int end = std::max(res.qStartPos, res.qEndPos);

, so I don't know - any ideas? Thanks again!