Description

bugfix for a segmentation fault in mmseqs prefilter, which occured in the following circumstances:

• seqLen of sequences larger than 32768 are overridden to 1 to lower their priority in diagonal score calculation
• min(seqs[i].seqLen - minDistToDiagonal, minSeqLen); returns an underflowed value for low seqLen, which holds for the above case (UngappedAlignment L:279)

=> this invalid seqLength value causes unrolledDiagonalScoring(...) to access memory out of bounds, causing a seg fault

Changes

• added a safety check to prevent seqLength having invalid(underflowed) values
• seqLen larger than 32768 are overridden to 0 instead of 1, in order to completely skip the processing inside unrolledDiagonalScoring(...)

Other Details

• error encountered when running mmseqs easy-cluster on uniparc DB
• stack trace:

#0  UngappedAlignment::unrolledDiagonalScoring<21u> (...) 
    at MMseqs2/src/prefiltering/UngappedAlignment.cpp:58
#1  UngappedAlignment::scoreDiagonalAndUpdateHits (...) 
    at MMseqs2/src/prefiltering/UngappedAlignment.cpp:281
#2  UngappedAlignment::computeScores (...) 
    at /MMseqs2/src/prefiltering/UngappedAlignment.cpp:349
#3  UngappedAlignment::align (...)
    at /MMseqs2/src/prefiltering/UngappedAlignment.cpp:26
#4  QueryMatcher::match (...)
    at /MMseqs2/src/prefiltering/QueryMatcher.cpp:295
#5  QueryMatcher::matchQuery (...)
    at /MMseqs2/src/prefiltering/QueryMatcher.cpp:103
#6  Prefiltering::runSplit (...)
    at /MMseqs2/src/prefiltering/Prefiltering.cpp:843