Hello All, I've been investigating an bwa issue #108[1]. I used valgrind and find two memory issues:
1) heap block overrun ==17130== Invalid write of size 4 ==17130== at 0x13753B: ksw_extend2 (ksw.c:395) ==17130== by 0x137BC5: ksw_extend (ksw.c:483) ==17130== by 0x124FFE: bsw2_extend_left (bwtsw2_aux.c:133) ==17130== by 0x125C3A: bsw2_aln1_core (bwtsw2_aux.c:283) ==17130== by 0x1278BC: bsw2_aln_core (bwtsw2_aux.c:598) ==17130== by 0x127E90: worker (bwtsw2_aux.c:660) ==17130== by 0x535D493: start_thread (pthread_create.c:333) It was an easy fix, i just allocated more memory for eh. As we always initiate eh[0] and eh[1] it seemed to be sensible to allocate at least 2 for eh. This issue didn't caused segmentation fault, but it still seemed to be right to fix it, so I created a patch fix_heap_block_overrun. 2) negative argument for malloc size ==10017== Thread 13: ==10017== Argument 'size' of function malloc has a fishy (possibly negative) value: -9223372036854775657 ==10017== at 0x4C2BBAF: malloc (vg_replace_malloc.c:299) ==10017== by 0x12E34A: bsw2_pair1 (bwtsw2_pair.c:123) ==10017== by 0x12EC37: bsw2_pair (bwtsw2_pair.c:193) ==10017== by 0x127B4E: bsw2_aln_core (bwtsw2_aux.c:621) ==10017== by 0x127E90: worker (bwtsw2_aux.c:660) ==10017== by 0x535D493: start_thread (pthread_create.c:333) Thanks to debug output I knew that this happens when l_mseq = 149, end = 0, beg = 1 Considering the line 120 in bwtsw2_pair.c : if(end - beg < l_mseq) return; the malloc on line 123 shouldn't happen at all when end=0, beg = 1, l_mseq=149. Further investigation showed that some incorrect implicit type casting happens at line 120. It seems that left part of comparison casts -1 to unsigned type. Although I don't have a clue why, as l_mseq is int, and both end and beg are int64_t. For now I created a patch with a workaround for this issue: if(end < beg || end - beg < l_mseq) return; which resolved upstream issue #108, but I'm still eager to find out why such type casting is happening. If someone with more experience with type casting in C have any ideas, please feel free to share. Regards, Nadiya Sitdykova [1] https://github.com/lh3/bwa/issues/108
