pmc build & memory corruption

[cbaakman]

Segmentation faults frequently occur while mrs is indexing the pubmed databank, using the raw xml files from ftp://ftp.ncbi.nih.gov/pub/pmc/articles*.tar.gz as input. The location of the segfaults vary, but they always co-occur with an allocation or deallocation. Inspection of some of the deallocation segfaults at https://github.com/cmbi/mrs/blob/master/src/M6Lexicon.cpp#L213 shows that the involved pointers are valid.

Facts about the update process:

• MRS always reports invalid checksums for the pumbed tars, though they are readable by commandline tar.
• pubmed updating occurs in multiple threads. One thread per archive.
• MRS uses a custom XML parser for pubmed
• The pubmed xml files are much larger than the average pdb file.