The new SplitNCigarReads doesn't handle existing supplementary reads correctly

[jamesemery]

The new version of SplitNCigarReads now sets reads that are split from one original read to be supplementary to each other. Unfortunately, the tool does not respect any existing supplementary reads as being supplementary to each other. Currently the tool clobbers any existing SA tag and sets one read from each read group as primary which is undesirable as it corresponds to loosing information from the aligner. One solution is to use the same "predicting" approach for existing SA information to attempt to repair the corresponding SA tag for each read based on how it would be split given its cigar string. Perhaps there is another solution that is more in line with what the SA tag gets used for.

In order to make these tags 100% accurate with overhang fixing on however, the OverhangFixingManager will probably need to be changed to store the mate information for every read it changes between file iterations. Currently it only stores information on a single read from every read group as it would be the only one that affects mate information but this assumption is invalidated as the SA tag needs to keep information on EVERY read in a canonical alignment to be correct, not just the first one.

[jamesemery]

Blocked by #2115

[chandrans]

Hey Megan @meganshand Is this related to the issue here: https://gatkforums.broadinstitute.org/gatk/discussion/comment/50915#Comment_50915 Should I get a test case from them?

[meganshand]

@chandrans I'm not sure if this is related or not, but I would love to get a test case from them. That's an unhelpful error message at the very least.

[chandrans]

Thanks Megan. I asked them for a test case and will report in a different thread if I find something that could be improved/fixed.