Fix CIGAR size calculation for insertions

[unode]

This fixes #109 but has potential implications in several code paths (calculation of % identity, which sequence is kept, ...) so requesting a review and further scrutiny before merge.

Tests will likely fail due to #110 .

[unode]

A small update:

On a local analysis 90 samples were failing due to CIGAR errors, the patch above fixed the problem for all of them without noticeable effects to other samples.

[luispedro]

This looks all good to me.

I am not sure about the backwards compatibility implications, though, and whether it needs to be have a check for versions.

[unode]

I am not sure about the backwards compatibility implications, though, and whether it needs to be have a check for versions.

I can't really assess backwards compatibility either. I'm still surprised we never ran into this before. Wondering if there's a samtools strictness change in the latest version that is now causing these to fail.

I agree that versioning this change makes sense. Perhaps with a warning much like the one in select() for 0.8. I'll look into it.

[unode]

Given this function isn't part of NGLessIO and is used in quite a few locations, most of them pure, what's your recommendation for implementing the warning and backwards compatibility behavior?

[luispedro]

I looked at this again: the reinjection can be a hard upgrade because the previous version was wrong. The tricky part is when this is used for computing match identities and sizes. I just threaded a boolean through the code. It's an inelegant solution, but it gets the job done.

[unode]

Cool, thanks for the fix!