Throw failed-io error

[arahuja]

At https://github.com/hammerlab/cohorts/blob/master/cohorts/load.py#L613-L619 - Is there a reason which catch this IOError and treat it the same as not having provided a VCF path?

Any objections to removing that catch and letting IOErrors bubble-up?

[tavinathanson]

@arahuja I think my logic was: I was running into issues where I had a few missing VCFs and wanted to load the others, but the IOError blew up the whole thing.

Open to changes that still allow me to load some data even if a few patients are missing?

[arahuja]

@tavinathanson It seems that's a problem when a VCF path has been specified but doesn't exist? I would think the solution to not populate snv_vcf_paths and then if it has been specified make sure we can read them.

[tavinathanson]

We discussed offline that it's okay for incorrect paths to blow up the whole thing, and any logic to the contrary can live outside of cohorts since it's counter-intuitive.