Low-level/general solution for --ignore-variants-starting-outside-interval?

[bbimber]

When running GATK with specific interval(s), the default behavior is to include any variant spanning those interval(s). When running scatter/gather jobs, this behavior is generally not what one wants, since this would result in variants spanning the job intervals getting included twice.

In a handful of GATK tools, there is support for something like --ignore-variants-starting-outside-interval, which is probably designed to solve this problem. GenotypeGVCFs supports this. However, the implementation/support is generally tool-level and I dont believe all tools support this. For example, SelectVariants does not appear to. If one wants to run a scatter/gather task that doesnt start with a GATK tool that supports --ignore-variants-starting-outside-interval, you're out of luck.

My questions are:

1. Am I completely missing some existing capability?

2. There is already some low-level support in the engine for control over intervals. Would you be receptive to a PR that pushes support for "--ignore-variants-starting-outside-intervals" lower into GATK? Perhaps into VariantWalkerBase? One possibility would be to create a StartsWithinIntervalsVariantFilter, and override makeVariantFilter() to inject it. I dont think this would be particularly invasive, and could be pretty useful across many tools. As part of this, MultiVariantWalkerGroupedOnStart's argument would get merged with this.

[droazen]

@bbimber Yes, this is an excellent suggestion and would be very useful! We actually already do have an open PR that adds such a general argument for VariantWalkers, but it's been languishing for a while: #6388

I'll see how difficult it would be to resurrect this old PR.