Create standardized, interpretable exceptions for error reporting

[tdanford]

I think that #297 is indicative of a slightly larger problem -- the way it manifests is that, when you run partitionAndJoin with some non-mappable records (i.e. records that can't be turned into a ReferenceRegion), you get a NullPointerException being thrown by an un-intelligible piece of code.

The challenge isn't just to catch this error condition earlier, it's to throw exceptions whose names/messages are interpretable to the end-user. So, for example, we could have an UnmappedException exception, which is thrown whenever "something that should be mappable, isn't."

Other ideas for exceptions --

• invalid sample identifier
• insufficient read coverage (or data coverage?)

thoughts?

[fnothaft]

@tdanford this sounds like a heavyweight fix where a lightweight fix would do. If we throw an NPE when we create a reference mapping for a read that is unmapped, it would be sufficient check whether the read is mapped, and to throw an IllegalArgumentException if the read is unmapped which clearly states that the read is unmapped.

To clarify my opinions, I don't have any problems with custom exceptions; it's just that standardized, interpretable exceptions framework sounds like SkyNet for exceptions, and is probably out of scope for us to build.

[tdanford]

Yeah, I initially used an IAE for the other fix.

I don't think this needs to be "skynet," just some standardized bioinformatics-specific names for exceptions that we could throw. Otherwise, you're packing it all into the message of the IllegalArgumentException and (it seems to me) you'll end up regex'ing it out sometimes.

I dunno. Just a thought.

[tdanford]

I took the word "framework" out of the title, does that sound a little less heavyweight now? :-D

It's basically about what you want to dispatch on. There are a lot of things that could throw an NPE inside some of this code.

[fnothaft]

+1; I think this:

I don't think this needs to be "skynet," just some standardized bioinformatics-specific names for exceptions that we could throw. Otherwise, you're packing it all into the message of the IllegalArgumentException and (it seems to me) you'll end up regex'ing it out sometimes.

Is a reasonable goal.

[tdanford]

This commit c8486eb has an example of what I'm intending.

[calvertj]

Any idea of what exceptions you would like? I think we could fold:
adam-core/src/main/scala/org/bdgenomics/adam/exceptions/UnmappedException.scala
into a
adam-core/src/main/scala/org/bdgenomics/adam/exceptions/Exceptions.scala file.
Or would you prefer a file per exception?

[laserson]

I like single file, as long as it doesn't get unwieldy