Tree.tmrca error when MRCA does not exist

[hyanwong]

When an MRCA does not exist between two nodes A and B (e.g. in deleted region), the tmrca(A, B) function errors out with ValueError: Node index out of bounds, which is not terribly informative.

This implies to me that this edge case has not been tested. I guess we could either have a more sensible error message, or return NaN or something?

[jeromekelleher]

I think we never figured out what the right thing to do is here, and then forgot about it.

A more meaningful error message is probably the least bad thing at this point, as people might be depending on the ValueError being thrown

[hyanwong]

Yes, throwing a meaningful error SGTM. If you don't want to catch the error you can simply use Tree.mrca and look for tskit.NULL. Worth having the unit test in for .tmrca either way, I guess.

[hyanwong]

Actually, in #2070 (comment) you suggested returning inf. I think that might be misleading (after all, we don't know the tMRCA here, but it's probably finite, so NaN is more descriptive). Returning a ValueError gets around the question of what we should return.