missing mutation parents is not detected

[petrelharp]

If a set of tables doesn't have mutation parents computed, one can do many things with the tree sequence without this ever raising an error. This could lead to wrong answers somewhere; it seems like this should be detected by check_integrity? And, maybe checked at load time?

Here's a test:

    def test_no_mutation_parents(self):
        tables = tskit.TableCollection(sequence_length=1)
        tables.nodes.add_row(time=0, flags=tskit.NODE_IS_SAMPLE)
        tables.sites.add_row(position=0, ancestral_state='A')
        tables.mutations.add_row(site=0, node=0, derived_state='X', time=0.5)
        tables.mutations.add_row(site=0, node=0, derived_state='C', time=0.5)
        ts = tables.tree_sequence()
        # ts.tables.mutations.parent, if computed, would be [-1, 0]

This would most easily be checked by verifying that the mutations.parent column is identical to what compute_mutation_parents would return.

[jeromekelleher]

I think we went around this originally and came the to the conclusion that this was a bit icky, but there's not much that could be done. If we run compute_mutation_parents each time we create a tree sequence from a set of tables, then there's no real point in storing the mutations.parent column at all.

[petrelharp]

Oh, right. This does sound familiar. Sounds good.