A discussion came up in PR #774 regarding the correct behavior when parsing an mmCif file without auth_seq_id.

BioJava 4.2.11 requires the auth_seq_id column. This is a problem because it is optional according to the spec and omitted by PyMOL.

In b207d34 I added code to use the label_seq_id column for creating the ResidueNumber for each group if auth_seq_id is missing. There was some concern that this could lead to inconsistent residue numbers if some residues used '?' (defaulting to label_) while the rest used the auth_ values specified. This worry is actually not justified due to another bug, which causes a NumberFormatException if '?' is used in that column.

@josemduarte suggested only doing the label_ fallback if ALL groups have null ResidueNumbers. This is probably the right solution, but it seems like such an edge case it might not be worth the hour it will take to fix it.