Resolution of ordering conflicts is probably too liberal

Right now, all type members are considered unordered. This isn't actually the case as fields may have dependencies on other fields, and so their order does matter.

Really, an ordering conflict can only be trivially automatically resolved if at most one of the conflicting node lists contains fields. If both contain fields, one has to analyze the dependencies between these fields.

It will probably work out quite often in practice regardless, but it would be better if fields were treated with more care.

Locate the code that resolves ordering conflicts and the existing tests for type-member ordering. Trace how fields and other members are represented, then determine the expected behavior when both conflicting lists contain fields. Done means automatic resolution is restricted unless dependency ordering can be established.

Written by the indexing model from the issue text.