Github user svanteschubert commented on the issue:
https://github.com/apache/tinkerpop/pull/892
a) Regression testing seems easier when there is a deterministic way of
serialization, in case this is already, the normalization is fine to be
optional but at least reachable by API via Gremlin console for GraphML
serialization. Would you agree to this? Otherwise, we can dismiss this patch.
Could you provide some console access for the normalization, Stephen?
b) Shall we check for integer and/or long and do the fall-back to the
simple string comparison or what would be your suggestion? Remember, the
integer ID is the default key of graphs created by Tinkerpop graph and the
current normalization is far from the human way of sorting integer. This seems
important to me, as GraphML is especially easy to view and edit the XML and do
testing based on this.
Again, if you do not like this at all, let's dismiss simply the patch and
save us the time. ;-)
---
