Github user robertdale commented on the issue:
https://github.com/apache/tinkerpop/pull/892
I don't like this patch. If the regression tests rely on some sort order
in xml, they're doing it wrong. If it's the graph being validated, then the
test should pull it into a graph engine and validate using graph queries. If
the xml is being validated, then the test should use xml tools, e.g. xmldiff,
schema validation, etc. If the test really requires sort order, which it
shouldn't because GraphML does not guarantee any order, then it should
transform the xml with xslt before validating it.
That said, the writer already provides an optional, **deterministic** order
feature. If someone doesn't like that order, then someone else won't like
this order. So, I think the right solution would be to be able to provide a
custom comparator as a parameter and let the user create any order they see fit.
---
