Eventually this might help fix #310 . Right now all it does is replace the SQL text comparison in the test suite with comparison of the underlying values, mostly without the SQL (the schema ddl is the exception).
This should enable use of a different floating point comparison logic which could allow the near-epsilon differences that seem to come up to pass through without failing the tests. That part is not implemented yet. I cannot reproduce these failures on Linux so I want to see some runs of this code on CI before proceeding.