Re: information_schema and not-null constraints

Alvaro Herrera <alvherre@alvh.no-ip.org> writes:
> In 0002, I took the tests added by Peter's proposed patch and put them
> in a separate test file that runs at the end.  There are some issues,
> however.  One is that the ORDER BY clause in the check_constraints view
> is not fully deterministic, because the table name is not part of the
> view definition, so we cannot sort by table name.

I object very very strongly to this proposed test method.  It
completely undoes the work I did in v15 (cc50080a8 and related)
to make the core regression test scripts mostly independent of each
other.  Even without considering the use-case of running a subset of
the tests, the new test's expected output will constantly be needing
updates as side effects of unrelated changes.

            regards, tom lane