Test are in files named NNNNN_A_*.  They are run in lexically sorted order.

The "A" in the name is one of:

  s   The test succeeds in altering the state of the database
  e   The test produces an error, and database state is unaltered
  n   The test does not alter database state

The database state established by earlier tests is used in later tests.

Note that sequence values can always be affected, for values "s" and
"e".  The "n" tests take special care to reset sequences to their
initial state.

The test/unsupported/ directory contains tests that the current
regression testing system cannot support.  This is because the output
necessarily contains timestamps containing the current time, which
necessarily differ when comparing the results of different test runs.


In the FWIW category, this is the logic, at present, behind the
numbering of the tests:

The tests start by testing per-table, with gaps in the sequence
between tables.  The tables that hold foreign keys used in other
tables are tested before the tables that use the keys.  The successful
inserts are left in place for use by later tests.  Within any one
table, an insert is first attempted, then tests of each of the columns
in the order in which they are defined in CREATE TABLE.  Then tests of
inter-column interaction within the one table.

Then there is a gap and testing of inter-table interactions.

Then another gap, all generated ISOK_RESULTS are deleted and the
ISOK_RESULTS sequence reset, and the functions are tested.  First, the
action of the function on columns is tested.  Then a gap and
exceptions are tested.  Then a gap and and testing of what happens
when saved queries generate invalid results.