Okushiri auto validation unit test broken on cyclone

[ole]

This unit test fails at present (17th January). When I tried revision 3929 it worked.

We need to bisect to find out when it stopped working

[duncan <duncan.gray@…>]

I just ran it and I got this error;

X:\anuga_validation\automated_validation_tests\okushiri_tank_validation>python v
alidate_okushiri.py
test_that_output_is_as_expected (__main__.Test_Okushiri) ... Traceback (most rec
ent call last):
  File "compare_timeseries_with_measures.py", line 305, in ?
    assert res-eps <= expected_covariance[name] <= res+eps
AssertionError
FAIL

======================================================================
FAIL: test_that_output_is_as_expected (__main__.Test_Okushiri)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "validate_okushiri.py", line 48, in test_that_output_is_as_expected
    assert res == 0
AssertionError

----------------------------------------------------------------------
Ran 1 test in 4527.049s

FAILED (failures=1)

[duncan <duncan.gray@…>]

More info on why the test failed This is from the stout file;

Validating Boundary

Covariance = 5.265269804796233900e-005

  Result is better than expected by: -4.479244941195278900e-008

  Expect = 5.269749049737429200e-005

[duncan]

Ole says it works on nautilus.

[ole]

Confirmed that it works on Nautilus.

Looking at the stdout output, it may well have to do with slight numerical differences and the way time series are compared. It is odd, though, that the test no longer passes and we need to find out what happened between changeset:3929 and now.

[duncan]

changeset:3929 fails for me, on cyclone.

[ole]

Changed the criteria and loosened the tolerances (after thorough inspections of timeseries on three platforms). The test passed on nautilus (Linux32), cyclone (Linux64) and Windows32.

This was completed in changeset:4340