[SciPy-dev] test error on Solaris using Sun's compilers

[Skip Montanaro]

    Travis> This is just a test issue.  Only single precision is used in the
    Travis> Fortran code and so the test should not be expecting more
    Travis> precision than about 6 or 7 digits.

I tried dropping decimal to 7 and then 6.  It now passes the first three
assert_almost_equal calls in test_shapiro, but fails during the fourth:

    ======================================================================
    FAIL: check_basic (test_morestats.test_shapiro)
    ----------------------------------------------------------------------
    Traceback (most recent call last):
      File "/home/skip/local/SunOS/lib/python2.2/site-packages/scipy/stats/tests/test_morestats.py", line 37, in check_basic
        assert_almost_equal(pw,0.52459925413131714,6)
      File "/home/skip/local/SunOS/lib/python2.2/site-packages/scipy_base/testing.py", line 304, in assert_almost_equal
        assert round(abs(desired - actual),decimal) == 0, msg
    AssertionError: 
    Items are not equal:
    DESIRED: 0.524599254131
    ACTUAL: 0.524600028992

Seems sort of like more than round-off error to my naive eyes.

Where do the constants used in those comparisons come from?  Presumably some
independent source and not the output of a run on another system (e.g., the
Linux/GCC combo).

Skip