Wiener_log: raise exception for tau >= y

[randommm]

@bob-carpenter commented 3 days ago (at stan-dev/stan#1578):

"Our general convention for probabilty functions is to raise an
exception. So rather than the comparisons, this should use
another check function.

And there's never any reason to set a value and then immediately
return it. It could be written as

return stan::math::negative_infinity();

But I really think it should be replaced with something that
throws an informative exception --- the idea is to trap errors
as early as possible and let the user know where they arise.

• Bob"

[randommm]

Something strange is happening: the test ValidValues for branch https://github.com/stan-dev/math/compare/feature/issue-150-raise-exception-for-tau-greater-than-or-equal-to-y won't pass:

  [ RUN      ] AgradDistributionWiener_v_95/AgradDistributionTestFixture/0.ValidValues
  ./test/prob/test_fixture_distr.hpp:110: Failure
  Expected: ({ lp = TestClass.template log_prob <true,T0,T1,T2,T3,T4,T5> (p0,p1,p2,p3,p4,p5); }) doesn't throw an exception.
    Actual: it throws.
  Valid parameters failed at index: 0 -- (1.1000000000000001, 2.1000000000000001, 0.29999999999999999, 0.55000000000000004, 0.40000000000000002)
  ./test/prob/test_fixture_distr.hpp:110: Failure
  Expected: ({ lp = TestClass.template log_prob <true,T0,T1,T2,T3,T4,T5> (p0,p1,p2,p3,p4,p5); }) doesn't throw an exception.
    Actual: it throws.
  Valid parameters failed at index: 1 -- (2.1000000000000001, 4.0999999999999996, 1.8999999999999999, 0.050000000000000003, 0.10000000000000001)
  ./test/prob/test_fixture_distr.hpp:110: Failure
  Expected: ({ lp = TestClass.template log_prob <true,T0,T1,T2,T3,T4,T5> (p0,p1,p2,p3,p4,p5); }) doesn't throw an exception.
    Actual: it throws.
  Valid parameters failed at index: 2 -- (0.10000000000000001, 10.1, 0.050000000000000003, 0.94999999999999996, 0.5)
  ./test/prob/test_fixture_distr.hpp:110: Failure
  Expected: ({ lp = TestClass.template log_prob <true,T0,T1,T2,T3,T4,T5> (p0,p1,p2,p3,p4,p5); }) doesn't throw an exception.
    Actual: it throws.
  Valid parameters failed at index: 4 -- (0.51000000000000001, 1.1000000000000001, 0.29999999999999999, 0.20000000000000001, 10.5)

But when I run

  parameters {
    real<lower=.3001> a;
  }
  model {
    a ~ wiener(2.1, .3, .55, .4);
    1.1 ~ wiener(2.1, .3, .55, .4);
    increment_log_prob(wiener_log(1.1, 2.1, .3, .55, .4));
  }

in CmdStan, no error message is shown at all. Any ideas?

[syclik]

@randommm -- sorry, I missed this comment earlier. I'll take a closer look later.

[bob-carpenter]

@syclik Did you ever take a closer look? Is this still an issue?