Problems running Boost Math tests on AIX + GCC

[Sohail Somani]

I've been running the Boost Math tests on AIX and nearly all of them fail with a floating point error.

For example, from test_bessel_i, I get:

boost/math/tools/test.hpp(213): error in "test_main_caller( argc, argv )": Unexpected exception. CAUTION: Gross error found at entry 2. Found: 0 Expected 2.27959 Error: 1.93926e+38 0

boost/math/tools/test.hpp(242): error in "test_main_caller( argc, argv )": Gross error 0, 4, 11.301922

....

Are there any special instructions for using the library on AIX? I couldn't see any in the Jamfile. The version of GCC here is 4.3.4

[John Maddock]

Unfortunately I don't have access to an AIX machine, so I can't really test this :-(

I believe that pretty much everything works OK with IBM's compiler though.

Can you let me have the full output of the regression test run - the output above seems mixed up somehow.

The usual solution for gcc problems is to define BOOST_MATH_NO_LONG_DOUBLE_MATH_FUNCTIONS in boost/math/tools/config.hpp which may be worth a try in this case, it would be nice to get to the cause of the problem though.

You can email me direct at john@… if it's easier than using the interface here as well,

Cheers, John.

[Sohail Somani]

Setting the define does make the tests start passing. I'll see if we need long double support or not to see if I should test further. Thanks a lot!

[John Maddock]

(In [61806]) Disable long double support when the C++ runtime does not support it either. Fixes #3825.