First an introduction; my name is Matt Markland and I support the Cray Compiling Environment (CCE) C/C++ compiler for Cray, Inc. We (Cray) are in the process of improving our Boost support and currently have an intern doing some work for us on this path. While working on Boost.config, he has run into a testcase failure with the limits_test function test_float_limits(). With default options, CCE fails the qnan tests due to the fact that we do not conform to the IEEE specification. CCE is a little different here in that by default it runs at a higher optimization level then other compilers. Gcc will fail this portion of the test if you compile with optimization on and -ffast-math.

My question is whether it is valid to test the qnan behavior when a platform does not claim compliance? It seems to me that these checks should be wrapped in an is_iec559 check.

I think that's a valid point to make - I've disabled the test for inequality between NaN's unless is_iec559 is true. Great to see you're working on compiling Boost too! John.