Hi Dan,
I think there is a bug in testRandomAngles33(); the acceptable
error limit is too low. If extractEuler(), which is called from
testRandomAngles33(), performs all its computations in single
precision, the assertion in line 229 of testExtractEuler.cpp fails.
On most systems, extractEuler() appears to compute a double-precision
arc tangent, and the assertion holds even with a very low error limit.
The Imath confidence tests should succeed if you change line 217 in
testExtractEuler.cpp to
float eps = 180.0 * 5.0 * limits<float>::epsilon();
Hope this helps,
Florian