The accuracy of the expression evaluation code was falsely downscaled in the code.
The false test did not pop up because of a bug in FortUTF. See #18 (closed)