JB>> There's nothing in the C++ Standard to say that it shouldn't (in 
 JB>> fact there is no no concept of "warnings" in the C++ Standard 
 JB>> at all), and portability enthusiasts would argue that a good 
 JB>> quality implementation *should* give such warnings.  After all, 
 JB>> it is not guaranteed that values of integral type can be 
 JB>> converted to values of floating-point type without loss of 
 JB>> precision and silent conversion to the next higher or 
 JB>> lower floating-point value that can represented.
 TM> I  agree  with  you  but I can't remember having met such a situation.
 TM> Have you?
You mean aside from the code that you originally posted, which did indeed 
produce such warnings, as Jan Bijsterbosch pointed out ?   (-:
Try converting from a 32-bit integer type to a single-precision floating 
point type on Intel x86 architectures.  See what happens, in terms of loss of 
precision, when the integer is greater than 2 to the 24.
 ¯ JdeBP ®
--- FleetStreet 1.19 NR
---------------