On 10/20/2009 04:50 PM, Eric Blake wrote:
An alternate patch would have been reordering the shell for loop to test long long, long, then int; that way, the checks are ordered widest to narrowest. But that has the potential for API changes if the macro selects a different type (but equivalent width) than what it did before the reordering, and it also penalizes the time taken for selecting the type to use for uint32_t.
ABI, actually, since it changes the C++ mangling. Paolo
