------- Comment #4 from bangerth at dealii dot org 2007-09-24 14:30 ------- My reasoning would be that in the call d%g, the type of the two expressions are 'double&' and 'A&'. So to call the user-defined operator%, only the first argument has to be converted to 'A' for which a conversion constructor is available. On the other hand, to use the built-in operator%, the first argument has to undergo a floating-integral conversion, and the second argument a conversion operator call to 'long'. That's two vs one conversion operation.
That said, this is only my gut feeling, I won't have the time right now to find out the exact paragraphs, and I don't know my way around in this part of the standard very well anyway... W. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33409
