http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50059
--- Comment #4 from Paolo Carlini <paolo.carlini at oracle dot com> 2011-08-12 18:01:43 UTC --- And I suspect that this diagnostic issue may be not unrelated to what I was noticing today together with PR50060: these badly printed intrinsics (remquo, modf, frexp) can be optimized by the middle-end but aren't really viable as constexpr functions because of the pointer parameter. Thus, by the time error.c is involved the middle-end has computed the constants and erased any info about the name of the intrinsics. Can be?