What parts of R5RS are generally considered not worthwhile to implement?

I have been looking around for good, small Scheme implementations. I notice it is very usual that they claim to follow "almost all" of R5RS, but never all of it. 开发者_开发百科What parts of R5RS are usually considered too hard/not worthwhile? Are they considered a mistake of R5RS?

If you're looking for things under the too-hard/not-worthwhile category, there are several tricky details of implementing continuations in various places (and some small implementations don't have support for real continuations either). Take a look at this (now outdated) chart that the SISC authors put together: http://sisc-scheme.org/r5rs_pitfall.php Some systems, while claiming to implement "(most) everything", have a few overlooked details like that.

There are a few that don't implement the full numerical tower (i.e different numerical types: complex, rational, exact vs inexact numbers).