I responded to this in another thread. I agree with your assessment that the xml parser is looking for something and java is throwing a fatal error when it returns nil after being forced to look for this. The 'fragments' as they call them I believe are user code and documentation for that code. However I think the bug is the fact that they are being forced to load, when in fact they should be optional and not critical to loading RM. This, among the other facets of RM and the R extension, are at best frustrating. I don't see a whole lot of replies to this - hence, my opinion is that they either haven't looked at it, can't solve it, or are waiting until we're frustrated enough to buy a support license. In the final case the problem will still exist for Mac users, but someone will be responsible for fixing it while-U-wait and pay on the license
That doesn't add alot of value in my book (and could even appear a bit underhanded) despite the other good tenants of the software.