armv4l bootstrap problem in libstdc++-v3

[Richard Earnshaw]

> 
> OK, fixed with this patch.  The problem *was* in the ARM back-end, in
> a manner of speaking.  The C++ front-ends wants the TYPE_ARG_TYPES for
> *all* functions that do not take variable numbers of argumenst to end
> with `void_list_node' -- not an equivalent node.  (That's used to
> marginally speed up things in the C++ front-end, and it saves memory.)
> 
> So, when making builtins you have to play along...

This, of course, was the vital bit of information I was missing when 
trying to construct a test case that failed with a standard builtin :-(

Many thanks for looking into this.  All I need to do now is fix all the 
other bugs that have recently started to cause bootstrap failures....

R.