Hopefully this is the last iteration before commit.  As discussed, I took whitespace processing out of PyUnicode_TransformDecimalToASCII() and made it public.  Whitespace conversion in int()/float()/complex() is repetitious and can be optimized by, for example only converting leading and trailing whitespace.  I erred on the side of correctness here and real optimization will come from making conversion algorithms operate directly on Py_UNICODE characters.  This looks like a relatively easy thing to do, but is definitely outside of the scope of this issue.