In improving JAXB performance, it helps to avoid unnecessary buffering.
Right now, when we unmarshal from XMLStreamReader, even though StAX API exposes
the parser-internal char buffer, JAXB still has to copy the characters into
JAXB's own buffer because the XML document might be:

  <foo>12<!-- comment -->34</foo>

When JAXB is looking at the CHARACTERS event, if we can know that the next token
is a start/end element. we can avoid copying the buffer. Looking at SJSXP source code,
I see that FragmentContentDriver actually has this information.

I wonder if SJSXP could expose this information as a proprietary extension to StAX,
so that we can take advantage of it in JAXB.
###@###.### 2005-05-11 17:31:43 GMT