When a CDATA element such as <style> or <script> has too much text HTMLparser
will produce the following error message:

    HTML parser error : Invalid char in CDATA 0x0

This is because this statement on line 2685 fails:

 if (((NXT(2) >= 'A') && (NXT(2) <= 'Z')) ||
     ((NXT(2) >= 'a') && (NXT(2) <= 'z')))

the problem here is that the end of the buffer has been reached, and so NXT(2)
is actually returning a null terminator or some other uninitialised value past
the end of the buffer.

I'm not sure why libxml 2.6.17 does not seem to exhibit this behaviour, even
though the code is reasonably similar. However it seems clear that there needs
to be some kind of check of whether the end of the buffer has been reached and
to refill it if necessary; perhaps one of the other character reading macros
performs this?