Hmmm, for what it's worth, Lark, both in its current form and
after the big performance update coming Real Soon Now, works
at approximately equal speed off byte and character streams...
the overhead of pouring a buffer's worth of bytes into the
internal character buffer that Lark will be reading from
is hardly detectable. The next Lark will not be as fast as XP but
it won't be that much slower. I'd be interested in what the other
parser builders have done in this area. -T.