Re: On Case and Performance

Tim Bray (tbray@textuality.com)
Sat, 06 Sep 1997 14:50:37 -0700

At 02:34 PM 06/09/97 -0700, Jon Bosak wrote:
>| Resulting performance improvement in Lark, in processing the XML spec:
>| a factor of 11.9.
>
>Now you've made me too curious to resist asking. What's the
>performance difference if you just compare codes directly and don't
>bother with case folding?

To test that I'd have to go regularize the case of all the tags in the
XML spec which <subtext>seems like an unreasonable amount of work</subtext>.
Anyhow, the routine that checks whether we've seen a GI (where this stuff is)
is taking 8.7% of the total time. So the gain from skipping the monocasing
entirely is not going to be dramatic. In fact, it's now spending more
time in BufferedInputStream.read() (oh for a good old-fashioned getc()
macro). -Tim