On Thursday, Aug 21, 2003, at 22:42 Europe/London, Bill de h�ra wrote:
Alex Blewitt wrote:
My point was not the parse pattern itself; you'll note that in both examples it's there. What my point was is that there's very little point in building it up in memory using a StringBuffer, only to then write out the string as a whole, when you can write out the individual parts to the stream and get the concatenation for free.
IIRC the Java text classes parse the pattern and then build a string internally as a StringBuffer, then convert it to the string all in one go. My point was that it would be possible to write a log method that did not do that, but instead wrote it out to a stream in one go.
I can provide demo code if I'm not making the point clear :-)
Alex,
Totally clear. Does this not fall under the category of small efficiencies tho'? That's assuming it is more efficient (I think it might well be with a well-written lexer).
It's in the same category of efficiencies as using a StringBuffer instead of concatenating strings with +. In any case, it does solve the problems of doing a log(message) where message has been calculated above using string concatenation.
If you have log(String pattern,Object[] data) then it really doesn't matter whether you test for isLogEnabled first; the (relatively expensive) string concatenation doesn't occur until write time.
Speaking of small efficiencies, it's still amazing how many people use 'println' to write out messages in a WebApp, even though println() is basically
try {
write(string);
write('\n\);
} catch (IOException e) {
// ignore
}In fact, the only reason that Sun introduced PrintStreams in the first place (and the esteemed println method) was (a) they couldn't get the Unicode->ASCII thing right (some would say they still haven't :-) and (b) they didn't want to force people to do a 'try/catch' in the traditional 'Hello World' application.[*]
But you're totally right; small efficiencies aren't really the big issue. Though I did pick up a tip from someone else (think it may have been Jeremy) that s.equals("") is actually slower than s.length() == 0. I did some timings and it works out between 2 times (with no JIT) and 10 times faster to do the zero length comparison. Thanks Jeremy!
Alex.
[*] Actually, the Java one is 'Hello World Wide Web' :-)
