ambassador at fourthworld.com
Fri Dec 5 13:10:27 EST 2003
>> What you're testing is a very fast operation. I found that if I
>> increase the number of iterations by one order of magnitude the
>> variance drops proportionately.
> Yes, sure. But sometimes the differences are such
> that testing for tweaking a handler is not really reliable.
In-place profiling is hard. I wrote RevBench to allow easy comprison of
snippets, but in many cases I need to test whole sequences of handlers.
With the HyperRESEARCH application I develop for ResearchWare, last year I
set out to remove some bottlenecks in legacy code. The best approach I
found at the time was to add calls to a logging handler at he beginning and
end of handlers I was profiling, noting the handler name, params, and
millisecs. It took a few minutes to set up but told me what I needed to
know. Before shipping I just commented the logging handler so it does
nothing; if I need it again in the future it's one de-comment away. :)
Not the most graceful solution but it got me through the day in the asence
of a built-in profiler.
Fourth World Media Corporation
Ambassador at FourthWorld.com http://www.FourthWorld.com
Tel: 323-225-3717 AIM: FourthWorldInc
More information about the Use-livecode