[plt-scheme] Scheme productivity data point

John Clements wrote at 11/10/2009 03:26 PM:
>
> 2) I think it would be illuminating to find the crossover point.  That 
> is: how many times would you have to run this job to make the C++ 
> version better?

I actually wouldn't assume that a C++ programmer would produce a faster 
implementation.  (They might do something less efficient because, e.g., 
the greater complexity of poorer abstractions and/or having to manage 
allocations means they choose a simpler approach to keep it manageable.  
Or because a better approach is obscured to them by the details.  Or 
because they don't have time to refine their approach as they gain a 
better understanding of the problem, since development productivity is 
lower, so they are stuck with one of their earlier ideas.)

To answer your question, this little program has probably run for less 
than 2 hours wall clock time so far (it took 26 minutes to run through 
the entire corpus each of 2 times), and perhaps less than 1 hour in the 
future.

Of course, as you said, we have a lot of guessing here as to what our 
imaginary C++ programmer might do, and this is just one anecdotal data 
point.

-- 
http://www.neilvandyke.org/