Okay, I think I'm brain damaged -- nope, I figured it out

[Richard Gaskin]

Mark Wieder wrote:
> Here's mine again, but I like your benchmarks. I did this with a
> single repeat loop rather than your nested one, but otherwise I think
> we're testing the same thing. I didn't stress-test this the way you
> did with large data clumps, so I think I'd trust your conclusions for
> real-world applications.
>
> on mouseUp
>     local temp
>     local s
>     local accum
>
>     put "XXX" into temp
>     put the ticks into s
>     repeat 1000000 -- yes, that's a million
>         append temp, accum
>     end repeat
>     put the ticks - s into field 1
>     answer length(accum)
> end mouseUp
>
> command append pNew, @pString
>     put pNew after pString
> end append

No doubt passing the variable by reference is much faster than requiring 
a copy, but both will be slower than doing the work inline (which is 
often non-copying by nature) because of the overhead of the separate 
function call.

To measure that overhead I made this quick test:

on mouseUp
    -- Number of iterations:
    put 10000 into n
    --
    -- Test 1: inline:
    put the millisecs into t
    repeat n
       get 1+1
    end repeat
    put the millisecs - t into t1
    --
    -- Test 2: function call:
    put the millisecs into t
    repeat n
       get foo()
    end repeat
    put the millisecs - t into t2
    --
    -- results:
    put t1 && t2
end mouseUp

function foo
    return 1+1
end foo

Here's the result:

0.0003 0.0014

On the one hand, the overhead seems quite impressive, with more than 
four times the performance when we can do the work inline rather than 
call out to a function.

But when we consider the absolute, rather than relative, time, that 
overhead is only ~0.0011 ms per call (on the relatively slow machine I 
tested on; on a Haswell i5 I'd expect it to be a small fraction of that).

So while it can add up in very lengthy processes like the monster files 
Geoff's working on, it's more than fast enough for the convenience 
function calls offer for factored, readable code.

--
  Richard Gaskin
  Fourth World
  LiveCode training and consulting: http://www.fourthworld.com
  Webzine for LiveCode developers: http://www.LiveCodeJournal.com
  Follow me on Twitter:  http://twitter.com/FourthWorldSys