quoting all names, the mystery (was Re: [ANN] News Reader Stack Available)

[Dar Scott]

On Thursday, July 3, 2003, at 12:15 PM, Alex Rice wrote:

> _HyperTalk 2.2 The Book_
>
> Says on p. 45. "Enclosing literals in quotation marks can result in 
> considerable time savings-- up to 20% for loops with several hundred 
> or thousand repetitions."

I think this means "--up to 20% which can be significant for loops with 
several hundred or thousand repetitions."  The loops shouldn't affect 
the percentage.

> In this test you are accessing a UI object. I suspect that access is 
> going to skew things. We know accessing fields is much slower than 
> accessing variables... maybe the same with for buttons?

Uh.  I thought the question was related to UI objects.

>
> But I ran my own test and can't discern any noticeable difference. 
> Sometimes quoted literals is faster, sometimes unquoted literals is 
> faster. The Hypertalk Book is so clear about this. Maybe we are not 
> writing correct benchmarks?

OK.

I trimmed mine down to minimal without a loop.


on mouseUp
   set the numberFormat to "0.000000"
   put empty into field "Report"
   put the long seconds into tEnd  -- throw this one away

   -- measure ambient time ;-)
   put the long seconds into tStart
	-- measure nothing
   put the long seconds into tEnd
   put tEnd-tStart into tCorrection

   -- measure minimal quoted
   put the long seconds into tStart
   get "dummy"
   put the long seconds into tEnd
   put ((tEnd - tStart) - tCorrection) & " s " & it & " quoted" & LF 
after field "Report"

   -- measure minimal unquoted
   put the long seconds into tStart
   get dummy
   put the long seconds into tEnd
   put ((tEnd - tStart) - tCorrection) & " s " & it & " unquoted" & LF 
after field "Report"
end mouseUp


The tCorrection simply is the time it takes to put the long seconds 
into a variable which I need to subtract from my measurement.  The 
first long seconds is thrown away because it sometimes takes longer 
than the others.

My typical results are like this:

0.000005 s dummy quoted
0.000014 s dummy unquoted

There is a consistent approximately 10-microsecond difference and about 
a 1 to 3 ratio in time.

Here is where I see a mystery--
Suppose I insert this line at the start of the the minimal unquoted 
test:
   if 0 = 1 then put "cow" into dummy
The report at the end shows that 'get dummy' got empty.  If I put that 
at the end, then 'get dummy' got "dummy".
The method used seems to depend on whether a variable was created by 
that line in compiling.  If that is the case, why aren't the times the 
same?  I suspect there might be some decisions that have to wait until 
execution time related to this but don't come into play in this case.

Dar Scott