On Monday, July 26, 2004, at 04:02  AM, Jay Reeve wrote:

> On Monday, July 26, 2004, at 01:57  AM, Larry Siebenmann wrote:
>>  --- (maybe) a FB flag that indicates that
>> the array has been changed since the last pointer array build.
>> If this flag is zero, then the lookup related functions
>> should use the pointer array.  The build mentioned
>> could reasonably zero this flag.
> I can work on an override to do this--but not tonight.

Larry,

I have some exciting news for you. Your proposal sparked a compromise 
solution, using a pointer list but with just one pointer for every 16 
handles. Much easier and faster to maintain while still giving most of 
the benefits of the pointers.

My tests show my new approach ranges from marginally slower to 
outrageously faster for the various INDEX$ functions. You'll be pleased 
to hear that reading 10,000 strings from the array dropped from 1.570 
seconds to 0.006 seconds!!!

Here's the whole results table. Times are in seconds.

1. Initialize INDEX$(0) to 200000 bytes.
	Mine = 0.000		FB = 0.000

2. Insert a short string at the beginning of the array 10,000 times:
	Mine = 2.342		FB = 0.403		--Worst case for mine : (

3. Replace each of the 10,000 short strings with a longer one.
	Mine = 0.672		FB = 2.516

4. Search every 10th  string for a word it contains.
	Mine = 0.047		FB = 0.055

5. Search for an unfindable string, starting with each of the 10,000 
strings.

6. Insert a string before each of the 10,000 strings, deleting the top 
one each time.	Mine = 6.001		FB = 4.815

7. Read each of the 10,000 strings into a variable.
	Mine = 0.005		FB = 1.153           --!

8. Read the 10,000th string into a variable 10,000 times.
	Mine = 0.005		FB = 2.612           --!!!!!!!!!!!!!!

9. Delete the first string in the array 10,000 times (forcing all the 
others to move down):
	Mine = 0.612		FB = 0.371

I'll send you my Subs Index.incl back channel, along with my test 
routine. (Also to anyone else who requests.) Run the test routine and 
record your results for the standard FB incl.  (Click on the first 
column head in the profiler to see them in order.) Then drop my incl 
into your User Libraries folder and run the routine again.

This still needs thorough testing. It appears to be working correctly, 
but there could be major bugs yet. If you have a program that makes use 
of INDEX$ strings, please run it with this and watch for any anomalies.

Any opinions on submitting this as a permanent replacement in R2?

  e-e
  =J= a  y
   "

--
To unsubscribe, send ANY message to: futurebasic-unsubscribe@...