Why even discuss this when it is so easy to look in the horse's mouth?
Similar questions keep coming up. Do people not know how to test this
stuff?
1. Create a loop that does Method 1 enough times so it becomes
measurable.
2. Create a 2nd loop doing Method 2 the same number of times.
3. Create a control loop that does all the same looping and set up but
doesn't
actually do anything useful so you can subtract out that.
(This is just for your own sanity.)
4. Compare 1 vs. 2.
5. Is it significant? I.e., how often do you expect the code to be
executed,
and how much does each (divide by loop iterations) individual
execution cost?
a. If not much, use the most maintainable code, disregarding
efficiency.
b. If lots, use the most efficient, even if it is a tricky, obscure
technique. Comment for maintainability.
The program is easy to write. The typical metrics are TIME() and
SYSTEM(9), cpu ticks.
Create one of the cases, then DUP and change the innards as needed.
Here is an example of what I mean, comparing
- trim()
- the original method
- WOL's STEP -1 method.
(Caveat: I used the PETE string someone else in this thread made up, but
you might loop through real representative data instead.)
ITERATIONS = 10000000
PETE = 'THIS}IS }}}A TEST}}}'
*
CRT TIMEDATE(), 'Control'
T0 = TIME() ; C0 = SYSTEM(9)
FOR I = 1 TO ITERATIONS
REC = PETE
NEXT I
C1 = SYSTEM(9) ; T1 = TIME()
C.CTRL = (C1-C0) ; T.CTRL = (T1-T0)
CRT TIMEDATE(), C.CTRL, T.CTRL, REC
*
CRT TIMEDATE(), 'TRIM'
T0 = TIME() ; C0 = SYSTEM(9)
FOR I = 1 TO ITERATIONS
REC = PETE
REC< 1 > = TRIM( REC< 1 >, @VM )
NEXT I
C1 = SYSTEM(9) ; T1 = TIME()
C = (C1-C0) ; T = (T1-T0)
CRT TIMEDATE(), C, T, REC
*
CRT TIMEDATE(), 'ORIGINAL'
T0 = TIME() ; C0 = SYSTEM(9)
FOR I = 1 TO ITERATIONS
REC = PETE
MAX = DCOUNT( REC< 1 > , @VM )
FOR LOP=1 TO MAX
IF LEN( REC<1,LOP>) ELSE
DEL REC< 1, LOP >
MAX-= 1
LOP-= 1
END
NEXT LOP
NEXT I
C1 = SYSTEM(9) ; T1 = TIME()
C = (C1-C0) ; T = (T1-T0)
CRT TIMEDATE(), C, T, REC
*
CRT TIMEDATE(), 'WOL'
T0 = TIME() ; C0 = SYSTEM(9)
FOR I = 1 TO ITERATIONS
REC = PETE
MAX = DCOUNT( REC, @VM )
FOR LOP = MAX TO 1 STEP -1
IF LEN( REC< 1, LOP > ) ELSE DEL REC< 1, LOP >
NEXT LOP
NEXT I
C1 = SYSTEM(9) ; T1 = TIME()
C = (C1-C0) ; T = (T1-T0)
CRT TIMEDATE(), C, T, REC
>RUN CDS.BP TRIMTEST
13:34:41 25 MAR 2004 Control
13:34:50 25 MAR 2004 9550 9.82 THIS}IS }}}A TEST}}}
13:34:50 25 MAR 2004 TRIM
13:35:32 25 MAR 2004 39490 41.5798 THIS}IS }A TEST
13:35:32 25 MAR 2004 ORIGINAL
13:38:53 25 MAR 2004 176770 200.5888 THIS}IS }A TEST
13:38:53 25 MAR 2004 WOL
13:41:29 25 MAR 2004 153400 156.3228 THIS}IS }A TEST
>
No brainer: Use TRIM().
QED,
cds
--
u2-users mailing list
[EMAIL PROTECTED]
http://www.oliver.com/mailman/listinfo/u2-users
