World: r3wp

just try my example

different results .. you're using 'random

let's try to compare following for R2 vs R3: 


 parse "abc^/def^/ghi^/" [start: any [end: newline (print copy/part 
 start end) start: | skip]]

well, just remove (print ...) ... I just want to find out the traversal 
time ....

I normally just parse short bits of text .. so R2 looks to be 2x 
faster than R3

s: now/time/precise loop 1000000 [parse "abc^/def^/ghi^/" [start: 
any [end: newline start: | skip]]] now/time/precise - s

Three consecutive runs:
R2:
0:00:03.11
0:00:02.941
0:00:03.089

R3:
0:00:01.732
0:00:01.679
0:00:01.724

the above code does not run in R3. Dunno if "end" is forbidden or 
became a keyword, but I get following error, if I don't change end: 
to en:


>> s: now/time/precise loop 1000000 [parse "abc^/def^/ghi^/" [start: 
any [end: n
ewline start: | skip]]] now/time/precise - s
** Script error: PARSE - command cannot be used as variable: end:
** Where: parse loop
** Near: parse {abc
def
ghi
} [start: any [end: newline start: | skip...

So - for basic traversal, R3 is faster .... for my example ....

even with my random generated data, the number of parts should be 
always the same.

But you can also parse the same data a few times. Every times it 
will be parsed in a different number of parts

end is a keyword!

The data I parse, are oftly in the 10s of MB and more.
With R2 I had no problems

>> start: now/time/precise loop 1000000 [parse "abc^/def^/ghi^/" 
[s: any [e: newline (append blk copy/part s e) s: | skip]]] now/time/precise 
- start
== 0:00:05.359

Hmm... tried to parse 4M chars a few times and r2 ran out of memory 
... r3 worked fine

There really is some problem:

R2:

>> start: now/time/precise loop 1000000 [parse "abc^/def^/ghi^/" 
"^/"] now/time/precise - start
== 0:00:00.933

R3:
==  0:00:04.259

I would curecode it, along with my example, and ask Carl about the 
possible cause ...

I did it already.

But your example does not show the real problem. just use my example 
and you will see why.

it shows very big time penalty ...

... which does not correspond to my manual parse example above ...

WTF? I get strange beeps when running your test code :-)

try running following code:

c: ""
 for i 0 255 1 [append c to-char i]

I get a beep here ....

each time I parse the dataset  ... I get it alternating between two 
lengths

I have no speaker at my pc.)

hmm, we are appending char into string ... who knows what it does 
internally :-) In charset, there are some escape/control codes, so 
maybe it causes a beep, I don't know :-)

I reported the strange beep into CC too ...

I copied the dataset before each parse ... very odd

Pekr, are you sure it's not just control chars printing to windows 
console causing beep?

could be a GC problem? Maybe above "beep" means there might be some 
memory leak?

maybe ... who knows ... should not beep ....

it surely is about control chars ...

why should it not beep?

Petr, most likely you get the beep when the string is printed to 
the windows console.

R2 does not beep (but it has its own console)

that is, windows does respect the BEL character.

ie. try print "^(07)"

yes, but I am denerved by that beeping :-)

and this is the windows console

don't include that char in your tests then, or don't print it out 
;)

Yes, it does so ... I will suggest the ticket dismissal ...

more unnerving is the non reproducible parse!

Graham - sqlab's code uses random sets, no? What is non reproducible 
exactly?

parse dataset "^/" returns a different value

for each consetutive run on the same dataset?

consecutive

length? parse data "^/"

do that a few times

>> parse "abc^/def^/ghi^/" "^/"
== ["abc" "def" "ghi"]

I get three value block for each run ...

on the large set ... not small sets

OK, I will try ...

he just generates a large set of random chars