>Hi Peter,
>
>try this,
>

>create cursor c_urls ( url c(100), thetime t DEFAULT DATETIME())
>
>PRIVATE lcAll, ln ,aUrl
>
>lcAll   = filetostr( 'index.dat' )
>
>* next line is for test
>*lcAll='AAhttp://1'+CHR(0)+'fsdfsdSDFFSDFSDFfhttp://2'+CHR(0)+'sdsdfBFGHDGHGHDGHsdf'
>*FOR K=1 TO 5
>*lcAll=m.lcAll+m.lcAll
>* NEXT
>*? LEN(lcAll)
>** END
>SET ESCAPE ON
>t       = seconds()
>for ln = 2 to ALINES(aUrl, SUBSTR(m.lcAll, AT('http://',m.lcAll)),'http://')
>   STORE AT(CHR(0),aUrl[m.ln]) TO lnNIL
>   if m.lnNIL > 1
>	   insert into c_urls values ( "http://"+LEFT(aUrl[m.ln],m.lnNIL-1), datetime() )
>   ENDIF
>next
>? ln -1, seconds() - t
>

Wow! This code is almost as fast as the RegExp algorithm of Gregory Adam (Re: VFP versus C++ Thread #842594 Message #842893).
They are both under 5 seconds on my machine and the major part of those seconds is used for the FOR/NEXT, which is essentially the same for both.

The interesting thing about your algorithm is that it's pure VFP code. It suggests that it's not only a matter of choosing the right toolbox, but also of choosing the right tools from the toolbox. And obviously VFP has at least one feature on board that does the job we're investigating perhaps (almost) as quick as C++.

For me personally your use of ALINES() is interesting, since I actually use it in the application (VFP versus C++ Thread #842594 Message #842594) in several ways. Basically, what I do is:

do while not feof( lnHR )
   lcC = fgets( lnHR )
   lnFieldsFound = alines( laField, lcC + chr(9), chr(9) )
   for ln = 1 to lnFieldsFound
      *  manipulate laField[ ln ]
   next 
enddo