Re: Less greedy pattern match

On 4/20/2008 9:30 AM, Janis Papanagnou wrote:
> pk wrote:
> 
>>On Saturday 19 April 2008 18:53, Janis Papanagnou wrote:
>>
>>
>>
>>>>{for(i=m+1;i<=NF;i+=2){printf "%s",$i};
>>>>if (NF) {m=(NF%2==m)?1:0}
>>>>if (m==0) print""}
>>>>
>>>>Not elegant as the gawk solution, but at least it should be standard.
>>>>
>>>
>>>Does removing some more characters make it a bit more elegant...?
>>>
>>>  {for(i=m+1;i<=NF;i+=2) printf "%s",$i}
>>>  NF {m=(NF%2==m)}
>>>  !m {print""}
>>
>>
>>Well yes...would that improve performance somewhat too?
> 
> 
> Haven't checked. Would be interesting to know, though; because
> generally I prefer  cond{action}  instead of  {if(cond)action}
> constructs when using awk (though primarily not for efficiency
> reasons).

Run on a 1-million line file with 11 fields per line where each line
starts with
the line number (ran each command 3 times, results concatenated):

$ time awk '/^1/{print}' big1mil > /dev/null
real    0m0.859s user    0m0.749s sys     0m0.124s
real    0m0.843s user    0m0.733s sys     0m0.155s
real    0m0.844s user    0m0.796s sys     0m0.077s

$ time awk '{if (/^1/) print}' big1mil > /dev/null
real    0m0.860s user    0m0.780s sys     0m0.093s
real    0m0.875s user    0m0.796s sys     0m0.093s
real    0m0.937s user    0m0.811s sys     0m0.093s

$ time awk '/^1/{print}' big1mil > /dev/null
real    0m0.844s user    0m0.780s sys     0m0.093s
real    0m0.890s user    0m0.811s sys     0m0.077s
real    0m0.922s user    0m0.827s sys     0m0.046s

-----------------------------------------------

$ time awk '{m=(NF%2==m)?1:0}' big1mil > /dev/null
real    0m1.609s user    0m1.405s sys     0m0.108s
real    0m1.485s user    0m1.405s sys     0m0.109s
real    0m1.484s user    0m1.436s sys     0m0.077s

$ time awk '{m=(NF%2==m)}' big1mil > /dev/null
real    0m1.469s user    0m1.327s sys     0m0.155s
real    0m1.484s user    0m1.405s sys     0m0.109s
real    0m1.469s user    0m1.405s sys     0m0.077s

$ time awk '{m=(NF%2==m)?1:0}' big1mil > /dev/null
real    0m1.469s user    0m1.452s sys     0m0.046s
real    0m1.484s user    0m1.436s sys     0m0.061s
real    0m1.485s user    0m1.390s sys     0m0.108s

Can't say I see any conclusive evidence for either approach being more
efficient, but I may be missing something.

Like Janis says, though, efficiency isn't the main reason to structure
your awk
code to look like awk...

	Ed.