Re: Re: Re: Re: Re: should this backspace removal code be done better?

[eyepopslikeamosquito]

Not any more. This new smack is a lot faster by eliminating the alternation.

#!/usr/bin/perl -w

use Benchmark qw(cmpthese);
$s = "\bthis is an\b correct\b\b\b usage\b";

sub uk {
  $a = $s;
  $a =~ s[(?:[^\cH]\cH|^\cH)][]g while 1+index $a, chr(8);
  $a;
}

sub new1 {
  $a = $s;
  while ($a =~ s/(?:[^\cH]\cH|^\cH+)//g) {}
  $a;
}

sub smack {
  $a = $s;
  $a =~ s/^\cH+//;
  1 while ($a =~ s/[^\cH]\cH//g);
  $a;
}

cmpthese(-5,{uk=>\&uk,new1=>\&new1,smack=>\&smack});
$s x= 100;
cmpthese(-5,{uk=>\&uk,new1=>\&new1,smack=>\&smack});

Benchmark: running new1, smack, uk, each for at least 5 CPU seconds...
      new1:  6 wallclock secs ( 5.39 usr +  0.03 sys =  5.42 CPU) @ 11
+344.28/s (n=61486)
     smack:  6 wallclock secs ( 5.24 usr +  0.02 sys =  5.26 CPU) @ 36
+814.64/s (n=193645)
        uk:  4 wallclock secs ( 5.26 usr +  0.06 sys =  5.32 CPU) @ 13
+261.09/s (n=70549)
         Rate  new1    uk smack
new1  11344/s    --  -14%  -69%
uk    13261/s   17%    --  -64%
smack 36815/s  225%  178%    --
Benchmark: running new1, smack, uk, each for at least 5 CPU seconds...
      new1:  5 wallclock secs ( 5.22 usr +  0.05 sys =  5.27 CPU) @ 14
+0.04/s (n=738)
     smack:  6 wallclock secs ( 5.26 usr +  0.07 sys =  5.33 CPU) @ 87
+3.55/s (n=4656)
        uk:  5 wallclock secs ( 5.18 usr +  0.04 sys =  5.22 CPU) @ 17
+6.44/s (n=921)
       Rate  new1    uk smack
new1  140/s    --  -21%  -84%
uk    176/s   26%    --  -80%
smack 874/s  524%  395%    --

/-\

[antirice]

Small problem. You may want to clear out the leading \b's after the while statement. Like many others on this thread, you've been bitten by the terminal's evaluation of \b.

#!/usr/bin/perl -l

$s = "\bt\b\bhello";

sub smack {
  $a = $s;
  $a =~ s/^\cH+//;
  1 while ($a =~ s/[^\cH]\cH//g);
  $a;
}

sub othersmack {
  $a = $s;
  1 while ($a =~ s/[^\cH]\cH//g);
  $a =~ s/^\cH+//;
  $a;
}

print smack();
print othermack();

print "smack correct" unless grep {ord==ord"\b"} split//,smack();
print "othersmack correct" unless grep {ord==ord"\b"} split//,othersma
+ck();
__END__
output:
hello
hello
othersmack correct

However, getting rid of the alternation is a good bit quicker. =)

[shenme]

Hell yes!   It's over 4 times faster than even my attempt, with your repair making it correct.   Gratuitous benchmark below:

               Rate     smack    new1      uk  badkcams othersmack
smack         188/s        --    -96%    -97%      -97%       -99%
new1         5088/s     2607%      --     -8%      -23%       -86%
uk           5517/s     2835%      8%      --      -16%       -85%
badkcams     6594/s     3409%     30%     20%        --       -82%
othersmack  37394/s    19797%    635%    578%      467%         --
            
               Rate     smack    new1      uk  badkcams othersmack
smack        42.6/s        --    -23%    -29%      -43%       -93%
new1         54.9/s       29%      --     -8%      -27%       -91%
uk           59.6/s       40%      9%      --      -21%       -90%
badkcams     75.0/s       76%     37%     26%        --       -88%
othersmack    626/s     1372%   1040%    951%      735%         --

(See shenme's scratchpad for oogly program)