# round(789.456,-2) => 800
# round(789.456,-1) => 790
# round(789.456)    => 789
# round(789.456,1)  => 789.5
# round(789.456,2)  => 789.46

sub round {
  my ($n, $p) = @_;
  my $sign = ($n > 0) ? 1 : -1;

  $p ||= 0;
  $n *= 10 ** $p;
  $n = int($n + .5 * $sign);
  return $n / 10**$p;
}
[download]

This sub has the same bug as sprintf.

round(5.245,2) returns 5.25

but

round(9.325,2) returns 9.32

But again, the root 'bug' is in floating point, which cannot exactly represent the numbers you give. See What Every Computer Scientist Should Know About Floating-Point Arithmetic.

c:\@Work\Perl>perl -wMstrict -le
"for my $f (5.245, 9.325) {
   print sprintf '%.20f', $f;
   }
"
5.24500000000000010000
9.32499999999999930000
[download]

Update: Slight cosmetic change to code example.

The good news is, add a very small number like .000001 and you'll get what you want.

$a = sprintf("%.2f", $a+.00001);
[download]

The default behaviour for routines like this is to round down to the nearest number, much like int() does for integers. As a result, rarely does any rounding occur in any Perl or C routines on which Perl is based, such as sprintf.

I don't think this is true, tadman. Testing:

% perl -e 'printf "%1.2f\n", 1.234'
1.23
[download]

and

% perl -e 'printf "%1.2f\n", 1.235'
1.24
[download]

Which gives me to think that rounding does in principle occur in sprintf and printf. The problem is the floating point representation of base-10 numbers, which occasionally gives you less precision than you though you had. If you really need that precision, hunt around for further discussion on the issue using Super Search, and you'll find plenty.

Update: or just read japhy's response on the original thread.




---

If God had meant us to fly, he would *never* have give us the railroads.
    --Michael Flanders