How do I round up a float whose mantissa is ending in 5? - perl

Question

I want to round up a floating-point number. For example, sprintf '%.4f', 0.12345 returns 0.1235 and sprintf '%.4f', 0.12325 returns 0.1232. For the second example, I want it to print out 0.1233, not 0.1232.
sprintf in Perl is not good enough. Math::BigFloat can do it, but it’s a little over-kill.
Does anyone know if there is other effective way to round up, or whether there is any other module in Perl?

Answer 1

The exact way sprintf will round a number is dependent on how the system libraries round numbers. If you need to control exactly how a number is rounded you will need to use a library that implements your desired rounding explicitly, or write a function to round as desired.
For example to round a positive number to an arbitrary precision, where 5 rounds up this would work.
sub round {
my $numer = shift;
my $precision = shift;
return int($numer * 10 ** $precision + 0.5) * 10 ** -$precision;
}
This however doesn't round correctly for negative numbers, -0.12324 incorrectly rounds to -0.1231. A solution where 5 should round up (that is towards positive infinity) would be to use floor instead of int.
use POSIX qw(floor);
sub round {
my $numer = shift;
my $precision = shift;
return floor($numer * 10 ** $precision + 0.5) * 10 ** -$precision;
}
If instead 5 should round to the largest absolute value (that is round away from 0) then you can add a simple check for negative numbers to round them in the correct direction.
sub round {
my $numer = shift;
my $precision = shift;
my $direction = $numer >= 0 ? 0.5 : -0.5;
return int($numer * 10 ** $precision + $direction) * 10 ** -$precision;
}
There are more complex rules for rounding used in some circumstances where the bias from rounding 5 in one direction for all numbers is unacceptable and any (decimal) rounding of floating point numbers is subject to possible errors due to imprecision in their binary format.

Answer 2

Perl's sprintf can only be as good as the data it's fed.
There is no floating point number equal exact to either 0.12345 or 0.12325:
The floating point number closest to the 0.12345 is exactly 2223877495995551/2**54 ≈ 0.123450000000000004, a value slightly greater than the input.
The floating point number closest to the 0.12325 is exactly 4440549232587309/2**55 ≈ 0.123249999999999998, a value slightly less than the input.
Note that the 5th decimal digit of the latter one is 4 not 5.
When rounded to 4 decimal places under the usual rules, they must come out as 0.1235 and 0.1232 respectively.
You say that using Math::BigFloat is a little overkill, but it seems that what you really want is decimal arithmetic so the overkill is inescapable.

Answer 3

This did work for me:
print sprintf '%.*f', 4, 0.12325
Your code does work for me too. Did you test that code on its own?
Could you tell us your perl version too (perl -v)?