On Sat, Sep 10, 2011 at 06:50:52PM +0200, goran kent wrote:
> Hmm, my sort floats are typically like this:
>
> min: 0.1254619125
> max: 3117.88289166118
> typical: 0.231372871201865 (15 dec places max)
>
> sprintf("%0.15f") in Perl produces:
>
> 0.125461912500000 ok
> 3117.882891661180111 screw up (should be 3117.882891661180000)
> 0.231372871201865 ok
>
> I imagine the large value 3117 with the added "0111" won't present a
> problem when sorting -- but I have a feeling I'm wrong, right?
;)
Sure, add the value "8.8" to the mix and you're in trouble, because it will
sort last:
$ perl -MData::Dumper -le 'my @stuff = ("0.125461912500000",
"3117.882891661180111", "0.231372871201865", "8.8"); warn Dumper([sort
@stuff])'
$VAR1 = [
'0.125461912500000',
'0.231372871201865',
'3117.882891661180111',
'8.8'
];
You need to change the format argument from "%0.15f" to something like
"%023.15f" to get a minimum number of total characters (the decimal point is
included in the count) and to fill with leading zeros.
$ perl -MData::Dumper -le 'my @stuff = map {sprintf "%023.15f", $_}
("0.125461912500000", "3117.882891661180111", "0.231372871201865", "8.8"); warn
Dumper([sort @stuff])'
$VAR1 = [
'0000000.125461912500000',
'0000000.231372871201865',
'0000008.800000000000001',
'0003117.882891661180111'
];
Watch out, though -- you have to know the range in advance in order to choose
the format so that you don't overflow. Effectively what we're doing is
converting a floating point representation to "fixed point", and fixed point
cannot represent the same range as floating point without occupying a lot more
space.
Marvin Humphrey
