Floating point issue

vitoco has asked for the wisdom of the Perl Monks concerning the following question:

If I try this oneliner:

perl -e "$a=-50.123456;for($i=0;$i<20;$i++){print qq!$i $a\n!;$a+=0.01}"

I get this:

0 -50.123456
1 -50.113456
2 -50.103456
3 -50.093456
4 -50.083456
5 -50.073456
6 -50.063456
7 -50.053456
8 -50.043456
9 -50.033456
10 -50.023456
11 -50.013456
12 -50.003456
13 -49.993456
14 -49.983456
15 -49.973456
16 -49.963456
17 -49.953456
18 -49.943456
19 -49.933456
[download]

But if I try this one:

perl -e "$a=-71.123456;for($i=0;$i<20;$i++){print qq!$i $a\n!;$a+=0.01}"

I get this:

0 -71.123456
1 -71.113456
2 -71.103456
3 -71.093456
4 -71.083456
5 -71.073456
6 -71.063456
7 -71.053456
8 -71.043456
9 -71.033456
10 -71.023456
11 -71.0134559999999
12 -71.0034559999999
13 -70.9934559999999
14 -70.9834559999999
15 -70.9734559999999
16 -70.9634559999999
17 -70.9534559999999
18 -70.9434559999999
19 -70.9334559999999
[download]

What's going on here?

Thanks!

Comment on Floating point issue Select or Download Code

Replies are listed 'Best First'.
Re: Floating point issue by pryrt (Abbot) on Aug 12, 2016 at 21:57 UTC
Did you Super Search for floating point first? Recent conversation which probably answers your question	[reply]
Re^2: Floating point issue by vitoco (Hermit) on Aug 12, 2016 at 22:04 UTC
Actually not. Just googled it... Thanks for the hint. But what made me get confused was that it happened sometimes and wanted to know why.	[reply]
Re^3: Floating point issue by pryrt (Abbot) on Aug 12, 2016 at 22:09 UTC
Depending on your starting value, sometimes the little tiny differences round to a value that Perl's default stringification will print differently than your expected precision, and sometimes they won't. If you want consistent precision in output, use sprintf or printf	[reply]
Re: Floating point issue by Marshall (Canon) on Aug 12, 2016 at 22:26 UTC
Binary floating point cannot represent 0.01 from base 10 exactly, close but not exact. Here is an example of rounding the printout using printf(): Update: I got a suggestion to make my warning about the use of $a and $b more prominent. $a and $b are Perl reserved variables and are used by sort amoungst other functions. Users should not use these variables. Use $x or $y or something else instead. Following my own advice, I changed from $a to $x below. Here the OP's use of $a "worked", but it is a bad habit to get into. #!usr/bin/perl use strict; use warnings; print "====float =======\n"; my $x=-50.123456; for my $i (0...19) { printf "%i\t%.3f\n", $i,$x; $x+=0.01; } $x=-71.123456; for my $i (0...19) { printf "%i\t%.3f\n", $i,$x; $x+=0.01; } __END__ ====float ======= 0 -50.123 1 -50.113 2 -50.103 3 -50.093 4 -50.083 5 -50.073 6 -50.063 7 -50.053 8 -50.043 9 -50.033 10 -50.023 11 -50.013 12 -50.003 13 -49.993 14 -49.983 15 -49.973 16 -49.963 17 -49.953 18 -49.943 19 -49.933 0 -71.123 1 -71.113 2 -71.103 3 -71.093 4 -71.083 5 -71.073 6 -71.063 7 -71.053 8 -71.043 9 -71.033 10 -71.023 11 -71.013 12 -71.003 13 -70.993 14 -70.983 15 -70.973 16 -70.963 17 -70.953 18 -70.943 19 -70.933 [download] Update: added similar comment above and changed $a to $x. Here it is ok, but $a and $b are reserved Perl variables and shouldn't be used by users. Use $x or $y instead.	[reply] [d/l]
Re: Floating point issue by Laurent_R (Canon) on Aug 13, 2016 at 07:44 UTC
Hi, consider this simple one-liner where the output should presumably be 0: `$ perl -e 'print .1 + .2 - .3, "\n"' 5.55111512312578e-17` [download] A very small number, indeed, but not zero. BTW, the same in the Perl 6 REPL: `> say .1 + .2 - .3 0` [download]	[reply] [d/l] [select]
Re: Floating point issue by vitoco (Hermit) on Aug 13, 2016 at 21:53 UTC
Thank you, guys. I'll use sprintf as a workaround...	[reply]
Re^2: Floating point issue by Laurent_R (Canon) on Aug 14, 2016 at 08:14 UTC
A more accurate alternative might be to use integer arithmetic: `$ perl -E '$c = -71123456; for my $i (0..19) {say $c/1000000; $c += 10 +000}' -71.123456 -71.113456 -71.103456 -71.093456 -71.083456 -71.073456 -71.063456 -71.053456 -71.043456 -71.033456 -71.023456 -71.013456 -71.003456 -70.993456 -70.983456 -70.973456 -70.963456 -70.953456 -70.943456 -70.933456` [download] You might also consider using the arbitrary big rational number module (Math::BigRat) or possibly the big float module (Math::BigFloat). Math::BigRat is accurate because it basically does integer arithmetic under the hood.	[reply] [d/l]