[s]printf %d oddity

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

I'm trying to understand (for my own sanity - and others I work with) why the following snippet does what it does:

my $x = 8.78; 
my $y = $x * 100;
my $z = $x * 1000;

printf "%08d\n", $y;     # prints 00000877 - wrong
printf "%08d\n", "$y";   # prints 00000878 - right 
printf "%3.5f\n", $y;    # prints 878.00000 - right
printf "%08d\n", $z;     # prints 00008780 - right
[download]

I can understand that since $y is a floating-point number it might be stored in memory as 877.99999999999... But, if so, why doesn't the %3.5f example print 877.99999? And why does multiplying by 1000 make a difference for the %08d version?

I've tested this code on version 5.005_2 on aix, 5.8.6 on win32, and 5.8.4 on linux. Also, I know that the simple "fix" is to convert it to a string first (as in the 2nd printf). 8.78 seems to be one of the magic numbers that triggers this phenomenon. Most do not. I'm sure that there are others... I'd just like to better understand what's happening.

TIA

Comment on [s]printf %d oddity Download Code

Replies are listed 'Best First'.
Re: [s]printf %d oddity by ikegami (Patriarch) on May 19, 2005 at 22:06 UTC
why doesn't the %3.5f example print 877.99999? Because %f rounds. (%d truncates.) `>perl -e "printf '%.16f', 8.78" 8.7799999999999994 >perl -e "printf '%.5f', 8.78" 8.78000 >perl -e "printf '%d', 4.7" 4 >perl -e "printf '%.0f', 4.7" 5 >perl -e "printf '%d', -4.7" -4 >perl -e "printf '%.0f', -4.7" -5` [download] And why does multiplying by 1000 make a difference for the %08d version? If I were to guess, the floating point processor is forced to discard low-precision bits, and it rounds the result when it does so.	[reply] [d/l]
Re: [s]printf %d oddity by Joost (Canon) on May 20, 2005 at 08:46 UTC
In addition to the above answers, note that some "simple" decimal floats cannot be represented by binary floats, which are used by most languages, including perl. You should always expect some small errors that you can usually work around by using sprintf("%xx.xxf"): http://support.microsoft.com/kb/q42980/ http://www-106.ibm.com/developerworks/java/library/j-jtp0114/ edit: and perldoc -q decimal Updated for clarity, this part added: Usually, the standard float -> string conversion as used by print() handles these inaccuracies fairly well, but printf("%d") does a "hard" truncate, as you can see here: `> perl -e 'print (8.78 * 100 ) 878 > perl -e 'printf("%d", (8.78 * 100 ))' 877` [download] This is documented in perlnumber, but the implications can be hard to predict (and hard to debug, since what is printed is not necissarily the same as what is in the number). "What should it profit a man, if he should win a flame war, yet lose his cool?"	[reply] [d/l]
Re: [s]printf %d oddity by chas (Priest) on May 19, 2005 at 22:15 UTC
Curious indeed! 8.78 does seem to behave peculiarly in this respect. For example: `if(8.78100==878){print "OK for 8.78\n"} #prints nothing if(7.78100==778){print "OK for 7.78\n"} #prints "OK for 7.78"` [download] etc chas	[reply] [d/l]
Re^2: [s]printf %d oddity by astroboy (Chaplain) on May 20, 2005 at 09:36 UTC
I'd love to see an explanation for this. It's got me doubting Perl as a language for doing even simple calculations. I can't see how this can be put down to rounding errors between binary and decimal	[reply]
Re^3: [s]printf %d oddity by Joost (Canon) on May 20, 2005 at 09:45 UTC
This is just what happens when you convert between binary and decimal - see the links I provided in the post below. update: Think of it like this: decimal floats are inaccurate too, since they can't represent a rational number like 1/3 in finite notation: `1/3 = 0.333333333333333333....` [download] If all arithmatic was done using decimal notation, assuming a maximum of 10 places you get: `$x = 1/3; # 0.3333333333 $y = $x * 3; $y == 1 -> false # $y == 0.9999999999` [download] The same thing happens here, only since humans have a hard time thinking in floating binary, it's much harder to predict where. "What should it profit a man, if he should win a flame war, yet lose his cool?"	[reply] [d/l] [select]
Re^4: [s]printf %d oddity by astroboy (Chaplain) on May 20, 2005 at 10:13 UTC
Re^5: [s]printf %d oddity by Joost (Canon) on May 20, 2005 at 11:45 UTC
Re^5: [s]printf %d oddity by Anonymous Monk on May 20, 2005 at 11:24 UTC