double to int issue. losing a penny

[Voski]

For class I have to write a program that will take in the conversion rate from euros to dollars and the amount of euros the person has. Then it is supposed to convert to US dollars and then distribute the money into dollars, quarters, dimes, nickels, and pennies.

So far I can get everything to work but I lose a penny.

For example I enter in the conversion rate of .7266 (rate as of last night) and put in that I have 5000 euros.

I get 6881.37 US dollars.

with 6881 dollars, 1 quarter, 1 dime, 0 nickes, 1 penny. This adds up to 6811.36.

I am not really sure why this is happening.

Here is the code

==============

#include <iostream>
using namespace std;

int main() {

cout << "Hello, travelers! \n\n" << "Enter the conversion rate from euros to dollars (i.e. how many euros in a US dollar):";

double numEurosPerDollar = 0;
cin >> numEurosPerDollar;

cout << "\n\nThe conversion rate you entered means that one US dollar is equivalent to " << numEurosPerDollar << " euros today. \n\n";

cout << "Now enter the total amount you'd like to convert from euros to USD (up to hundreths of a euro):";

double numEuros = 0;
cin >> numEuros;

double numDollars = numEuros / numEurosPerDollar;

cout << "\n\n" << numEuros << " euros will convert to " << numDollars << " US dollars.\n\n";

cout << "If you were to make this from change, using the biggest denomination at every step, it would be:\n\n";

const int PENNIES_PER_DOLLAR = 100;
const int PENNIES_PER_NICKEL = 5;
const int PENNIES_PER_DIME = 10;
const int PENNIES_PER_QUARTER = 25;

int numPennies = numDollars * PENNIES_PER_DOLLAR;
int numPenniesLeftOver = numPennies % PENNIES_PER_DOLLAR;
//this gives us whole dollars and left over pennies.

int numQuarters = numPenniesLeftOver / PENNIES_PER_QUARTER;
numPenniesLeftOver = numPenniesLeftOver % PENNIES_PER_QUARTER;
//this gives us whole quarters and left over pennies.

int numDimes = numPenniesLeftOver / PENNIES_PER_DIME ;
numPenniesLeftOver = numPenniesLeftOver % PENNIES_PER_DIME;
//this gives us whole dimes and left over pennies.

int numNickel = numPenniesLeftOver / PENNIES_PER_NICKEL;
numPenniesLeftOver = numPenniesLeftOver % PENNIES_PER_NICKEL;
//this gives uf whole nickels and left over pennies.

cout << (int)numDollars << " dollars, " << numQuarters << " quarters, " << numDimes << " dimes, " << numNickel << " nickels, and " << numPenniesLeftOver << " pennies.";

cout << "\n\n";
return 0;

}
============

[Eri523]

So far I can get everything to work but I lose a penny.

For example I enter in the conversion rate of .7266 (rate as of last night) and put in that I have 5000 euros.

I get 6881.37 US dollars.

with 6881 dollars, 1 quarter, 1 dime, 0 nickes, 1 penny. This adds up to 6811.36.

The output you get from the double variable is rounded to the nearest value with two decimals. Actually it's 6881.3652628681530415634461877236 $ (according to the Windows Calculator). But the conversion from double to int truncates the value downwards to the next integral value. That's where your cent gets lost.

BTW, please use code tags when posting code next time. This will make it much more readable.

Ah, and... Welcome to CodeGuru!

[kempofighter]

Use code tags and properly indent within the tags.
http://www.codeguru.com/forum/misc.php?do=bbcode#code

The calculation of change seems correct to me but the output stream is a rounded number to two decimal points. Sorry I don't know what the default precision is supposed to be or how you can stop it from rounding. Clearly there is no such thing as 0.5 pennies so in that kind of conversion anything less then whole pennies would have to be truncated. This is why your change calculation is correct but the output of dollars.cents is showing .37. You may have to manually truncate the variable prior to output because I cannot seem to find an io manipulator that will stop the rounding from occurring.

[kempofighter]

Take a look at this.
http://www.parashift.com/c++-faq-lit...html#faq-29.16

It seems annoying that the iomanip doesn't include anything to truncate but I think that it is impossible to implement due to problems across platforms with floating point representations. If I manually truncate in the following way I still get a strange response, consistent with that explanation. This code will give you what you want (xxxx.36) but if you step with the debugger you'll see that numDollars is now xxxx.35999999997 or something to that effect. It simply isn't possible to truncate the way you think you should be able to because in the computer it is not possible to represent 0.36 perfectly.

double numDollars = numEuros / numEurosPerDollar;
numDollars = floor(numDollars*100.0) / 100.0;
cout << "\n\n" << numEuros << " euros will convert to " << numDollars << " US dollars.\n\n";

// prints 0.36 as I expect but in the debugger value is 0.35999999999999999
double value(.36);
std::cout << "value : " << value << std::endl;

[rdrast]

Honestly, you shouldn't use ANY approximate data type (floating point) when dealing with money. Double's are still floating point numbers.

For any financial calculation, you should use a fixed point (even implied) numeric format.

Rather then summarize information, see here: http://docs.sun.com/source/806-3568/ncg_goldberg.html

[Lindley]

It seems annoying that the iomanip doesn't include anything to truncate but I think that it is impossible to implement due to problems across platforms with floating point representations. If I manually truncate in the following way I still get a strange response, consistent with that explanation. This code will give you what you want (xxxx.36) but if you step with the debugger you'll see that numDollars is now xxxx.35999999997 or something to that effect. It simply isn't possible to truncate the way you think you should be able to because in the computer it is not possible to represent 0.36 perfectly.

While the admonition against using floating point types for money stands, you can indeed specify a set number of fractional digits using iomanip. Note that the behavior of setprecision changes in the fixed and scientific modes:

double val = 12345.359999999;
cout << std::fixed << std::setprecision(2) << val;

[kempofighter]

While the admonition against using floating point types for money stands, you can indeed specify a set number of fractional digits using iomanip. Note that the behavior of setprecision changes in the fixed and scientific modes:

double val = 12345.359999999;
cout << std::fixed << std::setprecision(2) << val;

I'm well aware of that but those methods round. They do not help in any way whatsoever unless you manually perform the truncation yourself. In the OPs example he ends up with something like 12345.365. Using your method he would still end up with .37 which is not correct.

I do have some experience with fixed point math and it is not entirely precise either. You tend to lose some precision depending on how many bits you are using to represent each side of the decimal point. I'm not convinced that it is any better for these operations then floating point.

[Lindley]

I do have some experience with fixed point math and it is not entirely precise either. You tend to lose some precision depending on how many bits you are using to represent each side of the decimal point. I'm not convinced that it is any better for these operations then floating point.

Yes, but the errors are a lot more predictable with fixed-point, and generally will not accumulate unless you multiply excessively (to the point where you run out of bits) or you divide. Since neither of those is a common operation in finance, it generally isn't a problem. (Multiplying by an interest rate is fine, since it's unit-less the number of fractional bits doesn't change. You may compute a fractional cent which you can't store, but there are standard ways of dealing with that.)

[monarch_dodra]

I do have some experience with fixed point math and it is not entirely precise either. You tend to lose some precision depending on how many bits you are using to represent each side of the decimal point. I'm not convinced that it is any better for these operations then floating point.

No you don't. fixed point arithmetic is always 100% exact, 100% of the time. I grant you have a reduced range of values, and you can overflow, but that is a very different problem (and quasi nonexistent with the correct integer width).

[superbonzo]

fixed point arithmetic is always 100% exact, 100% of the time. I grant you have a reduced range of values, and you can overflow...

in principle, the same claim applies for floating point numbers, in the sense that you could in principle predict the exact amount of error induced by an operation with floating numbers ( although, as Lindley said, it's a much more difficult task ). The problem is that the IEEE standard for floating point numbers specifically admits "non strict" evaluation of floating point expressions, so, in practice, the actual error is essentially unpredictable ...

[monarch_dodra]

in principle, the same claim applies for floating point numbers, in the sense that you could in principle predict the exact amount of error induced by an operation with floating numbers ( although, as Lindley said, it's a much more difficult task ). The problem is that the IEEE standard for floating point numbers specifically admits "non strict" evaluation of floating point expressions, so, in practice, the actual error is essentially unpredictable ...

Can you? It was my understanding that the exact error was not only very difficult to work with, but literally non-deterministic.

here, http://www.parashift.com/c++-faq-lit...html#faq-29.18

That said, I'm not sure what you mean by "non strict".

[superbonzo]

It was my understanding that the exact error was not only very difficult to work with, but literally non-deterministic.

yes, I said "in principle" meaning "in theory" ... hence, I concluded saying that the error is "essentially unpredictable".

That said, I'm not sure what you mean by "non strict".

basically, the behavior described in the link in your post; moreover, compilers are free to "arithmetically" interpret floating point expressions as they like ( for example, x/y could be x*1./y and so on ... )