stringstream to float problem

[c0d3m0nk3y]

When I try and convert a stringstream to a float it seems that I encounter a problem with float arithmetic when using the extraction operator. The line of code that causes problems is as follows:

 ssConvert >> fValue

from what i have seen it starts acting weird at 32.1 it will turn it into 32.100006 and then any float value after that.

Thanks for any help.

[matsp]

You have to realize that floating point values are NOT PRECISE. It is very likely that you will get differences in the least significant digits of the number.

So, given that, I don't expect stringstream to magically solve this issue, and you have to make sure that you understand the difference.

--
Mats

[laserlight]

from what i have seen it starts acting weird at 32.1 it will turn it into 32.100006 and then any float value after that.

I cannot duplicate that for myself, but if it is just floating point inaccuracy, then there is little you can do about it as long as you want to use floating point.

[cpjust]

You could use a double and then this problem would happen less often than with floats.

[Codeplug]

C FAQ - 14. Floating Point

What Every Computer Scientist Should Know About Floating-Point Arithmetic

gg

[c0d3m0nk3y]

Does anyone then know of a way I could go from either a stringstream or a TCHAR array to a float value. I'm getting the value from a text box then converting into a float so I want to keep it as the value that is entered.

Thanks for the help.

[Elysia]

As some mentioned already, if float produces weird results, try a double. If a double doesn't work, then you just have to accept some inaccuracies. Or you could get a library to give you better precision.

[cpjust]

Does anyone then know of a way I could go from either a stringstream or a TCHAR array to a float value. I'm getting the value from a text box then converting into a float so I want to keep it as the value that is entered.

Thanks for the help.

If you want to keep it as the value entered, you'll need to keep it as a string.

[c0d3m0nk3y]

The double produces the same results as it is a value after a decimal point.

EDIT:With a float the value becomes 32.099998 after coming out of the stringstream.

It seems that after using the extration operator with the double the value becomes 32.100000000000001 when it should be 32.1 I'm not sure if that explains anything.

Thanks for all the help.

[anon]

Try to understand that there's an infinite amount of values - even between any two values. The floats and doubles only use 32 and 64 bits to store numbers, that is limited storage. That means that for most of the possible values the computer needs to use an approximation.

In addition computers work in binary. In binary 32.1 is an infinite fraction the same way the result of 1/3, 5/7 or pi etc cannot be represented exactly in decimals.

Anyway the difference between 32.1 and the value you are seeing is so tiny that you shouldn't be too much worried about it. If I'm not mistaken it would be the task of the output function to present it to the user rounded to some nice value again.

[matsp]

When producing the output from your application, you may want to restrict the precision of the number such that you only show, say, 5 decimal places. That way, 32.1 becomes 32.10000, which is close enough for most things - if that's not sufficiently precise, you will need to use a special library with extra precision.

--
Mats