type casting float to int/short

[acerunner316]

I ran into a bug in my app and eventually figured that it was a problem in the type casting. I wrote a short piece of code to confirm it. float tempflt = 3.8; int tempint = (int)(tempflt*100); CString Debug; Debug.Format("%d, %f", tempint, tempflt); MessageBox(Debug); Why is it that tempint becomes 379 and not 380?

[Christian Graus]

A floating point value is never precise.  Sorry, that's how it goes.  Try including math.h and do this: int tempint = (int )ceil(tempflt*100); I think that's the right syntax, but the Math class has floor and ceiling methods, which round up and down.  This will round your 379.xxx up to 380.00, then you cast it to int.

Christian Graus - C++ MVP 'Why don't we jump on a fad that hasn't already been widely discredited ?' - Dilbert

[acerunner316]

Christian Graus wrote:

A floating point value is never precise.

How predictable is a floats actual value? Because I don't think ceil will be a general solution. What about cases when I should be rounding down? Is double more precise? I really don't need too much precision as far as decimal points goes. The range I will be in will be -32.768 to 32.767 (the range of short divided by 1000), so I only need 3 numbers after the decimal, which is why i chose float instead of double. -- modified at 19:35 Tuesday 16th January, 2007

[Rick York]

This is because the float (or double) to int cast performs a truncated conversion. The actual answer was probably more like 379.9999999999999999 but the truncation gives a result of 379. If you want a rounded answer then do this : int tempint = (int)( tempflt * 100 + 0.5 );

[Christian Graus]

acerunner316 wrote:

What about cases when I should be rounding down?

floor

acerunner316 wrote:

Is double more precise?

Yes, but not exact.

Christian Graus - C++ MVP 'Why don't we jump on a fad that hasn't already been widely discredited ?' - Dilbert

[acerunner316]

Christian Graus wrote:

acerunner316 wrote: What about cases when I should be rounding down? floor

That's my point. I need a generic code that will accept a float and multiply by 100 and get the right integer. Without knowing how float will skew the result of the multiply operation, I won't know when ot use ceil, and when to use floor.

[acerunner316]

That's a good idea. :) As long as the float math operations don't skew the answers by more than 0.5, which it doesn't seem like it will do.

[Christian Graus]

They won't, that always guarentees you'll get what you want. ( I forgot that little trick )

Christian Graus - C++ MVP 'Why don't we jump on a fad that hasn't already been widely discredited ?' - Dilbert

[Christian Graus]

you always would need ceil.  Just like adding .5 assumes that you need to round up.

Christian Graus - C++ MVP 'Why don't we jump on a fad that hasn't already been widely discredited ?' - Dilbert

[malaugh]

The best method of making sure you get the correct value is to add 0.5 before casting.