Print

[sjegtp]

sjegtp

[C++] Problem involving division by zero

Suppose I have this statement in a code:

Code: [Select]

//code
//code
x = 1/y;
//code
//code
And y might equal zero. If that happens, x will be equalled to an improper value (it appears something like "x = 1.INF0000" in the debug mode).
In order to avoid this problem, setting x=0 whenever this division by zero occurs, I could simply do:

Code: [Select]

if(y==0)
     x = 0;
else
     x = 1/y;
But this code just prepares a condition to prevent the division from happening.


If I want to solve the problem after the division by zero was done; is there a way to fix the variable after it was set to an improper value? As in this code:

Code: [Select]

x = 1/y;
//code
//code
//code
if(x == <improper value> )
     x=0;


[Infinitus]

Infinitus

Re: [C++] Problem involving division by zero

Google for isnan and isinf macros.

If your using VS you may have to define them yourself (its far from C99 compliant). You can probably use something similar to the following code (nicked from google, haven't been bothered to test) to define them.

Code: [Select]

template<typename T>
inline bool isnan(T value)
{
return value != value;

}

// requires #include <limits>
template<typename T>
inline bool isinf(T value)
{
return std::numeric_limits<T>::has_infinity &&
value == std::numeric_limits<T>::infinity();
}



Its far better practice to prevent the problem occuring in the first place, rather than fixing it afterwards.

[PoliteProgrammer]

PoliteProgrammer

Re: [C++] Problem involving division by zero

What is y, specifically? It is good practice to check that y != 0 before carrying out the operation 1/y. If for some reason you are unable or unwilling to do this (more info on what y actually is would help) then a 'filter' for x would probably be best:

Code: [Select]

  if (x > 1e5) {
    x = 1e5;
  }

Where you could replace 1e5 to be some obscenely high, but finite value. If y is a float, then the chances of it being EXACTLY equal to zero are pretty negligible.

[sjegtp]

sjegtp

Re: [C++] Problem involving division by zero

@PoliteProgrammer:
No, in the situation I'm programming it's only when y is EXACTLY zero (and the chance is very high, even though y is a double), so that filter wouldn't work (because I can't use "if(x==1.INF0000)", since it's not a proper value).

@Infini:
The problem is that, in order to prevent the error, that condition would have to be in an enormous loop, and would make the program slower; so if I could just fix it out of the loop it wouldn't slow down the program.

But I see, I guess it's better to prevent it anyway.

[PoliteProgrammer]

PoliteProgrammer

Re: [C++] Problem involving division by zero

I guess it's better to prevent it anyway.

That's definitely the most sensible way to go. Just hope you never have to get your head around 0/0. I've had to work that out several times over the years  :'(

[MG-Zero]

MG-Zero

Re: [C++] Problem involving division by zero

^^haha..0/0..then you into calculus =D

hmm...try/catch blocks?

i love big weenies and i cannot lie

[sjegtp]

sjegtp

Re: [C++] Problem involving division by zero

Just hope you never have to get your head around 0/0. I've had to work that out several times over the years  :'(

In 0/0 you just use the limit. But in 1/0 there's no limit so you have to improvise

hmm...try/catch blocks?

I'm not very used to try/catch, but as far as I know that's preventing too, isn't it? Is there a way to fix it afterwards using try/catch?



But I just found a quick trick to fix the problem. if you have:

Code: [Select]

float y=0;
float x=1/y;
int z=x;
x=z;
then x equals zero. There is some loss of data, but if you want to keep the decimals you can do

Code: [Select]

int z=x*100;
x=z/100.0;


[MG-Zero]

MG-Zero

Re: [C++] Problem involving division by zero

You could probably fix it within the catch block.  Maybe the code you just posted would work in the catch block.

i love big weenies and i cannot lie

[sjegtp]

sjegtp

Re: [C++] Problem involving division by zero

You could probably fix it within the catch block.  Maybe the code you just posted would work in the catch block.

<.< Yeah, but you still have to put a try block before the division, don't you? So it involves prevention, doesn't it?


But I just found something very interesting, with no loss of data. You can do:

Code: [Select]

double y=0;
double x=1/y;
if(1/x==0)// if you do z=1/x;, z will equal zero. In addition, x is considered greater (>) than zero, but 1/x equals zero.
     x=0;
And for negative square roots:

Code: [Select]

double x=sqrt(-5);
if(x*x<0)//though here x*x doesn't equal -5. It goes through the <0 and ==0 conditions, but not for the >0 one.
     x=0;


[MG-Zero]

MG-Zero

Re: [C++] Problem involving division by zero

I dont see how thats going to work O_o if y = 0 and then you do 1/y...you're still dividing by 0.  Do you mean y/1?

i love big weenies and i cannot lie

[sjegtp]

sjegtp

Re: [C++] Problem involving division by zero

I dont see how thats going to work O_o if y = 0 and then you do 1/y...you're still dividing by 0.  Do you mean y/1?

It's really 1/y. It's because the program calculates x=1/y with no conditions preventing a possible division by zero; so y might be zero. When y==0 the program actually calculates the division, but x will equal 1.INF0000 (something like that) and all the following operations with x involving doubles or floats result in 1.INF0000. But what I've just found out (experimentally) that when you do y=0, x=1/y and z=1/x, then z equals zero in the end.

[Sophist]

Sophist

Re: [C++] Problem involving division by zero

in most cases where you may get division by zero the specific case requires a completely different solution or approach etc.
so, it doesn't really matter how "you want to do it", you should be thinking about what the result should be

My Child Is Student of The Month at Neverland Ranch!

• 

[sjegtp]

sjegtp

Re: [C++] Problem involving division by zero

in most cases where you may get division by zero the specific case requires a completely different solution or approach etc.
so, it doesn't really matter how "you want to do it", you should be thinking about what the result should be

That's another point - in my program, the default quotient for the division by zero is different for each situation, so fixing the problem in the end of the code in stead of preventing it every time in the loop runs the code faster.