Do you know there are two valid zero representations in JavaScript?
posZero = +0; negZero = -0;
In pure mathematics, zero means nothing and its sign doesn’t matter. +0 = -0 = 0. Computers can’t represent value well enough and mostly use the IEEE 754 standard.
Most languages have two zeros!
The IEEE 754 standard for floating point numbers allows for signed zeros, thus it is possible to have both -0 and +0. Correspondingly, 1 / +0 = +∞ while 1 / -0 = -∞ and these are values at opposite ends of the number line.
- They can be viewed as vectors with zero magnitude pointing in opposite directions.
- In the mathematical field of limits, negative and positive zeros show how zero was reached.
These two zeros can lead to issues as shown with the disparate ∞ results.
Why two zeros occur in IEEE 754
There is a bit representing the sign of each numeric value independent of its magnitude. Consequently if the magnitude of a number goes to zero without its sign changing then it becomes a -0.
So why does this matter? Well, JavaScript implements the IEEE 754 standard and this post goes into some of the details.
Keep in mind, the default zero value in JavaScript (and most languages) is actually the signed zero (+0).
The zeros in JavaScript
1. Representation
let a = -0; a; // -0 let b = +0; b; // 0
2. Creation
All mathematical operations give a signed zero result (+0 or -0) that depends on the operand values.
The only exception to this rule involves addition and subtraction involving +0 and -0.
- Adding two -0 values will always be -0
- Subtracting a 0 from -0 will also be -0
Any other combination of zero values gives a +0. Another thing to note is that negative zeros cannot be created as a result of addition or subtraction of non-zero operands. Thus -3 + 3 = 3 – 3 = +0.
The code below shows some more examples.
// Addition and Subtraction 3 - 3 // 0 -3 + 3 // 0 // Addition of zero values -0 + -0; // -0 -0 - 0; // -0 0 - 0; // 0 0 + -0; // 0 // Multiplication 3 * 0 // 0 3 * -0 // -0 // Division 3 / Infinity // 0 -3 / Infinity // -0 // Modulus 6 % 2 // 0 -6 % 2 // -0
3. The issue with zero strings
There is a minor niggle with stringifying -0. Calling toString will always give the result “0”. On the flip side, parseInt and parseFloat parse negative zero values.
Consequently, there is a loss of information in the stringify -> parseInt transformation. For example, if you convert values to strings (for example, via JSON.stringify), POST to some server and then retrieve those strings later.
let a = '-0';
a.toString(); // '0'
parseInt('-0', 10); // -0
parseFloat('-0', 10); // -0
4. Differentiating between +0 and -0
How would you tell one zero value apart from the other? Let’s try comparison.
-0 === 0; // true -0..toString(); // '0' 0..toString(); // '0' -0 < 0; // false 0 < -0; // false
0..toString() is valid JavaScript. Read this to know why.
ES2015’s Object.is method works
Object.is(0, -0); //false
The ES2015’s Math.sign method for checking the sign of a number is not of too much help since it returns 0 and -0 for +0 and -0 respectively.
Since ES5 has no such helper we can use the difference in behaviour of +0 and -0 to write a helper.
function isNegativeZero(value) {
value = +value; // cast to number
if(value) {
return false;
}
let infValue = 1 / value;
return infValue < 0;
}
isNegativeZero(0); // false
isNegativeZero(-0); // true
isNegativeZero('-0'); // true
5. Applications
What is the use of knowing all this?
1. One example could be say for example if you are doing some machine learning and need to differentiate between positive and negative values for branching. If a -0 result gets coerced into a positive zero; then this could lead to a tricky branching bug.
2. Another usage scenario would be for people who write compilers and try to optimize code. Expressions that would result in zero e.g. x * 0 cannot be optimized as the result now depends on the value of x. Optimizing such expressions and replacing them with a 0 will lead to a bug.
3. And know too that there are lots of languages that support IEEE 754. Let’s take C# and Java for example:
// Java System.out.print(1.0 / 0.0); // Infinity System.out.print(1.0 / -0.0); // -Infinity
// C# Console.WriteLine(1.0 / 0.0); // Infinity Console.WriteLine(1.0 / -0.0); // -Infinity;
Try it in your language too!
6. IEEE specifications
The IEEE specifications lead to the following results
Math.round(-0.4); // -0 Math.round(0.4); // 0 Math.sqrt(-0); // -0 Math.sqrt(0); // 0 1 / -Infinity; // -0 1 / Infinity; // 0
Rounding -0.4 leads to -0 because it is viewed as the limit of a value as it approaches 0 from the negative direction.
The square root rule is one I find strange; the specification says: “Except that squareRoot(–0) shall be –0, every valid squareRoot shall have a positive sign.”. If you are wondering, IEEE 754 is the same reason why 0.1 + 0.2 != 0.3 in most languages; but that’s another story.
Thoughts? Do share them in the comments.
