Is Floating-Point Equality Ever Reliable?

The Perplexing Case of Floating-Point Equality

In the realm of software development, the topic of floating-point comparison has sparked countless debates. While it's widely known that floating-point precision presents challenges for determining equality, there remain scenarios where the use of == might not be as problematic as one might assume.

Let's dissect the specific code example raised:

// Defined in somewhere.h
static const double BAR = 3.14;

// Code elsewhere.cpp
void foo(double d)
{
    if (d == BAR)
        ...
}

The question arises whether there are situations where this comparison would be valid. The answer lies in the fundamental nature of floating-point data representation.

IEEE 754, the prevailing standard for floating-point operations, guarantees that integer representations within a specific range are exact when stored as floats. In other words, whole numbers, including 0.0, can be directly compared using == without fear of precision errors.

Therefore, in the given code, as long as BAR is initialized with an integer value within the float's range, the comparison d == BAR is safe and reliable.

However, when dealing with variables that have undergone mathematical operations or calculations, the situation becomes more complex. While some calculations may yield exact integer values, others may introduce rounding errors that invalidate float equality.

Similarly, the call foo(BAR) will always compare equal, assuming that BAR is indeed an integer constant. This is because both variables use the same static constant, which is guaranteed to represent the same underlying integer value.

In conclusion, while floating-point equality should generally be treated with caution, there are specific scenarios where it can be used with confidence. When comparing integer numbers within the float's range or using static constants, programmers can rely on the precise nature of floating-point representation to perform reliable equality checks.

The above is the detailed content of Is Floating-Point Equality Ever Reliable?. For more information, please follow other related articles on the PHP Chinese website!