Floating point types in Rust - Deep Dive

Overview - Floating point types

What is it?

Floating point types are used in programming to represent numbers with decimal points, like 3.14 or -0.001. They allow computers to handle fractions and very large or very small numbers approximately. In Rust, the main floating point types are f32 and f64, which store numbers with different levels of precision.

Why it matters

Without floating point types, computers could only work with whole numbers, which would make many tasks like scientific calculations, graphics, and measurements impossible or very inaccurate. Floating point types let us model real-world values that are not whole numbers, enabling precise and flexible computing.

Where it fits

Before learning floating point types, you should understand basic number types like integers and variables. After mastering floating point types, you can learn about numeric operations, rounding errors, and how to handle precision in calculations.

Mental Model

Core Idea

Floating point types store numbers as a combination of a base number and an exponent to represent a wide range of decimal values approximately.

Think of it like...

Imagine a floating point number like a scientific calculator display showing a number in scientific notation, such as 1.23 × 10^4, where the decimal part and the exponent together represent the full number.

┌───────────────┐
│ Floating Point │
│   Number      │
├───────────────┤
│ Sign (±)      │
│ Mantissa (1.23)│
│ Exponent (10^4)│
└───────────────┘

Build-Up - 7 Steps

1

Foundation

What are floating point numbers

Concept: Introduce the idea of numbers with decimals and why they need special types.

In Rust, floating point numbers represent decimal numbers. Unlike integers, they can store fractions like 0.5 or 3.1415. Rust has two main types: f32 (32-bit) and f64 (64-bit), where f64 is more precise.

Result

You can store and use decimal numbers in your Rust programs using f32 or f64.

Understanding that floating point types let you work with decimals is the first step to handling real-world numeric data.

2

Foundation

Difference between f32 and f64

Concept: Explain the two floating point types in Rust and their precision differences.

f32 uses 32 bits to store a number, giving about 6-7 decimal digits of precision. f64 uses 64 bits, giving about 15-16 digits. More bits mean more precise numbers but use more memory.

Result

Choosing f32 or f64 affects how accurate your decimal numbers are and how much memory they use.

Knowing the precision difference helps you pick the right type for your needs, balancing accuracy and performance.

3

Intermediate

How floating point numbers are stored

Concept: Introduce the internal structure: sign, mantissa, and exponent.

Floating point numbers store a sign (positive or negative), a mantissa (the main digits), and an exponent (which shifts the decimal point). This lets them represent very large or very small numbers approximately.

Result

You understand why floating point numbers can represent a wide range but only approximately.

Knowing the parts inside floating point numbers explains why some decimal numbers can't be stored exactly.

4

Intermediate

Precision and rounding errors

🤔Before reading on: do you think floating point numbers always store decimals exactly? Commit to yes or no.

Concept: Explain why floating point numbers sometimes give surprising results due to limited precision.

Because floating point numbers have limited bits, some decimals like 0.1 can't be stored exactly. This causes small rounding errors in calculations, which can add up or cause unexpected results.

Result

You see why 0.1 + 0.2 might not exactly equal 0.3 in Rust programs.

Understanding rounding errors helps you avoid bugs and choose the right methods for comparing floating point numbers.

5

Intermediate

Using floating point literals and operations

Concept: Show how to write floating point numbers and do math with them in Rust.

You can write floating point numbers like 3.14 or 2.0 in Rust. Operations like +, -, *, / work as expected but remember about precision limits. You can specify type with suffixes like 3.14f32 or 2.0f64.

Result

You can perform arithmetic with floating point numbers in Rust safely and clearly.

Knowing how to write and operate on floats is essential for practical programming with decimals.

6

Advanced

Handling floating point comparisons safely

🤔Before reading on: do you think comparing floats with == is always reliable? Commit to yes or no.

Concept: Teach why direct equality checks can fail and how to compare floats properly.

Because of rounding errors, comparing floats with == can be unreliable. Instead, check if the difference between numbers is smaller than a tiny threshold (epsilon). Rust crates like 'float-cmp' help with this.

Result

You avoid bugs caused by direct float comparisons and write more robust code.

Knowing safe comparison methods prevents subtle bugs in programs that use floating point numbers.

7

Expert

Floating point internals and IEEE 754 standard

🤔Before reading on: do you think Rust invented its own floating point format? Commit to yes or no.

Concept: Explain that Rust uses the IEEE 754 standard for floating point representation and what that means.

Rust's f32 and f64 follow the IEEE 754 standard, which defines how bits represent sign, exponent, and mantissa. This standard ensures consistency across platforms but also defines special values like NaN (not a number) and infinity.

Result

You understand the exact format and special cases of floating point numbers in Rust.

Knowing the IEEE 754 basis explains many floating point behaviors and helps debug tricky numeric issues.

Under the Hood

Floating point numbers in Rust are stored as binary values following the IEEE 754 standard. Each number has a sign bit, an exponent field that shifts the decimal point, and a mantissa (fraction) field that holds the significant digits. The exponent uses a bias to allow positive and negative powers of two. This structure lets floats represent a wide range of values but only approximately, causing rounding errors for some decimals.

Why designed this way?

The IEEE 754 standard was created to unify floating point representation across hardware and software, ensuring consistent behavior. It balances range, precision, and performance. Rust uses this standard to leverage hardware support and interoperability. Alternatives like fixed-point or arbitrary precision exist but have different tradeoffs in speed and complexity.

┌───────────────┐
│ Floating Point │
│   Number      │
├───────────────┤
│ Sign (1 bit)  │
├───────────────┤
│ Exponent      │
│ (8 bits f32)  │
│ (11 bits f64) │
├───────────────┤
│ Mantissa     │
│ (23 bits f32) │
│ (52 bits f64) │
└───────────────┘

Myth Busters - 4 Common Misconceptions

Quick: Do you think 0.1 + 0.2 == 0.3 is always true in Rust? Commit to yes or no.

Common Belief: Floating point numbers store decimal values exactly, so 0.1 + 0.2 equals 0.3 exactly.

Reality: Due to binary representation, 0.1 and 0.2 cannot be stored exactly, so their sum is close but not exactly 0.3.

Why it matters: Assuming exact equality causes bugs in comparisons and calculations, leading to incorrect program behavior.

Quick: Do you think f32 is always better because it uses less memory? Commit to yes or no.

Common Belief: Using f32 is always better because it uses less memory and is faster.

Reality: f32 is less precise and can cause more rounding errors; sometimes f64 is necessary for accuracy despite using more memory.

Why it matters: Choosing f32 blindly can cause subtle bugs in calculations that require precision.

Quick: Do you think comparing floats with == is reliable? Commit to yes or no.

Common Belief: You can safely compare floating point numbers using == just like integers.

Reality: Direct equality checks often fail due to tiny rounding differences; approximate comparisons are needed.

Why it matters: Incorrect comparisons cause logic errors, especially in conditions and loops.

Quick: Do you think NaN equals itself? Commit to yes or no.

Common Belief: NaN (Not a Number) is equal to itself like any other number.

Reality: NaN is never equal to anything, including itself, by IEEE 754 rules.

Why it matters: Misunderstanding NaN behavior can cause unexpected bugs in numeric computations and comparisons.

Expert Zone

1

Floating point arithmetic is not associative or distributive, meaning (a + b) + c may differ from a + (b + c), which affects algorithm design.

2

Rust provides methods like is_nan(), is_infinite(), and total_cmp() to handle special float cases and ordering correctly.

3

Compiler optimizations and hardware instructions can affect floating point behavior subtly, so understanding IEEE 754 helps debug performance and correctness issues.

When NOT to use

Floating point types are not suitable when exact decimal representation is required, such as in financial calculations. In those cases, use fixed-point types, arbitrary precision libraries like 'rust_decimal', or integer-based representations.

Production Patterns

In production Rust code, f64 is often preferred for scientific and engineering calculations due to its precision. Developers use epsilon-based comparisons for equality and handle special values explicitly. Libraries and crates provide utilities for safe float math, and tests include tolerance checks to avoid flaky failures.

Connections

Scientific Notation

Floating point numbers implement a binary form of scientific notation.

Understanding scientific notation in math helps grasp how floating point numbers represent very large or small values compactly.

Error Propagation in Measurements

Floating point rounding errors are a form of error propagation in numerical calculations.

Knowing how measurement errors accumulate in physics or engineering clarifies why floating point precision matters in computing.

Human Memory Approximation

Floating point approximation is similar to how human memory stores imperfect snapshots of experiences.

Recognizing that both computers and humans approximate information helps appreciate the limits and design of numeric types.

Common Pitfalls

#1 Comparing floats directly for equality.

Wrong approach:

let a = 0.1f64 + 0.2f64;
if a == 0.3 {
println!("Equal");
} else {
println!("Not equal");
}

Correct approach:

let a = 0.1f64 + 0.2f64;
let epsilon = 1e-10;
if (a - 0.3).abs() < epsilon {
println!("Equal within tolerance");
} else {
println!("Not equal");
}

Root cause:

Floating point numbers cannot represent some decimals exactly, so direct equality often fails.

#2 Using f32 when high precision is needed.

Wrong approach:

let pi: f32 = 3.141592653589793238;
println!("Pi: {}", pi);

Correct approach:

let pi: f64 = 3.141592653589793238;
println!("Pi: {}", pi);

Root cause:

f32 has limited precision and truncates many decimal digits, causing loss of accuracy.

#3 Ignoring special float values like NaN and infinity.

Wrong approach:

let x = 0.0f64 / 0.0f64;
if x == x {
println!("x equals itself");
} else {
println!("x does not equal itself");
}

Correct approach:

let x = 0.0f64 / 0.0f64;
if x.is_nan() {
println!("x is NaN");
} else {
println!("x is a number");
}

Root cause:

NaN is never equal to itself, so direct comparisons can mislead.

Key Takeaways

Floating point types let you store decimal numbers approximately using a sign, mantissa, and exponent.

Rust provides two main floating point types: f32 for less precision and f64 for more precision and range.

Because of limited bits, some decimal numbers cannot be represented exactly, causing rounding errors.

Directly comparing floating point numbers for equality is unreliable; use approximate comparisons instead.

Rust follows the IEEE 754 standard, which defines special values like NaN and infinity and ensures consistent behavior.