Hegel: Property-Based Testing

Hegel is a universal family of property-based testing, supporting a variety of languages all powered by Hypothesis. Tests integrate with standard test runners (e.g. cargo test, pytest, jest, etc.). Hegel generates random inputs for your code and automatically shrinks failing cases to minimal counterexamples.

Workflow

Follow these steps when writing property-based tests.

1. Detect Language and Load References

Identify the project language from build files:

Load the corresponding reference file for API details and idiomatic patterns.

2. Explore the Code Under Test

Before writing any test, understand what you're testing:

• Read the source code of the function/module under test
• Read existing tests to understand expected behavior and edge cases
• Read docstrings, comments, and type signatures for documented contracts
• Read usage sites to see how callers use the code and what they expect

The goal is to find evidence for properties, not to invent them.

3. Identify Valuable Properties

Look for properties that are:

• Grounded in evidence from the code, docs, or usage patterns
• Non-trivial — they test real behavior, not tautologies, and do not duplicate the code being tested
• Falsifiable — a buggy implementation could actually violate them

Write one test per property. Don't cram multiple properties into one test.

4. Check for Existing Tests to Evolve or Port

Before writing tests from scratch, always check existing tests:

• Existing PBTs in another framework (proptest, quickcheck, etc.) should be ported to hegel. Load references/porting.md for general guidance and the language-specific porting reference (e.g., references/porting-rust.md). Don't carry over narrow generator bounds from the old framework — use broader generators unless bounds are justified by the function's contract.
• Unit tests and example-based tests can often be evolved into PBTs. Load references/evolving-tests.md for guidance. Tests with hardcoded seeds, parameterized examples, or multiple similar test cases are prime candidates.
• Tests that use rand with fixed seeds are especially good candidates — the randomness should come from hegel instead so failures produce shrinkable counterexamples.

When you evolve an existing test, modify the existing test file rather than creating a new one. Add hegel tests alongside (or replacing) the existing tests in the same file where the original tests live. Do not create a separate test_hegel.rs or similar — property-based tests are tests like any other and belong with the code they're testing.

5. Write the Tests

For each property:

1. Add tests to the appropriate existing test file. If there's already a test_foo.rs covering the module, add hegel tests there. Only create a new file if no relevant test file exists.
2. Choose the simplest possible generators — start with no bounds, unless bounds are logically necessary (e.g. if a number has to be non-zero it's fine to force it to be, but lists should not have max_size set unless there is a compelling correctness reason to set them or poor performance has been observed when actually running the test)
3. Draw values using tc.draw()
4. Run the code under test
5. Assert the property

6. Run and Reflect

Run the tests. When a test fails, ask:

• Is this a real bug? If the code violates its own contract, flag the bug to the user and ask what to do, or fix the code if instructed to do so.
• Is the property unsound? If you asserted something the code never promised, fix the test.
• Is the generator too broad? Only if the failing input is genuinely outside the function's domain, add constraints. Investigate before constraining.

Property Categories

Use this taxonomy to identify what to test. Not every category applies to every function — pick the ones supported by evidence.

High-Value Patterns (Field-Tested)

These patterns are ranked by how often they found real bugs when tested across many popular Rust crate libraries. See references/field-tested-patterns.md for detailed examples.

1. Model Tests (Highest Value for Data Structures)

For any data structure, the highest-value first test is a model test — run the same operations on the library under test and a known-good reference (usually a std type), then assert they agree after every operation.

Choose the right oracle:

• Vec for sequential containers (fixed-capacity vecs, small vecs)
• HashMap for hash maps (alternative/concurrent hash maps)
• BTreeMap for ordered maps (tree maps, persistent maps)
• BTreeSet for ordered sets / bitmaps (compressed bitmaps, tree sets)
• HashSet for unordered sets (indexed sets, bit sets)

2. Idempotence Tests (Highest Value for String/Text Processing)

Any normalization, case conversion, or formatting function should be idempotent: f(f(x)) == f(x). Use generators::text() (not ASCII-only generators) because Unicode edge cases like ß → SS and combining characters are where bugs hide.

3. Parse Robustness (Universal — Test Every Parser)

Every from_str, parse, or decode function should be tested with generators::text(). The property is simple: it should never panic. Parsers that delegate to constructors which panic on invalid values (instead of returning errors) are a common source of bugs.

4. Roundtrip Tests (High Value for Serialization)

parse(format(x)) == x for any serialize/deserialize pair. Test with the full input domain — don't restrict to "reasonable" values. Bugs hide at boundaries like zero (e.g. scientific notation missing the coefficient), large integers (precision loss through f64 intermediaries for values > 2^53), and unusual string content (double slashes in paths, control characters).

5. Boundary Value Tests (High Value for Numeric Code)

Integer operations should be tested with MIN, MAX, 0, and unconstrained ranges. Negating i32::MIN overflows, dividing by i64::MIN overflows, and many libraries forget to handle these. Don't add .min_value(-100).max_value(100) — those bounds hide real bugs.

Choosing Properties

Properties must be evidence-based. Find evidence in:

• Names and Type signatures: A function fn merge(a: Vec<T>, b: Vec<T>) -> Vec<T> implies the output length might equal the sum of input lengths.
• Docstrings and comments: "Returns a sorted list" directly gives you an invariant.
• Assertions and debug_asserts in the source: These are properties the author already identified, and do not need to be duplicated in the tests, but may suggest other invariants.
• Usage patterns: If callers always assume a result is non-empty, assert that the result is always non-empty.
• Existing tests: Unit tests often encode specific instances of general properties.

Err on the side of creating more properties rather than fewer, and if they fail investigate whether the failure is legitimate behaviour or not.

Beware of properties that seem universal but aren't. Read the docs carefully before asserting a property. Examples from real testing:

• Grapheme-based string reverse is NOT an involution (reverse(reverse("\n\r")) ≠ "\n\r" because \r\n is one grapheme cluster while \n\r is two).
• A method called difference might mean symmetric difference (A △ B), not set difference (A \ B) — check the docs.
• A function documented as "returns the largest key ≤ k" means ≤, not <.

When a property fails, investigate whether it's a real bug or a genuine edge case in the domain. A weaker property often still holds.

Generator Discipline

A common mistake agents make when writing property-based tests is over-constraining generators. This leads to tests that are weaker than they need to be.

Start With No Bounds

If the function accepts any i32, use:

generators::integers::<i32>()  // no min_value, no max_value

Do NOT preemptively write:

generators::integers::<i32>().min_value(0).max_value(100)  // WRONG unless justified

Edge Cases Are the Point

Don't narrow ranges to "avoid edge cases." Edge cases are exactly what PBT is for. If a function claims to work on all i32 values, test it on all i32 values — including i32::MIN, i32::MAX, 0, -1, and 1.

Don't Add .min_size(1) by Default

Unless the function's contract explicitly requires non-empty input, test with empty collections too. If a function panics on an empty vec, that might be a bug worth knowing about.

When a Test Fails on Extreme Values

Your first reaction should be: is this a real bug?

You should assume that it is unless you have strong evidence that it is not. If in doubt, ask the user.

• If the function's documentation says it handles all integers but it overflows on i32::MAX, that's a bug in the code, not in your test.
• Only add bounds after investigating and confirming the input is outside the function's documented domain.

When to Add Constraints

Add generator bounds only when:

1. The function's contract explicitly excludes some inputs. For example, fn sqrt(x: f64) documents that x >= 0 is required.
2. You need to avoid undefined behavior. For example, division by zero.
3. A test failure has been investigated and confirmed to be outside the function's domain.

Avoid rejection sampling where possible

When a constraint involves relationships between multiple generated values, you may use tc.assume():

let a = tc.draw(generators::integers::<i32>());
let b = tc.draw(generators::integers::<i32>());
tc.assume(a != b);  // constraint relates two values

This example is perfectly fine, but it is better to avoid assume altogether when you can:

e.g.

let a = tc.draw(generators::integers::<i32>());
let b = tc.draw(generators::integers::<i32>().min_value(a));

is better than

let a = tc.draw(generators::integers::<i32>());
let b = tc.draw(generators::integers::<i32>());
tc.assume(a <= b)

Even better is:

let mut a = tc.draw(generators::integers::<i32>());
let mut b = tc.draw(generators::integers::<i32>());
if (a > b) {
    (a, b) = (b, a);
}

It is particularly important to avoid rejection sampling in cases where the rejection rate is likely to be high.

For example st.integers().map(|n| n * 2) is much better than st.integers().filter(|n| n % 2 == 0), as the former constructs an even number directly, while the latter throws away around 50% of test cases.

Getting Large Collections

Hegel's default collection size is small. If you need large collections (e.g., to exercise deep tree paths), draw the size separately and pass it as min_size:

// GOOD — can generate large collections, shrinks well
let n = tc.draw(generators::integers::<usize>().max_value(300));
let keys: Vec<i32> = tc.draw(generators::vecs(generators::integers())
    .min_size(n));  // no max_size — let hegel go bigger if it wants

// BAD — hegel's default size distribution rarely produces 100+ elements
let keys: Vec<i32> = tc.draw(generators::vecs(generators::integers()));

Setting min_size but not max_size is a shrinking optimization: hegel can shrink n to find the minimal collection size that triggers the bug, while still being able to add extra elements if needed.

Use .unique() for Key Generation

When testing maps/sets that need unique keys:

let keys: Vec<i32> = tc.draw(generators::vecs(generators::integers::<i32>())
    .max_size(30).unique());

This avoids confusion about which value wins for duplicate keys.

Handling Randomness in Code Under Test

When the code under test requires an RNG (e.g., fn sample(&self, rng: &mut impl Rng)), do not create a seeded RNG like ChaCha8Rng::seed_from_u64(seed) with a hegel-generated seed. This defeats shrinking — hegel can only shrink the seed integer, not the actual random decisions the RNG makes.

Instead, use hegel's rand feature to get a hegel-controlled RNG. See the language-specific reference for API details.

Rand version mismatch

Hegel's rand feature uses rand 0.9. If the project uses an older version of rand (e.g., 0.8), the RNG traits will be incompatible. In this case, ask the user whether they'd like to upgrade the project's rand dependency to 0.9. This is usually straightforward (the main API changes are gen_range -> random_range, gen::<T>() -> random::<T>(), thread_rng() -> rng(), from_entropy -> from_os_rng). Do not silently fall back to seeded ChaCha — that defeats the purpose.

Two modes: artificial vs true randomness

generators::randoms() has two modes:

• Default (artificial randomness): Every random decision goes through hegel, enabling fine-grained shrinking of individual random values. This is the best option for most code.

• generators::randoms().use_true_random(true): Generates a single seed via hegel, then creates a real StdRng from it. Hegel can only shrink the seed, not individual random decisions. Use this when the code under test does rejection sampling or otherwise depends on the RNG producing statistically random-looking output. Artificial randomness can cause rejection loops to hang because the controlled byte sequences don't look random enough.

How to choose: Start with the default. If tests hang or time out because the code does rejection sampling internally, switch to .use_true_random(true).

Refactoring concrete RNG types

If the code under test takes a concrete RNG type (e.g., rng: &mut ChaCha8Rng) rather than a trait bound, consider whether it should be refactored to accept impl Rng or &mut dyn RngCore instead. This is both better API design and makes the code testable with hegel's random generator. Suggest this refactoring to the user.

Common Mistakes

1. Over-constraining generators — Adding bounds "just in case." This hides bugs and makes tests less valuable. See Generator Discipline above.
2. Testing trivial properties — assert!(x == x) or assert!(vec.len() >= 0) test nothing. Every property should be falsifiable by a buggy implementation.
3. Using the implementation as the oracle — If your test calls the same function to compute the expected result, it can never fail. Use an independent reference implementation (do not just copy the code to write this!), a simpler algorithm, or a structural property.
4. Generating too broadly then filtering almost everything — If .filter() or tc.assume() rejects most inputs, Hegel will give up. Restructure your generators instead (e.g., use .map() or dependent generation).
5. Creating a separate test file for hegel tests — Property-based tests belong alongside the existing tests for the same code. Don't put them in test_hegel.rs or test_properties.rs — add them to the existing test files.
6. Using manually seeded RNGs — Don't generate a seed with hegel then create ChaCha8Rng::seed_from_u64(seed). Use generators::randoms() with the rand feature so hegel controls the random decisions and can shrink them. See "Handling Randomness" above.
7. Overflowing in test code — When computing values from generated data (e.g., map.insert(k, k * 10)), your test code itself can overflow before the library has a chance to be buggy. Use wrapping arithmetic (k.wrapping_mul(10)) or smaller intermediate types (draw i16, cast to i32 for multiplication) to prevent this. Distinguish "this constraint protects the library's contract" (keep it) from "this constraint prevents my test from overflowing" (use wrapping arithmetic instead).
8. Adding .max_size() for performance — If a test is slow with large collections, lower test_cases rather than restricting the input space. A slow test that finds bugs beats a fast test that can't. Many tree/trie bugs only manifest at 50-200+ elements.

Quick Setup

Rust

# Cargo.toml
[dev-dependencies]
hegel = { git = "https://github.com/hegeldev/hegel-rust" }

If the code under test uses rand:

[dev-dependencies]
hegel = { git = "https://github.com/hegeldev/hegel-rust", features = ["rand"] }

Requires uv on PATH. Run with cargo test.