Test Framework

Botan uses a custom-built test framework. Some portions of it are quite similar to assertion-based test frameworks such as Catch or Gtest, but it also includes many features which are well suited for testing cryptographic algorithms.

The intent is that the test framework and the test suite evolve symbiotically; as a general rule of thumb if a new function would make the implementation of just a few distinct tests simpler, it is worth adding to the framework on the assumption it will prove useful again. Feel free to propose changes to the test system.

When writing a new test, there are three key classes that are used, namely Test, Test::Result, and Text_Based_Test. A Test (or Text_Based_Test) runs and returns one or more Test::Result.

Namespaces in Test

The test code lives in a distinct namespace (Botan_Tests) and all code in the tests which calls into the library should use the namespace prefix Botan:: rather than a using namespace declaration. This makes it easier to see where the test is actually invoking the library, and makes it easier to reuse test code for applications.

Test Data

The test framework is heavily data driven. As of this writing, there is about 2.5 Mib of test code and 28 MiB of test data. For most (though certainly not all) tests, it is better to add a data file representing the input and outputs, and run the tests over it. Data driven tests make adding or editing tests easier, for example by writing scripts which produce new test data and output it in the expected format.

Test data lives in src/tests/data.

Test

class Test

virtual std::vector<Test::Result> run() = 0

This is the key function of a Test: it executes and returns a list of results. Almost all other functions on Test are static functions which just serve as helper functions for run.

static std::string read_data_file(const std::string &path)

Return the contents of a data file and return it as a string.

static std::vector<uint8_t> read_binary_data_file(const std::string &path)

Return the contents of a data file and return it as a vector of bytes.

static std::string data_file(const std::string &what)

An alternative to read_data_file and read_binary_file, use only as a last result, typically for library APIs which themselves accept a filename rather than a data blob.

static bool run_long_tests() const

Returns true if the user gave option --run-long-tests. Use this to gate particularly time-intensive tests.

static Botan::RandomNumberGenerator &rng()

Returns a reference to a fast, not cryptographically secure random number generator. It is deterministically seeded with the seed logged by the test runner, so it is possible to reproduce results in “random” tests.

Tests are registered using the macro BOTAN_REGISTER_TEST which takes 2 arguments: the name of the test and the name of the test class. For example given a Test instance named MyTest, use:

BOTAN_REGISTER_TEST("mytest", MyTest);

All test names should contain only lowercase letters, numbers, and underscore.

Test::Result

class Test::Result

A Test::Result records one or more tests on a particular topic (say “AES-128/CBC” or “ASN.1 date parsing”). Most of the test functions return true or false if the test was successful or not; this allows performing conditional blocks as a result of earlier tests:

if(result.test_str_eq("first value", produced, expected))
   {
   // further tests that rely on the initial test being correct
   }

Only the most commonly used functions on Test::Result are documented here, see the header tests.h for more.

Test::Result(const std::string &who)

Create a test report on a particular topic. This will be displayed in the test results.

bool test_success()

Report a test that was successful.

bool test_success(std::string_view note)

Report a test that was successful, including some comment.

bool test_failure(std::string_view err)

Report a test failure of some kind. The error string will be logged.

bool test_failure(std::string_view what, std::string_view error)

Report a test failure of some kind, with a description of what failed and what the error was.

void test_failure(std::string_view what, std::span<const uint8_t> context)

Report a test failure due to some particular input, which is provided as context. Normally this is only used if the test was using some randomized input which unexpectedly failed, since if the input is hardcoded or from a file it is easier to just reference the test number.

bool test_str_eq(std::string_view what, std::string_view produced, std::string_view expected)

Compare two strings for equality.

bool test_str_ne(std::string_view what, std::string_view produced, std::string_view expected)

Compare two strings for non-equality.

bool test_bin_eq(std::string_view what, std::span<const uint8_t> produced, std::span<const uint8_t> expected);

Compare two arrays for equality.

bool test_bin_eq(std::string_view what, std::span<const uint8_t> produced, std::string_view expected_hex);

Compare two arrays for equality, with the expected value provided as a hex string.

template<typename T>
bool test_not_null(std::string_view what, T *ptr)

Verify that the pointer is not null.

bool test_u8_eq(std::string_view what, uint8_t produced, uint8_t expected)

Test that produced == expected.

bool test_u16_eq(std::string_view what, uint16_t produced, uint16_t expected)

Test that produced == expected.

bool test_u32_eq(std::string_view what, uint32_t produced, uint32_t expected)

Test that produced == expected.

bool test_u64_eq(std::string_view what, uint64_t produced, uint64_t expected)

Test that produced == expected.

bool test_sz_eq(std::string_view what, size_t produced, size_t expected)

Test that produced == expected.

bool test_sz_lt(std::string_view what, size_t produced, size_t expected)

Test that produced < expected.

bool test_sz_lte(std::string_view what, size_t produced, size_t expected)

Test that produced <= expected.

bool test_sz_gt(std::string_view what, size_t produced, size_t expected)

Test that produced > expected.

bool test_sz_gte(std::string_view what, size_t produced, size_t expected)

Test that produced >= expected.

bool test_throws(std::string_view what, std::function<void()> fn)

Call a function and verify it throws an exception of some kind.

bool test_throws(std::string_view what, std::string_view expected, std::function<void()> fn)

Call a function and verify it throws an exception of some kind and that the exception message exactly equals expected.

There is also a comparison function for arbitrary types, which is defined in the separate header test_arb_eq.h because it drags in some additional headers.

template<typename T>
bool test_arb_eq(Test::Result &result, std::string_view what, const T &produced, const T &expected)

Compare some arbitrary Ts for equality.

It is required that T not be something that is handled by one of the existing comparison functions (eg not a string, integer, or bytestring type) and also that test_arb_eq is able to deduce some way of printing values of T. Depending on your T you may need to extend the implementation of detail::to_string in that header.

Text_Based_Test

A Text_Based_Text runs tests that are produced from a text file with a particular format which looks somewhat like an INI-file:

# Comments begin with # and continue to end of line
[Header]
# Test 1
Key1 = Value1
Key2 = Value2

# Test 2
Key1 = Value1
Key2 = Value2

class VarMap

An object of this type is passed to each invocation of the text-based test. It is used to access the test variables. All access takes a key, which is one of the strings which was passed to the constructor of Text_Based_Text. Accesses are either required (get_req_foo), in which case an exception is throwing if the key is not set, or optional (get_opt_foo) in which case the test provides a default value which is returned if the key was not set for this particular instance of the test.

std::vector<uint8_t> get_req_bin(std::string_view key) const

Return a required binary string. The input is assumed to be hex encoded.

std::vector<uint8_t> get_opt_bin(std::string_view key) const

Return an optional binary string. The input is assumed to be hex encoded. Returns empty if the value was not provided.

Botan::BigInt get_req_bn(std::string_view key) const

Return a required BigInt. The input can be decimal or (with “0x” prefix) hex encoded.

Botan::BigInt get_opt_bn(std::string_view key, const Botan::BigInt &def_value) const

Return an optional BigInt. The input can be decimal or (with “0x” prefix) hex encoded.

std::string get_req_str(std::string_view key) const

Return a required text string.

std::string get_opt_str(std::string_view key, std::string_view def_value) const

Return an optional text string, or the specified default value if not set.

size_t get_req_sz(std::string_view key) const

Return a required integer. The input should be decimal.

size_t get_opt_sz(std::string_view key, const size_t def_value) const

Return an optional integer. The input should be decimal.

class Text_Based_Test : public Test

Text_Based_Test(const std::string &input_file, const std::string &required_keys, const std::string &optional_keys = "")

Note

The final element of required_keys is the “output key”, that is the key which signifies the boundary between one test and the next. When this key is seen, run_one_test will be invoked. In the test input file, this key must always appear least for any particular test. All the other keys may appear in any order.

Test::Result run_one_test(const std::string &header, const VarMap &vars)

Runs a single test and returns the result of it. The header parameter gives the value (if any) set in a [Header] block. This can be useful to distinguish several types of tests within a single file, for example “[Valid]” and “[Invalid]”.

bool clear_between_callbacks() const

By default this function returns false. If it returns true, then when processing the data in the file, variables are not cleared between tests. This can be useful when several tests all use some common parameters.

Test Runner

If you are simply writing a new test there should be no need to modify the runner, however it can be useful to be aware of its abilities.

The runner can run tests concurrently across many cores, and does so by default on most systems. If you want single-threaded testing for some reason, use the option --test-threads=1. If not specified then (as of this writing) at most 16 threads will be used; you can use eg --test-threads=128 if running on a large system.

The RNG used in the tests is deterministic, and the seed is logged for each execution. You can cause the random sequence to repeat using --drbg-seed option. It’s not necessary that the same sequence of tests be executed in order to replay the state; if you see a test foo fail with a specific DRBG seed, for example in a CI run, you should be able to replicate that with just botan-test --drbg-seed=<seed> foo.

If you are trying to track down a bug that happens only occasionally, two very useful options are --test-runs and --abort-on-first-fail. The first takes an integer and runs the specified test cases that many times. The second causes abort to be called on the very first failed test. This is sometimes useful when tracing a memory corruption bug.