Skip to content

Style Guide

Eric Voskuil edited this page Jan 8, 2017 · 11 revisions

Maximum of 80 characters per line [exceptions allowed for test vectors], 4 space indents, no tabs.

Each continued line should be nested by 1 indent. Parameter stacking is discouraged.

If you're nesting more than 3 levels deep then generally you are doing something wrong. Reorganize and/or break it into functions. Always prefer flat over nested, dumb over clever, slow+clear over fast+obtuse.

Use _ as opposed to - in source file names.

Variables, functions, class, enum, and typedef are lower case, define (macros) upper case, CamelCase identifiers for template types, no other mixed case.

Write self-documenting code. Names should be explanatory, not cryptic or ambiguous. Avoid abbreviations and acronyms.

Avoid Hungarian notation. Private class member variables are suffixed with _.

Functions should never be defined in headers unless required to be in every translation unit (i.e template, inline or constexpr functions).

It's nice if function names read like a sentence when used: write_to_stream(data_packet); or say_hello_to(my_friend);.

The using keyword should never be used in headers. Never declare using namespace std.

Every { begins a newline except for namespaces [there is some inconsistency here with respect to initializers].

Spaces between all operators: x = 5 * 8 + 4 * 2.

If the order is ambiguous or confusing then use parenthesis: x = (5 * 8) + (4 * 2).

Avoid pre/postfix operators in compound expressions: auto total = count++.

Avoid raw pointers in C++, using smart pointers if pointers are necessary.

Use standard integer types where possible:

  • Fixed-Length Unsigned Integers (<cstdint> / <stdint.h>)
    • int8_t
    • int16_t
    • int32_t
    • int64_t
  • Fixed-Length Signed Integers (<cstdint> / <stdint.h>)
    • uint8_t
    • uint16_t
    • uint32_t
    • uint64_t
  • Unspecified Length Unsigned Integer (<cstddef> / <stddef.h>)
    • size_t

Don't use long integers as this type varies in length by platform.

Don't assume sizeof(void*) == sizeof(size_t) as it is not guaranteed and it's semantically incorrect.

Generally you should use size_t for arbitrary length representation.

Don't treat non-boolean values or expressions as boolean, even in assertions, use comparison operators.

Use char for "characters", don't use unsigned char when you mean "byte", use uint8_t.

Assume char is interpreted as Utf-8, even in VC++! We have implemented "Utf-8 everywhere".

Use boost::filesystem::path for any part of a filesystem path, not std::string.

Each :, ; (in for loops) and , has a space after it.

No spaces between function/macros name and left parenthesis: do_bar(...). This greatly facilitates search.

Keywords such as if and forare not functions, one space before left parenthesis: while (!done) ....

Flow control should transition on new lines.

    // Bad
    if (foo) do_bar();

    // Good
    if (foo)
        do_bar();

Don't put more than one logical line on a line.

    // Bad
    something(); another_thing(); yada_yada();

    // Good
    something(); 
    another_thing(); 
    yada_yada();

Be careful with the use of assertions, they get removed in NDEBUG builds and can cause breaks and unused variable/parameter warnings.

Use assertions only to facilitate debugging, not for error handling.

Guard all parameters of published API calls against bad data. Assersions are insufficient guards for API parameters.

Avoid direct use of numeric literals other than zero in a zero-based index. Use sizeof() if you are referring to a type size. Use constexpr or const for magic numbers.

Do not accept implicit narrowing casts. These are bugs that haven't been discovered (and produce warnings). Use static_cast but only if the conversion is first guarded.

Test subtraction for underflow and addition for overflow, before performing the operation.

Don't use global variables or goto, we are grown-ups now.

Don't tolerate warnings and don't suppress them due to laziness.

Use //// to temporarily disable code.

Use /// for public API documentation (short form) or javadoc format.

Use // for source code commentary.

Follow the closing of a scope with a blank line before a subsequent statement.

Precede comments with a blank line.

Define only one class or struct per file and match the file name to the object.

Use the following source file layouts.

IPP Format

    // license statement
    #ifndef LIBBITCOIN_FOO_IPP
    #define LIBBITCOIN_FOO_IPP

    // system includes (sorted)
    // other library includes (sorted)
    // bitcoin library includes (sorted)

    namespace libbitcoin {

    template <size_t Size>
    bool do_foo(bytes<Size>& out)
    {
        // ...
    }

    } // namespace libbitcoin

    #endif

HPP Format

    // license statement
    #ifndef LIBBITCOIN_FOO_HPP
    #define LIBBITCOIN_FOO_HPP

    // system includes (sorted)
    // other library includes (sorted)
    // bitcoin library includes (sorted)

    namespace libbitcoin {
    
    class foo
      : public bar
    {
    public:
        // typedefs
        // method declarations
        // member declarations

    private:
        // typedefs
        // method declarations
        // member declarations
    };

    } // namespace libbitcoin
    
    // function declarations
    // template includes (ipp) 

    #endif

CPP Format

    // license statement
    #include <bitcoin/foo.hpp>

    // system includes (sorted)
    // other library includes (sorted)
    // bitcoin library includes (sorted)

    namespace libbitcoin {

    // function implementations 
    // method implementations

    } // libbitcoin

H Format

    // license statement
    #ifndef LIBBITCOIN_BAR_H
    #define LIBBITCOIN_BAR_H

    // system includes (sorted)
    // other library includes (sorted)
    // bitcoin library includes (sorted)

#ifdef __cplusplus
extern "C" 
{
#endif

    // constant declarations
    // function declarations

#ifdef __cplusplus
}
#endif

#endif

C Format

    // license statement
    #include <bitcoin/bar.h>

    // system includes (sorted)
    // other library includes (sorted)
    // bitcoin library includes (sorted)
    
    // constant definitions
    // function implementations

Test Considerations

Tests have not been fully converted to this style in all repositories.

It is important that tests can be evaluated visually. Otherwise it becomes impossible to determine if the test itself is correct. To achieve this result we apply certain style rules as follows:

Declarativeness.

This means that test execution is unconditional, there is no branching. Loops, if statements and the ternatory operator are prohibited. Test helpers may include conditional logic, but then these should be tested for correctness.

Transparent Names

When a test or set of tests fail this helps a developer to quickly isolate the cause. The naming convention is:

BOOST_AUTO_TEST_CASE([class|file]__[method|function]__[condition]__[expectation])

For example:

BOOST_AUTO_TEST_CASE(reservation__stopped__import_last_block__true)

Tests against a method, constructor or function override should use a simple ordinal numbering scheme to group test of the same signature. For example:

BOOST_AUTO_TEST_CASE(configuration__construct1__none_context__expected)
BOOST_AUTO_TEST_CASE(configuration__construct1__mainnet_context__expected)
BOOST_AUTO_TEST_CASE(configuration__construct1__testnet_context__expected)
BOOST_AUTO_TEST_CASE(configuration__construct2__none_context__expected)

Group tests in a single .cpp file to mirror the source file naming convention. If it becomes necessary to break up the test file into independent files for a single corresponding class or source file, use the following test file naming convention:

class__method.cpp
file__function.cpp

Group tests semantically using BOOST_AUTO_TEST_SUITE(...). This allows tests to be executed independently. These can be grouped hierarchically if necessary. Generally there is one suite per file or class under test.

Test One Thing

Test just one thing. The test name indicates the conditions and expectation of the test. When a test or set of tests fail this allows a developer to quickly isolate the cause. Sometimes it is convenient to test a set of results under one condition, such as a true result code and an expected out value. But these combinations should be semantically associated in that they are necessary to test together.

Line Length

A line of test code should not be wrapped. For this reason line length is not limited as in source files. However intermediate variables can and should be used to limit line length.

Test Helpers

The BOOST test helpers expose information about the test when there is a failure. It is preferred to use the least general test helper available. For example, use:

BOOST_REQUIRE(!foo.is_valid())

instead of

BOOST_REQUIRE_EQUAL(foo.is_valid(), false)

If values are serializable then use:

BOOST_REQUIRE_EQUAL(foo.size(), 5u)

as opposed to

BOOST_REQUIRE(foo.size() == 5u)

and in all cases placing the condition under test one the left side and the test expectation on the right.

Libbitcoin Menu

Clone this wiki locally