6. Coding style

When writing code to be contributed to the ns-3 open source project, we ask that you follow the coding standards, guidelines, and recommendations found below.

6.1. Clang-format

The ns-3 project uses clang-format to define and enforce the C++ coding style. Clang-format can be easily integrated with modern IDEs or run manually on the command-line.

Besides clang-format, ns-3 adopts other coding-style guidelines that are not covered by clang-format, which are explained in this document. Read the check-style-clang-format.py section below for information on how to use this Python script to check and fix all formatting guidelines followed by ns-3.

6.1.1. Clang-format installation

Clang-format can be installed using your OS’s package manager. Please note that you should install one of the supported versions of clang-format, which are listed in the following section.

6.1.2. Supported versions of clang-format

Since each new major version of clang-format can add or modify properties, newer versions of clang-format might produce different outputs compared to previous versions.

The following list contains the set of clang-format versions that are verified to produce consistent output among themselves.

• clang-format-17

• clang-format-16

• clang-format-15

• clang-format-14

6.1.3. Integration with IDEs

Clang-format can be integrated with modern IDEs (e.g., VS Code) that are able to read the .clang-format file and automatically format the code on save or on typing.

Please refer to the documentation of your IDE for more information. Some examples of IDE integration are provided in clang-format documentation

As an example, VS Code can be configured to automatically format code on save, on paste and on type by enabling the following settings:

{
  "editor.formatOnSave": true,
  "editor.formatOnPaste": true,
  "editor.formatOnType": true,
}

6.1.4. Clang-format usage in the terminal

In addition to IDE support, clang-format can be manually run on the terminal.

To format a file in-place, run the following command:

clang-format -i $FILE

To check that a file is well formatted, run the following command on the terminal. If the file is not well formatted, clang-format indicates the lines that need to be formatted.

clang-format --dry-run $FILE

6.1.5. Clang-format Git integration

Clang-format integrates with Git to format Git commits or changes not yet committed, such as pending merge requests on the GitLab repository. The full documentation is available on clang-format Git integration

To fix the formatting of files with Git, run the following commands in the ns-3 main directory. These commands do not change past commits. Instead, the reformatted files are left in the workspace. These changes should be squashed to the corresponding commits, in order to fix them.

# Fix all commits of the current branch, relative to the master branch
git clang-format master

# Fix all staged changes (i.e., changes that have been `git add`ed):
git clang-format

# Fix all changes staged and unstaged:
git clang-format -f

# Fix specific files:
git clang-format path_to_file

# Check what formatting changes are needed (if no files provided, check all staged files):
git clang-format --diff

Note that this only fixes formatting issues related to clang-format. For other ns-3 coding style guidelines, read the check-style-clang-format.py section below.

In addition to Git patches, clang-format-diff can also be used to reformat existing patches produced with the diff tool.

6.1.6. Disable formatting in specific files or lines

To disable formatting in specific lines, surround them with the following C++ comments:

// clang-format off
...
// clang-format on

To exclude an entire file from being formatted, surround the whole file with the special comments.

6.2. check-style-clang-format.py

To facilitate checking and fixing source code files according to the ns-3 coding style, ns-3 maintains the check-style-clang-format.py Python script (located in utils/). This script is a wrapper to clang-format and provides useful options to check and fix source code files. Additionally, it performs other manual checks and fixes in text files (described below).

We recommend running this script over your newly introduced C++ files prior to submission as a Merge Request.

The script performs multiple style checks. By default, the script runs the following checks:

• Check code formatting using clang-format. Respects clang-format guards.

• Check if local #include headers do not use the “ns3/” prefix. Respects clang-format guards.

• Check if Doxygen tags use @ rather than \\. Respects clang-format guards.

• Check if there are no trailing whitespace. Always checked.

• Check if there are no tabs. Respects clang-format guards.

• Check if source code files use SPDX licenses rather than GPL license text. Respects clang-format guards.

• Check if files have the correct encoding (UTF-8). Always checked.

The process returns a zero exit code if all files adhere to these rules. If there are files that do not comply with the rules, the process returns a non-zero exit code and lists the respective files. This mode is useful for developers editing their code and for the GitLab CI/CD pipeline to check if the codebase is well formatted. All checks are enabled by default. Users can disable specific checks using the corresponding flags:

• --no-formatting

• --no-include-prefixes

• --no-doxygen-tags

• --no-whitespace

• --no-tabs

• --no-licenses

• --no-encoding

Additional information about the formatting issues detected by the script can be enabled by adding the -v, --verbose flag.

In addition to checking the files, the script can automatically fix detected issues in-place. This mode is enabled by adding the --fix flag.

The formatting and tabs checks respect clang-format guards, which mark code blocks that should not be checked. Trailing whitespace is always checked regardless of clang-format guards.

The complete API of the check-style-clang-format.py script can be obtained with the following command:

./utils/check-style-clang-format.py --help

For quick-reference, the most used commands are listed below:

# Entire codebase (using paths relative to the ns-3 main directory)
./utils/check-style-clang-format.py --fix .

# Entire codebase (using absolute paths)
/path/to/utils/check-style-clang-format.py --fix /path/to/ns3

# Specific directory or file
/path/to/utils/check-style-clang-format.py --fix absolute_or_relative/path/to/directory_or_file

# Files modified by the current branch, relative to the master branch
git diff --name-only master | xargs ./utils/check-style-clang-format.py --fix

6.3. Clang-tidy

The ns-3 project uses clang-tidy to statically analyze (lint) C++ code and help developers write better code. Clang-tidy can be easily integrated with modern IDEs or run manually on the command-line.

The list of clang-tidy checks currently enabled in the ns-3 project is saved on the .clang-tidy file. The full list of clang-tidy checks and their detailed explanations can be seen in Clang-tidy checks.

6.3.1. Clang-tidy installation

Clang-format can be installed using your OS’s package manager. Please note that you should install one of the supported versions of clang-format, which are listed in the following section.

6.3.2. Minimum clang-tidy version

Since clang-tidy is a linter that analyzes code and outputs errors found during the analysis, developers can use different versions of clang-tidy on the workflow. Newer versions of clang-tidy might produce better results than older versions. Therefore, it is recommended to use the latest version available.

To ensure consistency among developers, ns-3 defines a minimum version of clang-tidy, whose warnings must not be ignored. Therefore, developers should, at least, scan their code with the minimum version of clang-tidy.

The minimum version is clang-tidy-14.

6.3.3. Integration with IDEs

Clang-tidy automatically integrates with modern IDEs (e.g., VS Code) that read the .clang-tidy file and automatically checks the code of the currently open file.

Please refer to the documentation of your IDE for more information. Some examples of IDE integration are provided in clang-tidy documentation

6.3.4. Clang-tidy usage

In order to use clang-tidy, ns-3 must first be configured with the flag --enable-clang-tidy. To configure ns-3 with tests, examples and clang-tidy enabled, run the following command:

./ns3 configure --enable-examples --enable-tests --enable-clang-tidy

Due to the integration of clang-tidy with CMake, clang-tidy can be run while building ns-3. In this way, clang-tidy errors will be shown alongside build errors on the terminal. To build ns-3 and run clang-tidy, run the the following command:

./ns3 build

To run clang-tidy without building ns-3, use the following commands on the ns-3 main directory:

# Analyze (and fix) a single file with clang-tidy
clang-tidy -p cmake-cache/ [--fix] [--format-style=file] [--quiet] $FILE

# Analyze (and fix) multiple files in parallel
run-clang-tidy -p cmake-cache/ [-fix] [-format] [-quiet] $FILE1 $FILE2 ...

# Analyze (and fix) the entire ns-3 codebase in parallel
run-clang-tidy -p cmake-cache/ [-fix] [-format] [-quiet]

When running clang-tidy, please note that:

• Clang-tidy only analyzes implementation files (i.e., *.cc files). Header files are analyzed when they are included by implementation files, with the #include "..." directive.

• Not all clang-tidy checks provide automatic fixes. For those cases, a manual fix must be made by the developer.

• Enabling clang-tidy will add time to the build process (in the order of minutes).

6.3.5. Disable clang-tidy analysis in specific lines

To disable clang-tidy analysis of a particular rule in a specific function, specific clang-tidy comments have to be added to the corresponding function. Please refer to the official clang-tidy documentation for more information.

To disable modernize-use-override checking on func() only, use one of the following two “special comment” syntaxes:

//
// Syntax 1: Comment above the function
//

// NOLINTNEXTLINE(modernize-use-override)
void func();

//
// Syntax 2: Trailing comment
//

void func(); // NOLINT(modernize-use-override)

To disable a specific clang-tidy check on a block of code, for instance the modernize-use-override check, surround the code block with the following “special comments”:

// NOLINTBEGIN(modernize-use-override)
void func1();
void func2();
// NOLINTEND(modernize-use-override)

To disable all clang-tidy checks on a block of code, surround it with the following “special comments”:

// NOLINTBEGIN
void func1();
void func2();
// NOLINTEND

To exclude an entire file from being checked, surround the whole file with the “special comments”.

6.4. Source code formatting

The ns-3 coding style was changed between the ns-3.36 and ns-3.37 release. Prior to ns-3.37, ns-3 used a base GNU coding style. Since ns-3.37, ns-3 changed the base coding style to what is known in the industry as Allman-style braces, with four-space indentation. In clang-format, this is configured by selecting the Microsoft base style. The following examples illustrate the style.

6.4.1. Indentation

Indent code with 4 spaces. When breaking statements into multiple lines, indent the following lines with 4 spaces.

void
Func()
{
    int x = 1;
}

Indent constructor’s initialization list with 4 spaces.

MyClass::MyClass(int x, int y)
    : m_x(x),
      m_y(y)
{
}

Do not use tabs in source code. Always use spaces for indentation and alignment.

6.4.2. Line endings

Files use LF (\n) line endings.

6.4.3. Column limit

Code lines should not extend past 100 characters. This allows reading code in wide-screen monitors without having to scroll horizontally, while also allowing editing two files side-by-side.

6.4.4. Braces style

Braces should be formatted according to the Allman style. Braces are always on a new line and aligned with the start of the corresponding block. The main body is indented with 4 spaces.

Always surround conditional or loop blocks (e.g., if, for, while) with braces, and always add a space before the condition’s opening parentheses.

void Foo()
{
    if (condition)
    {
        // do stuff here
    }
    else
    {
        // do other stuff here
    }

    for (int i = 0; i < 100; i++)
    {
        // do loop
    }

    while (condition)
    {
        // do while
    }

    do
    {
        // do stuff
    } while (condition);
}

6.4.5. Spacing

To increase readability, functions, classes and namespaces are separated by one new line. This spacing is optional when declaring variables or functions. Declare one variable per line. Do not mix multiple statements on the same line.

Do not add a space between the function name and the opening parentheses. This rule applies to both function (and method) declarations and invocations.

void Func(const T&);  // OK
void Func (const T&); // Not OK

6.4.6. Trailing whitespace

Source code and text files must not have trailing whitespace.

6.4.7. Code alignment

To improve code readability, trailing comments should be aligned.

int varOne;    // Variable one
double varTwo; // Variable two

The trailing \ character of macros should be aligned to the far right (equal to the column limit). This increases the readability of the macro’s body, without forcing unnecessary whitespace diffs on surrounding lines when only one line is changed.

#define MY_MACRO(msg)                      \
    do                                     \
    {                                      \
        std::cout << msg << std::endl;     \
    } while (false);

6.4.8. Class members

Definition blocks within a class should be organized in descending order of public exposure, that is: static > public > protected > private. Separate each block with a new line.

class MyClass
{
public:
    static int m_counter = 0;

    MyClass(int x, int y);

private:
    int x;
    int y;
};

6.4.9. Function arguments bin packing

Function arguments should be declared in the same line as the function declaration. If the arguments list does not fit the maximum column width, declare each one on a separate line and align them vertically.

void ShortFunction(int x, int y);

void VeryLongFunctionWithLongArgumentList(int x,
                                          int y,
                                          int z);

The constructor initializers are always declared one per line, with a trailing comma:

void
MyClass::MyClass(int x, int y)
    : m_x(x),
      m_y(y)
{
}

6.4.10. Function return types

In function declarations, return types are declared on the same line. In function implementations, return types are declared on a separate line.

// Function declaration
void Func(int x, int y);

// Function implementation
void
Func(int x, int y)
{
    // (...)
}

6.4.11. Templates

Template definitions are always declared in a separate line from the main function declaration:

template <class T>
void Func(T t);

6.4.12. Naming

6.4.12.1. Name encoding

Function, method, and type names should follow the CamelCase convention: words are joined without spaces and are capitalized. For example, “my computer” is transformed into MyComputer. Do not use all capital letters such as MAC or PHY, but choose instead Mac or Phy. Do not use all capital letters, even for acronyms such as EDCA; use Edca instead. This applies also to two-letter acronyms, such as IP (which becomes Ip). The goal of the CamelCase convention is to ensure that the words which make up a name can be separated by the eye: the initial Caps fills that role. Use PascalCasing (CamelCase with first letter capitalized) for function, property, event, and class names.

Variable names should follow a slight variation on the base CamelCase convention: camelBack. For example, the variable user name would be named userName. This variation on the basic naming pattern is used to allow a reader to distinguish a variable name from its type. For example, UserName userName would be used to declare a variable named userName of type UserName.

Global variables should be prefixed with a g_ and member variables (including static member variables) should be prefixed with a m_. The goal of that prefix is to give a reader a sense of where a variable of a given name is declared to allow the reader to locate the variable declaration and infer the variable type from that declaration. Defined types will start with an upper case letter, consist of upper and lower case letters, and may optionally end with a _t. Defined constants (such as static const class members, or enum constants) will be all uppercase letters or numeric digits, with an underscore character separating words. Otherwise, the underscore character should not be used in a variable name. For example, you could declare in your class header my-class.h:

typedef int NewTypeOfInt_t;
constexpr uint8_t PORT_NUMBER = 17;

class MyClass
{
    void MyMethod(int aVar);
    int m_aVar;
    static int m_anotherVar;
};

and implement in your class file my-class.cc:

int MyClass::m_anotherVar = 10;
static int g_aStaticVar = 100;
int g_aGlobalVar = 1000;

void
MyClass::MyMethod(int aVar)
{
    m_aVar = aVar;
}

As an exception to the above, the members of structures do not need to be prefixed with an m_.

Finally, do not use Hungarian notation, and do not prefix enums, classes, or delegates with any letter.

6.4.12.2. Choosing names

Variable, function, method, and type names should be based on the English language, American spelling. Furthermore, always try to choose descriptive names for them. Types are often english names such as: Packet, Buffer, Mac, or Phy. Functions and methods are often named based on verbs and adjectives: GetX, DoDispose, ClearArray, etc.

A long descriptive name which requires a lot of typing is always better than a short name which is hard to decipher. Do not use abbreviations in names unless the abbreviation is really unambiguous and obvious to everyone (e.g., use size over sz). Do not use short inappropriate names such as foo, bar, or baz. The name of an item should always match its purpose. As such, names such as tmp to identify a temporary variable, or such as i to identify a loop index are OK.

If you use predicates (that is, functions, variables or methods which return a single boolean value), prefix the name with “is” or “has”.

6.4.13. File layout and code organization

A class named MyClass should be declared in a header named my-class.h and implemented in a source file named my-class.cc. The goal of this naming pattern is to allow a reader to quickly navigate through the ns-3 codebase to locate the source file relevant to a specific type.

Each my-class.h header should start with the following comment to ensure that your code is licensed under the GPL, that the copyright holders are properly identified (typically, you or your employer), and that the actual author of the code is identified. The latter is purely informational and we use it to try to track the most appropriate person to review a patch or fix a bug. Please do not add the “All Rights Reserved” phrase after the copyright statement.

/*
 * Copyright (c) YEAR COPYRIGHTHOLDER
 *
 * SPDX-License-Identifier: GPL-2.0-only
 *
 * Author: MyName <myemail@example.com>
 */

Below these C-style comments, always include the following which defines a set of header guards (MY_CLASS_H) used to avoid multiple header includes, which ensures that your code is included in the ns-3 namespace and which provides a set of Doxygen comments for the public part of your class API. Detailed information on the set of tags available for doxygen documentation is described in the Doxygen website.

#ifndef MY_CLASS_H
#define MY_CLASS_H

namespace ns3
{

/**
 * @brief short one-line description of the purpose of your class
 *
 * A longer description of the purpose of your class after a blank
 * empty line.
 */
class MyClass
{
public:
    MyClass();

    /**
     * A detailed description of the purpose of the method.
     *
     * @param firstParam a short description of the purpose of this parameter
     * @return a short description of what is returned from this function.
     */
    int DoSomething(int firstParam);

private:
    /**
     * Private method doxygen is also recommended
     */
    void MyPrivateMethod();

    int m_myPrivateMemberVariable; ///< Brief description of member variable
};

} // namespace ns3

#endif // MY_CLASS_H

The my-class.cc file is structured similarly:

/*
 * Copyright (c) YEAR COPYRIGHTHOLDER
 *
 * SPDX-License-Identifier: GPL-2.0-only
 *
 * Author: MyName <myemail@foo.com>
 */

#include "my-class.h"

namespace ns3
{

MyClass::MyClass()
{
}

...

} // namespace ns3

6.4.14. Header file includes

Included header files should be organized by source location. The sorting order is as follows:

// Header class (applicable for *.cc files)
#include "my-class.h"

// Includes from the same module
#include "header-from-same-module.h"

// Includes from other modules
#include "ns3/header-from-different-module.h"

// External headers (e.g., STL libraries)
#include <iostream>

Groups should be separated by a new line. Within each group, headers should be sorted alphabetically.

For standard headers, use the C++ style of inclusion:

#include <cstring>  // OK
#include <string.h> // Avoid

• inside .h files, always use

  #include <ns3/header.h>
  

• inside .cc files, use

  #include "header.h"
  

  if file is in same directory, otherwise use

  #include <ns3/header.h>
  

6.4.15. Variables and constants

Each variable declaration is on a separate line. Variables should be declared at the point in the code where they are needed, and should be assigned an initial value at the time of declaration.

// Do not declare multiple variables per line
int x, y;

// Declare one variable per line and assign an initial value
int x = 0;
int y = 0;

Named constants defined in classes should be declared as static constexpr instead of macros, const, or enums. Use of static constexpr allows a single instance to be evaluated at compile-time. Declaring the constant in the class enables it to share the scope of the class.

If the constant is only used in one file, consider declaring the constant in the implementation file (*.cc).

// Avoid declaring constants as enum
class LteRlcAmHeader : public Header
{
    enum ControlPduType_t
    {
        STATUS_PDU = 000,
    };
};

// Prefer to declare them as static constexpr (in class)
class LteRlcAmHeader : public Header
{
    static constexpr uint8_t STATUS_PDU{0};
};

// Or as constexpr (in implementation files)
constexpr uint8_t STATUS_PDU{0};

When declaring variables that are easily deducible from context, prefer to declare them with auto instead of repeating the type name. Not only does this improve code readability, by making lines shorter, but it also facilitates future code refactoring.

// Avoid repeating the type name when declaring iterators or pointers, and casting variables
std::map<uint32_t, std::string>::const_iterator it = myMap.find(key);
int* ptr = new int[10];
uint8_t m = static_cast<uint8_t>(97 + (i % 26));

// Prefer to declare them with auto
auto it = myMap.find(key);
auto* ptr = new int[10];
auto m = static_cast<uint8_t>(97 + (i % 26));

6.4.16. Initialization

When declaring variables, prefer to use direct-initialization, to avoid repeating the type name.

// Avoid splitting the declaration and initialization of variables
Ipv4Address ipv4Address = Ipv4Address("192.168.0.1")

// Prefer to use direct-initialization
Ipv4Address ipv4Address("192.168.0.1")

Variables with no default constructor or of primitive types should be initialized when declared.

Variables with default constructors do not need to be explicitly initialized, since the compiler already does that. An example of this is the ns3::Time class, which will initialize to zero.

Member variables of structs and classes should be initialized unless the member has a default constructor that guarantees initialization. Preferably, variables should be initialized together with the declaration (in the header file). Alternatively, they can be initialized in the default constructor (in the implementation file), and you may see instances of this in the codebase, but direct initialization upon declaration is preferred going forward.

If all member variables of a class / struct are directly initialized (see above), they do not require explicit default initialization. But if not all variables are initialized, those non-initialized variables will contain garbage. Therefore, initializing the class object with {} allows all member variables to always be initialized – either with the provided default initialization or with the primitive type’s default value (typically 0).

C++ supports two syntax choices for direct initialization, either () or {}. There are various tradeoffs in the choices for more complicated types (consult the C++ literature on brace vs. parentheses initialization), but for the fundamental types like double, either is acceptable (please use consistently within files).

Regarding ns3::Time, do not initialize to non-zero integer values as follows, assuming that it will be converted to nanoseconds:

Time t{1000000};  // This is disallowed

The value will be interpreted according to the current resolution, which is ambiguous. A user’s program may have already changed the resolution from the default of nanoseconds to something else by the time of this initialization, and it will be instead interpreted according to 10^6 * the new resolution unit.

Time initialization to raw floating-point values is additionally fraught, because of rounding. Doing so with small values has led to bugs in practice such as timer timeout values of zero time.

When declaring or manipulating Time objects with known values, prefer to use integer-based representations and arguments over floating-point fractions, where possible, because integer-based is faster. This means preferring the use of NanoSeconds, MicroSeconds, and MilliSeconds over Seconds. For example, to represent a tenth of a second, prefer MilliSeconds(100) to Seconds(0.1).

To summarize Time declaration and initialization, consider the following examples and comments:

Time t;  // OK, will be value-initialized to integer zero
Time t{MilliSeconds(100)};  // OK, fastest, no floating point involved
Time t{"100ms"}; // OK, will perform a string conversion; integer would be faster
Time t{Seconds(0.1)};  // OK, will invoke Seconds(double); integer would be faster
Time t{100000000}; // NOT OK, is interpreted differently when ``Time::SetResolution()`` called
Time t{0.1}; // NOT OK, will round to zero; see above and also merge request !2007

6.4.17. Comments

The project uses Doxygen to document the interfaces, and uses comments for improving the clarity of the code internally. All classes, methods, and members should have Doxygen comments. Doxygen comments should use the C-style comment (also known as Javadoc) style. For comments that are intended to not be exposed publicly in the Doxygen output, use the @internal and @endinternal tags. Please use the @see tag for cross-referencing. All parameters and return values should be documented. The ns-3 codebase prefers the @ character for tag identification. This character is recognized by clang-format as the start of Doxygen tags, which enables it to keep tags properly formatted; therefore please don’t use \ as the delimiter.

/**
 * MyClass description.
 */
class MyClass
{
    /**
     * Constructor.
     *
     * @param n Number of elements.
     */
    MyClass(int n);
};

All the functions and variables must be documented, with the exception of member functions inherited from parent classes (the documentation is copied automatically from the parent class), and default constructor/destructor.

It is strongly suggested to use grouping to bind together logically related classes (e.g., all the classes in a module). E.g.;

/**
 * @defgroup mynewmodule This is a new module
 */

/**
 * @ingroup mynewmodule
 *
 * MyClassOne description.
 */
class MyClassOne
{
};

/**
 * @ingroup mynewmodule
 *
 * MyClassTwo description.
 */
class MyClassTwo
{
};

In the tests for the module, it is suggested to add an ancillary group:

/**
 * @defgroup mynewmodule-test Tests for new module
 * @ingroup mynewmodule
 * @ingroup tests
 */

/**
 * @ingroup mynewmodule-tests
 * @brief MyNewModule Test
 */
class MyNewModuleTest : public TestCase
{
};

/**
 * @ingroup mynewmodule-tests
 * @brief MyNewModule TestSuite
 */
class MyNewModuleTestSuite : public TestSuite
{
  public:
    MyNewModuleTestSuite();
};

/**
 * @ingroup mynewmodule-tests
 * Static variable for test initialization
 */
static MyNewModuleTestSuite g_myNewModuleTestSuite;

As for comments within the code, comments should be used to describe intention or algorithmic overview where is it not immediately obvious from reading the code alone. There are no minimum comment requirements and small routines probably need no commenting at all, but it is hoped that many larger routines will have commenting to aid future maintainers. Please write complete English sentences and capitalize the first word unless a lower-case identifier begins the sentence. Two spaces after each sentence helps to make emacs sentence commands work. Sometimes NS_LOG_DEBUG statements can be also used in place of comments.

Short one-line comments and long comments can use the C++ comment style; that is, //, but longer comments may use C-style comments. Use one space after // or /*.

/*
 * A longer comment,
 * with multiple lines.
 */

Variable declaration should have a short, one or two line comment describing the purpose of the variable, unless it is a local variable whose use is obvious from the context. The short comment should be on the same line as the variable declaration, unless it is too long, in which case it should be on the preceding lines.

int nNodes = 3; // Number of nodes

/// Node container with the Wi-Fi stations
NodeContainer wifiStations(3);

6.4.18. Casts

Where casts are necessary, use the Google C++ guidance: “Use C++-style casts like static_cast<float>(double_value), or brace initialization for conversion of arithmetic types like int64 y = int64{1} << 42.” Do not use C-style casts, since they can be unsafe.

Try to avoid (and remove current instances of) casting of uint8_t type to larger integers in our logging output by overriding these types within the logging system itself. Also, the unary + operator can be used to print the numeric value of any variable, such as:

uint8_t flags = 5;
std::cout << "Flags numeric value: " << +flags << std::endl;

Avoid unnecessary casts if minor changes to variable declarations can solve the issue. In the following example, x can be declared as float instead of int to avoid the cast, or write numbers in decimal format:

// Do not declare x as int, to avoid casting it to float
int x = 3;
float y = 1 / static_cast<float>(x);

// Prefer to declare x as float
float x = 3.0;
float y = 1 / x;

// Or use 1.0 instead of just 1
int x = 3;
float y = 1.0 / x;

6.4.19. Namespaces

Code should always be included in a given namespace, namely ns3. In order to avoid exposing internal symbols, consider placing the code in an anonymous namespace, which can only be accessed by functions in the same file.

Code within namespaces should not be indented. To more easily identify the end of a namespace, add a trailing comment to its closing brace.

namespace ns3
{

// (...)

} // namespace ns3

Namespace names should follow the snake_case convention.

6.4.20. Unused variables

Compilers will typically issue warnings on unused entities (e.g., variables, function parameters). Use the [[maybe_unused]] attribute to suppress such warnings when the entity may be unused depending on how the code is compiled (e.g., if the entity is only used in a logging statement or an assert statement).

The general guidelines are as follows:

• If a function’s or a method’s parameter is definitely unused, prefer to leave it unnamed. In the following example, the second parameter is unnamed.

  void
  UanMacAloha::RxPacketGood(Ptr<Packet> pkt, double, UanTxMode txMode)
  {
      UanHeaderCommon header;
      pkt->RemoveHeader(header);
      ...
  }
  

  In this case, the parameter is also not referenced by Doxygen; e.g.,:

  /**
   * Receive packet from lower layer (passed to PHY as callback).
   *
   * @param pkt Packet being received.
   * @param txMode Mode of received packet.
   */
  void RxPacketGood(Ptr<Packet> pkt, double, UanTxMode txMode);
  

  The omission is preferred to commenting out unused parameters, such as:

  void
  UanMacAloha::RxPacketGood(Ptr<Packet> pkt, double /*sinr*/, UanTxMode txMode)
  {
      UanHeaderCommon header;
      pkt->RemoveHeader(header);
      ...
  }
  

• If a function’s parameter is only used in certain cases (e.g., logging), or it is part of the function’s Doxygen, mark it as [[maybe_unused]].

  void
  TcpSocketBase::CompleteFork(Ptr<Packet> p [[maybe_unused]],
                              const TcpHeader& h,
                              const Address& fromAddress,
                              const Address& toAddress)
  {
      NS_LOG_FUNCTION(this << p << h << fromAddress << toAddress);
  
      // Remaining code that definitely uses 'h', 'fromAddress' and 'toAddress'
      ...
  }
  

• If a local variable saves the result of a function that must always run, but whose value may not be used, declare it [[maybe_unused]].

  void
  MyFunction()
  {
      int result [[maybe_unused]] = MandatoryFunction();
      NS_LOG_DEBUG("result = " << result);
  }
  

• If a local variable saves the result of a function that is only run in certain cases, prefer to not declare the variable and use the function’s return value directly where needed. This avoids unnecessarily calling the function if its result is not used.

  void
  MyFunction()
  {
      // Prefer to call GetDebugInfo() directly on the log statement
      NS_LOG_DEBUG("Debug information: " << GetDebugInfo());
  
      // Avoid declaring a local variable with the result of GetDebugInfo()
      int debugInfo [[maybe_unused]] = GetDebugInfo();
      NS_LOG_DEBUG("Debug information: " << debugInfo);
  }
  

  If the calculation of the maybe unused variable is complex, consider wrapping the calculation of its value in a conditional block that is only run if the variable is used.

  if (g_log.IsEnabled(ns3::LOG_DEBUG))
  {
      auto debugInfo = GetDebugInfo();
      auto value = DoComplexCalculation(debugInfo);
  
      NS_LOG_DEBUG("The value is " << value);
  }
  

6.4.21. Unnecessary else after return

In order to increase readability and avoid deep code nests, consider not adding an else block if the if block breaks the control flow (i.e., when using return, break, continue, etc.).

For instance, the following code:

if (n < 0)
{
    return false;
}
else
{
    n += 3;
    return n;
}

can be rewritten as:

if (n < 0)
{
    return false;
}

n += 3;
return n;

6.4.22. Boolean Simplifications

In order to increase readability and performance, avoid unnecessarily complex boolean expressions in if statements and variable declarations.

For instance, the following code:

bool IsPositive(int n)
{
    if (n > 0)
    {
        return true;
    }
    else
    {
        return false;
    }
}

void ProcessNumber(int n)
{
    if (IsPositive(n) == true)
    {
        ...
    }
}

can be rewritten as:

bool IsPositive(int n)
{
    return n > 0;
}

void ProcessNumber(int n)
{
    if (IsPositive(n))
    {
        ...
    }
}

6.4.23. Smart pointer boolean comparisons

As explained in this issue, the ns-3 smart pointer class Ptr should be used in boolean comparisons as follows:

for Ptr<> p, do not use:            use instead:
========================            =================================
if (p != nullptr) {...}             if (p)      {...}
if (p != NULL)    {...}
if (p != 0)       {...}             if (p)      {...}

if (p == nullptr) {...}             if (!p)     {...}
if (p == NULL)    {...}
if (p == 0)       {...}

NS_ASSERT...(p != nullptr, ...)    NS_ASSERT...(p, ...)
NS_ABORT... (p != nullptr, ...)    NS_ABORT... (p, ...)

NS_ASSERT...(p == nullptr, ...)    NS_ASSERT...(!p, ...)
NS_ABORT... (p == nullptr, ...)    NS_ABORT... (!p, ...)

NS_TEST...  (p, nullptr, ...)      NS_TEST...  (p, nullptr, ...)

6.4.24. Code performance tips

While developing code, consider the following tips to improve the code’s performance. Some tips are general recommendations, but are not strictly enforced. Other tips are enforced by clang-tidy. Please refer to the clang-tidy section below for more details.

• Prefer to use .emplace_back() over .push_back() to optimize performance.

• When initializing STL containers (e.g., std::vector) with known size, reserve memory to store all items, before pushing them in a loop.

  constexpr int N_ITEMS = 5;
  
  std::vector<int> myVector;
  myVector.reserve(N_ITEMS); // Reserve memory to store all items
  
  for (int i = 0; i < N_ITEMS; i++)
  {
      myVector.emplace_back(i);
  }
  

• Prefer to initialize STL containers (e.g., std::vector, std::map, etc.) directly through the constructor or with a braced-init-list, instead of pushing elements one-by-one.

  // Prefer to initialize containers directly
  std::vector<int> myVector1{1, 2, 3};
  std::vector<int> myVector2(myVector1.begin(), myVector1.end());
  std::vector<bool> myVector3(myVector2.size(), true);
  
  // Avoid pushing elements one-by-one
  std::vector<int> myVector1;
  myVector1.reserve(3);
  myVector1.emplace_back(1);
  myVector1.emplace_back(2);
  myVector1.emplace_back(3);
  
  std::vector<int> myVector2;
  myVector2.reserve(myVector1.size());
  for (const auto& v : myVector1)
  {
      myVector2.emplace_back(v);
  }
  
  std::vector<bool> myVector3;
  myVector3.reserve(myVector1.size());
  for (std::size_t i = 0; i < myVector1.size(); i++)
  {
      myVector3.emplace_back(true);
  }
  

• When looping through containers, prefer to use const-ref syntax over copying elements.

  std::vector<int> myVector{1, 2, 3};
  
  for (const auto& v : myVector) { ... }  // OK
  for (auto v : myVector) { ... }         // Avoid
  

• Prefer to use the empty() function of STL containers (e.g., std::vector), instead of the condition size() > 0, to avoid unnecessarily calculating the size of the container.

• Avoid unnecessary calls to the functions .c_str() and .data() of std::string.

• Avoid unnecessarily dereferencing std smart pointers (std::shared_ptr, std::unique_ptr) with calls to their member function .get(). Prefer to use the std smart pointer directly where needed.

  auto ptr = std::make_shared<Node>();
  
  // OK
  if (ptr) { ... }
  
  // Avoid
  if (ptr.get()) { ... }
  

• Consider caching frequently-used results (especially expensive calculations, such as mathematical functions) in a temporary variable, instead of calculating them in every loop.

  // Prefer to cache intermediate results
  const double sinTheta = std::sin(theta);
  const double cosTheta = std::cos(theta);
  
  for (uint8_t i = 0; i < NUM_VALUES; i++)
  {
      double power = std::pow(2, i);
  
      array1[i] = (power * sinTheta) + cosTheta;
      array2[i] = (power * cosTheta) + sinTheta;
  }
  
  // Avoid repeating calculations
  for (uint8_t i = 0; i < NUM_VALUES; i++)
  {
      array1[i] = (std::pow(2, i) * std::sin(theta)) + std::cos(theta);
      array2[i] = (std::pow(2, i) * std::cos(theta)) + std::sin(theta);
  }
  

• Do not include inline implementations in header files; put all implementation in a .cc file (unless implementation in the header file brings demonstrable and significant performance improvement).

• Avoid declaring trivial destructors, to optimize performance.

6.4.25. C++ standard

As of ns-3.36, ns-3 permits the use of C++-17 (or earlier) features in the implementation files.

If a developer would like to propose to raise this bar to include more features than this, please email the developers list. We will move this language support forward as our minimally supported compiler moves forward.

6.4.26. Guidelines for using maps

Maps (associative containers) are used heavily in ns-3 models to store key/value pairs. The C++ standard, over time, has added various methods to insert elements to maps, and the ns-3 codebase has made use of most or all of these constructs. For the sake of uniformity and readability, the following guidelines are recommended for any new code.

Prefer the use of std::map to std::unordered_map unless there is a measurable performance advantage. Use std::unordered_map only for use cases in which the map does not need to be iterated or the iteration order does not affect the results of the operation (because different implementations of the hash function may lead to different iteration orders on different systems).

Keep in mind that C++ now allows several methods to insert values into maps, and the behavior can be different when a value already exists for a key. If the intended behavior is that the insertion should not overwrite an existing value for the key, try_emplace() can be a good choice. If the intention is to allow the overwriting of a key/value pair, insert_or_assign() can be a good choice. Both of the above methods provide return values that can be checked– in the case of try_emplace(), whether the insertion succeeded or did not occur, and in the case of insert_or_assign(), whether an insertion or assignment occurred.

6.4.27. Miscellaneous items

• NS_LOG_COMPONENT_DEFINE("log-component-name"); statements should be placed within namespace ns3 (for module code) and after the using namespace ns3;. In examples, NS_OBJECT_ENSURE_REGISTERED() should also be placed within namespace ns3.

• Pointers and references are left-aligned:

  int x = 1;
  int* ptr = &x;
  int& ref = x;
  

• Use a trailing comma in braced-init-lists, so that each item is positioned in a new line.

  const std::vector<std::string> myVector{
    "string-1",
    "string-2",
    "string-3",
  };
  
  const std::map<int, std::string> myMap{
    {1, "string-1"},
    {2, "string-2"},
    {3, "string-3"},
  };
  

• Const reference syntax:

  void MySub(const T& t);  // OK
  void MySub(T const& t);  // Not OK
  

• Do not use NULL, nil or 0 constants; use nullptr (improves portability)

• Consider whether you want the default constructor, copy constructor, or assignment operator in your class. If not, explicitly mark them as deleted and make the declaration public:

  class MyClass
  {
    public:
      // Allowed constructors
      MyClass(int i);
  
      // Deleted constructors.
      // Explain why they are not supported.
      MyClass() = delete;
      MyClass(const MyClass&) = delete;
      MyClass& operator=(const MyClass&) = delete;
  };
  

• Avoid returning a reference to an internal or local member of an object:

  MyType& foo();        // Avoid. Prefer to return a pointer or an object.
  const MyType& foo();  // Same as above.
  

  This guidance does not apply to the use of references to implement operators.

• Expose class members through access functions, rather than direct access to a public object. The access functions are typically named Get and Set followed by the member’s name. For example, a member m_delayTime might have accessor functions GetDelayTime() and SetDelayTime().

• Do not bring the C++ standard library namespace into ns-3 source files by using the using namespace std; directive.

• Do not use the C++ goto statement.

• Do not add the enum or struct specifiers when declaring the variable’s type.

• Do not unnecessarily add typedef to struct or enum.

  // Avoid
  typedef struct
  {
      ...
  } MyStruct;
  
  // Prefer
  struct MyStruct
  {
      ...
  };
  

• When checking whether a Time value is zero, use Time::IsZero() rather than comparing it to a zero-valued time object with operator==, to avoid construction of a temporary. Similar guidance applies to the related functions Time::IsPositive(), Time::IsNegative(), Time::IsStrictlyPositive, and Time::IsStrictlyNegative().

  Time t = ...;
  // prefer the below:
  if (t.IsStrictlyPositive())
  {...}
  // to this alternative:
  if (t > Seconds(0))
  {...}
  

6.4.28. Clang-tidy rules

Please refer to the .clang-tidy file in the ns-3 main directory for the full list of rules that should be observed while developing code.

Some rules are explained in the corresponding sections above. The remaining rules are explained here.

• Explicitly mark inherited functions with the override specifier.

• When creating STL smart pointers, prefer to use std::make_shared or std::make_unique, instead of creating the smart pointer with new.

  auto node = std::make_shared<Node>();           // OK
  auto node = std::shared_ptr<Node>(new Node());  // Avoid
  

• When looping through containers, prefer to use range-based for loops rather than index-based loops.

  std::vector<int> myVector{1, 2, 3};
  
  for (const auto& v : myVector) { ... }             // Prefer
  for (int i = 0; i < myVector.size(); i++) { ... }  // Avoid
  

• Avoid accessing class static functions and members through objects. Instead, prefer to access them through the class.

  // OK
  MyClass::StaticFunction();
  
  // Avoid
  MyClass myClass;
  MyClass.StaticFunction();
  

• Prefer using type traits in short form traits_t<...> and traits_v<...>, instead of the long form traits<...>::type and traits<...>::value, respectively.

  // Prefer using the shorter version of type traits
  std::is_same_v<int, float>
  std::is_integral_v<T>
  std::enable_if_t<std::is_integral_v<T>, Time>
  
  // Avoid the longer form of type traits
  std::is_same<int, float>::value
  std::is_integral<T>::value
  std::enable_if<std::is_integral<T>::value, Time>::type
  

• Avoid using integer values (1 or 0) to represent boolean variables (true or false), to improve code readability and avoid implicit conversions.

• Prefer to use static_assert() over NS_ASSERT() when conditions can be evaluated at compile-time.

• Prefer using transparent functors to non-transparent ones, to avoid repeating the type name. This improves readability and avoids errors when refactoring code.

  // Prefer using transparent functors
  std::map<MyClass, int, std::less<>> myMap;
  
  // Avoid repeating the type name "MyClass" in std::less<>
  std::map<MyClass, int, std::less<MyClass>> myMap;
  

• In conditional control blocks (i.e., if-else and switch-case), avoid declaring multiple branch conditions with the same content to avoid duplicating code.

  In if-else blocks, prefer grouping the identical bodies in a single if condition with a disjunction of the multiple conditions.

  if (condition1)
  {
      Foo();
  }
  else if (condition2)
  {
      // Same body as condition 1
      Foo();
  }
  else
  {
      Bar();
  }
  
  // Prefer grouping the two conditions
  if (condition1 || condition2)
  {
      Foo();
  }
  else
  {
      Bar();
  }
  

  In switch-case blocks, prefer grouping identical case labels by removing the duplicate bodies of the former case labels.

  switch (condition)
  {
  case 1:
      Foo();
      break;
  case 2: // case 2 has the same body as case 1
      Foo();
      break;
  case 3:
      Bar();
      break;
  }
  
  switch (condition)
  {
  // Group identical cases by removing the content of case 1 and letting it fallthrough to case 2
  case 1:
  case 2:
      Foo();
      break;
  case 3:
      Bar();
      break;
  }
  

6.5. CMake file formatting

The CMakeLists.txt and other *.cmake files follow the formatting rules defined in build-support/cmake-format.yaml and build-support/cmake-format-modules.yaml.

The first set of rules applies to CMake files in all directories that are not modules, while the second one applies to files within modules.

Those rules are enforced via the cmake-format tool, that can be installed via Pip.

pip install cmake-format pyyaml

After installing cmake-format, it can be called to fix the formatting of a CMake file with the following command:

cmake-format -c ./build-support/cmake-format.yaml CMakeLists.txt

To check the formatting, add the –check option to the command, before specifying the list of CMake files.

Instead of calling this command for every single CMake file, it is recommended to use the ns3 script to run the custom targets that do that automatically.

# Check CMake formatting
./ns3 build cmake-format-check

# Check and fix CMake formatting
./ns3 build cmake-format

Custom functions and macros need to be explicitly configured in the cmake-format.yaml files, otherwise their formatting will be broken.

6.6. Python file formatting

Python format style and rule enforcement is based on the default settings for the Black formatter tool and Isort import sorter tool. Black default format is detailed in Black current style.

The custom settings for both tools are set in the pyproject.toml file.

These tools that can be installed via Pip, using the following command:

pip install black isort

To check the formatting, add the –check option to the command:

black --check .
isort --check .

To check and fix the formatting, run the commands as follows:

black .
isort .

For VS Code users, MS Black formatter and MS Isort extensions, which repackage Black and Isort for VS Code, can be installed to apply fixes regularly. To configure VS Code to automatically format code when saving, editing or pasting code, add the following configuration to .vscode/settings.json:

{
  "editor.formatOnPaste": true,
  "editor.formatOnSave": true,
  "editor.formatOnType": true,
  "[python]": {
    "editor.defaultFormatter": "ms-python.black-formatter",
    "editor.codeActionsOnSave": {
        "source.organizeImports": "explicit",
    },
  },
  "black-formatter.args": [
    "--config",
    "pyproject.toml",
  ],
  "isort.check": true,
  "isort.args": [
    "--sp",
    "pyproject.toml",
  ],
}

6.7. Markdown Lint

ns-3 uses Markdownlint as a linter of Markdown files. This linter checks if Markdown files follow a set of defined rules, in order to encourage standardization and consistency of Markdown files across parsers. It also ensures that Markdown files are correctly interpreted and rendered.

Markdownlint detects linting issues, but it can not fix them automatically. The issues must be fixed manually.

6.7.1. Markdownlint configuration

Markdownlint’s settings are saved in the file .mdl_style.rb. This file is defined in Markdownlint Configuration Style File, which explains how to customize the tool to enable / disable rules or customize its parameters.

The list of Markdown rules supported by Markdownlint is available in Markdownlint Rules.

6.7.2. Install and Run Markdownlint

Markdownlint is written in Ruby. To run Markdownlint, either use the official Markdownlint Docker image or install Ruby and Markdownlint.

6.7.2.1. Run Markdownlint with Docker image

Markdownlint has an official Docker image in Markdownlint Docker Hub with the tool and all dependencies installed. The instructions to use the Docker image are available in Markdownlint Docker Instructions.

To run Markdownlint in a Docker container, use the following command:

docker run -v .:/data markdownlint/markdownlint -s .mdl_style.rb .

6.7.2.2. Install and Run Markdownlint with Ruby

To install Markdownlint natively, you need to have Ruby installed in your system. Check the installation instructions in the Ruby’s official documentation.

After installing Ruby in your system, install Markdownlint using the following command:

gem install mdl

To run Markdownlint and check Markdown files for linting issues, run Markdownlint using the following command:

mdl -s .mdl_style.rb .

6.7.3. VS Code Extension

For VS Code users, the Markdownlint VS Code Extension extension is available in the marketplace. This extension is inspired in Markdownlint and follows the same set of rules.

The Markdownlint extension automatically analyzes files open in the editor and provides inline hints when issues are detected. It can automatically fix most issues related with formatting.