Style Guide#


Pigweed runs pw format as part of pw presubmit to perform some code formatting checks. To speed up the review process, consider adding pw presubmit as a git push hook using the following command: pw presubmit --install

C++ style#

The Pigweed C++ style guide is closely based on Google’s external C++ Style Guide, which is found on the web at The Google C++ Style Guide applies to Pigweed except as described in this document.

The Pigweed style guide only applies to Pigweed itself. It does not apply to projects that use Pigweed or to the third-party code included with Pigweed. Non-Pigweed code is free to use features restricted by Pigweed, such as dynamic memory allocation and the entirety of the C++ Standard Library.

Recommendations in the Embedded C++ Guide are considered part of the Pigweed style guide, but are separated out since it covers more general embedded development beyond just C++ style.

C++ standard#

All Pigweed C++ code must compile with -std=c++17 in Clang and GCC. C++20 features may be used as long as the code still compiles unmodified with C++17. See pw_polyfill/language_feature_macros.h for macros that provide C++20 features when supported.

Compiler extensions should not be used unless wrapped in a macro or properly guarded in the preprocessor. See pw_processor/compiler.h for macros that wrap compiler-specific features.

Automatic formatting#

Pigweed uses clang-format to automatically format Pigweed source code. A .clang-format configuration is provided with the Pigweed repository.

Automatic formatting is essential to facilitate large-scale, automated changes in Pigweed. Therefore, all code in Pigweed is expected to be formatted with clang-format prior to submission. Existing code may be reformatted at any time.

If clang-format formats code in an undesirable or incorrect way, it can be disabled for the affected lines by adding // clang-format off. clang-format must then be re-enabled with a // clang-format on comment.

// clang-format off
constexpr int kMyMatrix[] = {
    100,  23,   0,
      0, 542,  38,
      1,   2, 201,
// clang-format on

C Standard Library#

In C++ headers, always use the C++ versions of C Standard Library headers (e.g. <cstdlib> instead of <stdlib.h>). If the header is used by both C and C++ code, only the C header should be used.

In C++ code, it is preferred to use C functions from the std namespace. For example, use std::memcpy instead of memcpy. The C++ standard does not require the global namespace versions of the functions to be provided. Using std:: is more consistent with the C++ Standard Library and makes it easier to distinguish Pigweed functions from library functions.

Within core Pigweed, do not use C standard library functions that allocate memory, such as std::malloc. There are exceptions to this for when dynamic allocation is enabled for a system; Pigweed modules are allowed to add extra functionality when a heap is present; but this must be optional.

C++ Standard Library#

Much of the C++ Standard Library is not a good fit for embedded software. Many of the classes and functions were not designed with the RAM, flash, and performance constraints of a microcontroller in mind. For example, simply adding the line #include <iostream> can increase the binary size by 150 KB! This is larger than many microcontrollers’ entire internal storage.

However, with appropriate caution, a limited set of standard C++ libraries can be used to great effect. Developers can leverage familiar, well-tested abstractions instead of writing their own. C++ library algorithms and classes can give equivalent or better performance than hand-written C code.

A limited subset of the C++ Standard Library is permitted in Pigweed. To keep Pigweed small, flexible, and portable, functions that allocate dynamic memory must be avoided. Care must be exercised when using multiple instantiations of a template function, which can lead to code bloat.

Permitted Headers#

The following C++ Standard Library headers are always permitted:

  • <array>

  • <complex>

  • <initializer_list>

  • <iterator>

  • <limits>

  • <optional>

  • <random>

  • <ratio>

  • <string_view>

  • <tuple>

  • <type_traits>

  • <utility>

  • <variant>

  • C Standard Library headers (<c*>)

With caution, parts of the following headers can be used:

  • <algorithm> – be wary of potential memory allocation

  • <atomic> – not all MCUs natively support atomic operations

  • <bitset> – conversions to or from strings are disallowed

  • <functional> – do not use std::function; use pw_function

  • <mutex> – can use std::lock_guard, use pw_sync for mutexes

  • <new> – for placement new

  • <numeric> – be wary of code size with multiple template instantiations

Never use any of these headers:

  • Dynamic containers (<list>, <map>, <set>, <vector>, etc.)

  • Streams (<iostream>, <ostream>, <fstream>, <sstream> etc.) – in some cases pw_stream can work instead

  • <span> – use pw_span instead. Downstream projects may want to directly use std::span if it is available, but upstream must use the pw::span version for compatability

  • <string> – can use pw_string

  • <thread> – can use pw_thread

  • <future> – eventually pw_async will offer this

  • <exception>, <stdexcept> – no exceptions

  • <memory>, <scoped_allocator> – no allocations

  • <regex>

  • <valarray>

Headers not listed here should be carefully evaluated before they are used.

These restrictions do not apply to third party code or to projects that use Pigweed.

Combining C and C++#

Prefer to write C++ code over C code, using extern "C" for symbols that must have C linkage. extern "C" functions should be defined within C++ namespaces to simplify referring to other code.

C++ functions with no parameters do not include void in the parameter list. C functions with no parameters must include void.

namespace pw {

bool ThisIsACppFunction() { return true; }

extern "C" int pw_ThisIsACFunction(void) { return -1; }

extern "C" {

int pw_ThisIsAlsoACFunction(void) {
  return ThisIsACppFunction() ? 100 : 0;

}  // extern "C"

}  // namespace pw


Prefer C++-style (//) comments over C-style comments (/* */). C-style comments should only be used for inline comments.

// Use C++-style comments, except where C-style comments are necessary.
// This returns a random number using an algorithm I found on the internet.
#define RANDOM_NUMBER() [] {                \
  return 4;  /* chosen by fair dice roll */ \

Indent code in comments with two additional spaces, making a total of three spaces after the //. All code blocks must begin and end with an empty comment line, even if the blank comment line is the last line in the block.

// Here is an example of code in comments.
//   int indentation_spaces = 2;
//   int total_spaces = 3;
//   engine_1.thrust = RANDOM_NUMBER() * indentation_spaces + total_spaces;
bool SomeFunction();

Control statements#

Loops and conditionals#

All loops and conditional statements must use braces, and be on their own line.

Yes: Always use braces for line conditionals and loops:

while (SomeCondition()) {
  x += 2;
if (OtherCondition()) {

No: Missing braces

while (SomeCondition())
  x += 2;
if (OtherCondition())

No: Statement on same line as condition

while (SomeCondition()) { x += 2; }
if (OtherCondition()) { DoTheThing(); }

The syntax while (true) is preferred over for (;;) for infinite loops.


while (true) {


for (;;) {

Prefer early exit with return and continue#

Prefer to exit early from functions and loops to simplify code. This is the same same conventions as LLVM. We find this approach is superior to the “one return per function” style for a multitude of reasons:

  • Visually, the code is easier to follow, and takes less horizontal screen space.

  • It makes it clear what part of the code is the “main business” versus “edge case handling”.

  • For functions, parameter checking is in its own section at the top of the function, rather than scattered around in the fuction body.

  • For loops, element checking is in its own section at the top of the loop, rather than scattered around in the loop body.

  • Commit deltas are simpler to follow in code reviews; since adding a new parameter check or loop element condition doesn’t cause an indentation change in the rest of the function.

The guidance applies in two cases:

  • Function early exit - Early exits are for function parameter checking and edge case checking at the top. The main functionality follows.

  • Loop early exit - Early exits in loops are for skipping an iteration due to some edge case with an item getting iterated over. Loops may also contain function exits, which should be structured the same way (see example below).

Yes: Exit early from functions; keeping the main handling at the bottom and de-dentend.

Status DoSomething(Parameter parameter) {
  // Parameter validation first; detecting incoming use errors.
  PW_CHECK_INT_EQ(, 3, "Programmer error: frobnitz");

  // Error case: Not in correct state.
  if (parameter.other() == MyEnum::kBrokenState) {
    LOG_ERROR("Device in strange state: %s", parametr.state_str());
    return Status::InvalidPrecondition();

  // Error case: Not in low power mode; shouldn't do anything.
  if (parameter.power() != MyEnum::kLowPower) {
    LOG_ERROR("Not in low power mode");
    return Status::InvalidPrecondition();

  // Main business for the function here.

No: Main body of function is buried and right creeping. Even though this is shorter than the version preferred by Pigweed due to factoring the return statement, the logical structure is less obvious. A function in Pigweed containing nested conditionals indicates that something complicated is happening with the flow; otherwise it would have the early bail structure; so pay close attention.

Status DoSomething(Parameter parameter) {
  // Parameter validation first; detecting incoming use errors.
  PW_CHECK_INT_EQ(, 3, "Programmer error: frobnitz");

  // Error case: Not in correct state.
  if (parameter.other() != MyEnum::kBrokenState) {
    // Error case: Not in low power mode; shouldn't do anything.
    if (parameter.power() == MyEnum::kLowPower) {
      // Main business for the function here.
    } else {
      LOG_ERROR("Not in low power mode");
  } else {
    LOG_ERROR("Device in strange state: %s", parametr.state_str());
  return Status::InvalidPrecondition();

Yes: Bail early from loops; keeping the main handling at the bottom and de-dentend.

for (int i = 0; i < LoopSize(); ++i) {
  // Early skip of item based on edge condition.
  if (!CommonCase()) {
  // Early exit of function based on error case.
  int my_measurement = GetSomeMeasurement();
  if (my_measurement < 10) {
    LOG_ERROR("Found something strange; bailing");
    return Status::Unknown();

  // Main body of the loop.
  ProcessItem(my_items[i], my_measurement);
  ProcessItemMore(my_items[i], my_measurement, other_details);

No: Right-creeping code with the main body buried inside some nested conditional. This makes it harder to understand what is the main purpose of the loop versus what is edge case handling.

for (int i = 0; i < LoopSize(); ++i) {
  if (CommonCase()) {
    int my_measurement = GetSomeMeasurement();
    if (my_measurement >= 10) {
      // Main body of the loop.
      ProcessItem(my_items[i], my_measurement);
      ProcessItemMore(my_items[i], my_measurement, other_details);
    } else {
      LOG_ERROR("Found something strange; bailing");
      return Status::Unknown();

There are cases where this structure doesn’t work, and in those cases, it is fine to structure the code differently.

No else after return or continue#

Do not put unnecessary } else { blocks after blocks that terminate with a return, since this causes unnecessary rightward indentation creep. This guidance pairs with the preference for early exits to reduce code duplication and standardize loop/function structure.

Yes: No else after return or continue

// Note lack of else block due to return.
if (Failure()) {
  return Status::ResourceExausted();

// Note lack of else block due to continue.
while (MyCondition()) {
  if (SomeEarlyBail()) {
  // Main handling of item

return OkStatus();

No: Else after return needlessly creeps right

if (Failure()) {
  return Status::ResourceExausted();
} else {
  while (MyCondition()) {
    if (SomeEarlyBail()) {
    } else {
      // Main handling of item
  return OkStatus();

Include guards#

The first non-comment line of every header file must be #pragma once. Do not use traditional macro include guards. The #pragma once should come directly after the Pigweed copyright block, with no blank line, followed by a blank, like this:

// Copyright 2021 The Pigweed Authors
// Licensed under the Apache License, Version 2.0 (the "License"); you may not
// use this file except in compliance with the License. You may obtain a copy of
// the License at
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
// WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
// License for the specific language governing permissions and limitations under
// the License.
#pragma once

// Header file-level comment goes here...

Memory allocation#

Dynamic memory allocation can be problematic. Heap allocations and deallocations occupy valuable CPU cycles. Memory usage becomes nondeterministic, which can result in a system crashing without a clear culprit.

To keep Pigweed portable, core Pigweed code is not permitted to dynamically (heap) allocate memory, such as with malloc or new. All memory should be allocated with automatic (stack) or static (global) storage duration. Pigweed must not use C++ libraries that use dynamic allocation.

Projects that use Pigweed are free to use dynamic allocation, provided they have selected a target that enables the heap.


Entities shall be named according to the Google style guide, with the following additional requirements.

C++ code#

  • All Pigweed C++ code must be in the pw namespace. Namespaces for modules should be nested under pw. For example, pw::string::Format().

  • Whenever possible, private code should be in a source (.cc) file and placed in anonymous namespace nested under pw.

  • If private code must be exposed in a header file, it must be in a namespace nested under pw. The namespace may be named for its subsystem or use a name that designates it as private, such as internal.

  • Template arguments for non-type names (e.g. template <int kFooBar>) should follow the constexpr and const variable Google naming convention, which means k prefixed camel case (e.g. kCamelCase). This matches the Google C++ style for variable naming, however the wording in the official style guide isn’t explicit for template arguments and could be interpreted to use foo_bar style naming. For consistency with other variables whose value is always fixed for the duration of the program, the naming convention is kCamelCase, and so that is the style we use in Pigweed.

  • Trivial membor accessors should be named with snake_case(). The Google C++ style allows either snake_case() or CapsCase(), but Pigweed always uses snake_case().

  • Abstract base classes should be named generically, with derived types named specifically. For example, Stream is an abstract base, and SocketStream and StdioStream are an implementations of that interface. Any prefix or postfix indicating whether something is abstract or concrete is not permitted; for example, IStream or SocketStreamImpl are both not permitted. These pre-/post-fixes add additional visual noise and are irrelevant to consumers of these interfaces.

C code#

In general, C symbols should be prefixed with the module name. If the symbol is not associated with a module, use just pw as the module name. Facade backends may chose to prefix symbols with the facade’s name to help reduce the length of the prefix.

  • Public names used by C code must be prefixed with the module name (e.g. pw_tokenizer_*).

  • If private code must be exposed in a header, private names used by C code must be prefixed with an underscore followed by the module name (e.g. _pw_assert_*).

  • Avoid writing C source (.c) files in Pigweed. Prefer to write C++ code with C linkage using extern "C". Within C source, private C functions and variables must be named with the _pw_my_module_* prefix and should be declared static whenever possible; for example, _pw_my_module_MyPrivateFunction.

  • The C prefix rules apply to

    • C functions (int pw_foo_FunctionName(void);),

    • variables used by C code (int pw_foo_variable_name;),

    • constant variables used by C code (const int pw_foo_kConstantName;),

    • structs used by C code (typedef struct {} pw_foo_StructName;), and

    • all of the above for extern "C" names in C++ code.

    The prefix does not apply to struct members, which use normal Google style.

Preprocessor macros#

  • Public Pigweed macros must be prefixed with the module name (e.g. PW_MY_MODULE_*).

  • Private Pigweed macros must be prefixed with an underscore followed by the module name (e.g. _PW_MY_MODULE_*). (This style may change, see b/234886184).


namespace pw::my_module {
namespace nested_namespace {

// C++ names (types, variables, functions) must be in the pw namespace.
// They are named according to the Google style guide.
constexpr int kGlobalConstant = 123;

// Prefer using functions over extern global variables.
extern int global_variable;

class Class {};

void Function();

extern "C" {

// Public Pigweed code used from C must be prefixed with pw_.
extern const int pw_my_module_kGlobalConstant;

extern int pw_my_module_global_variable;

void pw_my_module_Function(void);

typedef struct {
  int member_variable;
} pw_my_module_Struct;

// Private Pigweed code used from C must be prefixed with _pw_.
extern const int _pw_my_module_kPrivateGlobalConstant;

extern int _pw_my_module_private_global_variable;

void _pw_my_module_PrivateFunction(void);

typedef struct {
  int member_variable;
} _pw_my_module_PrivateStruct;

}  // extern "C"

// Public macros must be prefixed with PW_.
#define PW_MY_MODULE_PUBLIC_MACRO(arg) arg

// Private macros must be prefixed with _PW_.
#define _PW_MY_MODULE_PRIVATE_MACRO(arg) arg

}  // namespace nested_namespace
}  // namespace pw::my_module

See Preprocessor macros for details about macro usage.

Namespace scope formatting#

All non-indented blocks (namespaces, extern "C" blocks, and preprocessor conditionals) must have a comment on their closing line with the contents of the starting line.

All nested namespaces should be declared together with no blank lines between them.

#include "some/header.h"

namespace pw::nested {
namespace {

constexpr int kAnonConstantGoesHere = 0;

}  // namespace

namespace other {

const char* SomeClass::yes = "no";

bool ThisIsAFunction() {
  return true;
  return false;
#endif  // PW_CONFIG_IS_SET

extern "C" {

const int pw_kSomeConstant = 10;
int pw_some_global_variable = 600;

void pw_CFunction() { ... }

}  // extern "C"

}  // namespace
}  // namespace pw::nested

Using directives for literals#

Using-directives (e.g. using namespace ...) are permitted in implementation files only for the purposes of importing literals such as std::chrono_literals or pw::bytes::unit_literals. Namespaces that contain any symbols other than literals are not permitted in a using-directive. This guidance also has no impact on using-declarations (e.g. using foo::Bar;).

Rationale: Literals improve code readability, making units clearer at the point of definition.

using namespace std::chrono;                    // Not allowed
using namespace std::literals::chrono_literals; // Allowed

constexpr std::chrono::duration delay = 250ms;

Pointers and references#

For pointer and reference types, place the asterisk or ampersand next to the type.

int* const number = &that_thing;
constexpr const char* kString = "theory!"

bool FindTheOneRing(const Region& where_to_look) { ... }

Prefer storing references over storing pointers. Pointers are required when the pointer can change its target or may be nullptr. Otherwise, a reference or const reference should be used.

Preprocessor macros#

Macros should only be used when they significantly improve upon the C++ code they replace. Macros should make code more readable, robust, and safe, or provide features not possible with standard C++, such as stringification, line number capturing, or conditional compilation. When possible, use C++ constructs like constexpr variables in place of macros. Never use macros as constants, except when a string literal is needed or the value must be used by C code.

When macros are needed, the macros should be accompanied with extensive tests to ensure the macros are hard to use wrong.

Stand-alone statement macros#

Macros that are standalone statements must require the caller to terminate the macro invocation with a semicolon (see Swalling the Semicolon). For example, the following does not conform to Pigweed’s macro style:

// BAD! Definition has built-in semicolon.
#define PW_LOG_IF_BAD(mj) \

// BAD! Compiles without error; semicolon is missing.

Here’s how to do this instead:

// GOOD; requires semicolon to compile.
#define PW_LOG_IF_BAD(mj) \

// GOOD; fails to compile due to lacking semicolon.

For macros in function scope that do not already require a semicolon, the contents can be placed in a do { ... } while (0) loop.

#define PW_LOG_IF_BAD(mj)  \
  do {                     \
    if (mj.Bad()) {        \
      Log(#mj " is bad")   \
    }                      \
  } while (0)

Standalone macros at global scope that do not already require a semicolon can add a static_assert declaration statement as their last line.

#define PW_NEAT_THING(thing)             \
  bool IsNeat_##thing() { return true; } \
  static_assert(true, "Macros must be terminated with a semicolon")

Private macros in public headers#

Private macros in public headers must be prefixed with _PW_, even if they are undefined after use; this prevents collisions with downstream users. For example:

#define _PW_MY_SPECIAL_MACRO(op) ...
// Code that uses _PW_MY_SPECIAL_MACRO()

Macros in private implementation files (.cc)#

Macros within .cc files that should only be used within one file should be undefined after their last use; for example:

#define DEFINE_OPERATOR(op) \
  T operator ## op(T x, T y) { return x op y; } \
  static_assert(true, "Macros must be terminated with a semicolon") \



Preprocessor conditional statements#

When using macros for conditional compilation, prefer to use #if over #ifdef. This checks the value of the macro rather than whether it exists.

  • #if handles undefined macros equivalently to #ifdef. Undefined macros expand to 0 in preprocessor conditional statements.

  • #if evaluates false for macros defined as 0, while #ifdef evaluates true.

  • Macros defined using compiler flags have a default value of 1 in GCC and Clang, so they work equivalently for #if and #ifdef.

  • Macros defined to an empty statement cause compile-time errors in #if statements, which avoids ambiguity about how the macro should be used.

All #endif statements should be commented with the expression from their corresponding #if. Do not indent within preprocessor conditional statements.

using Word = uint64_t;
using Word = uint32_t;
#endif  // USE_64_BIT_WORD

Unsigned integers#

Unsigned integers are permitted in Pigweed. Aim for consistency with existing code and the C++ Standard Library. Be very careful mixing signed and unsigned integers.

Features not in the C++ standard#

Avoid features not available in standard C++. This includes compiler extensions and features from other standards like POSIX.

For example, use ptrdiff_t instead of POSIX’s ssize_t, unless interacting with a POSIX API in intentionally non-portable code. Never use POSIX functions with suitable standard or Pigweed alternatives, such as strnlen (use pw::string::NullTerminatedLength instead).

Code Owners (OWNERS)#

If you use Gerrit for source code hosting and have the code-owners plug-in enabled Pigweed can help you enforce consistent styling on those files and also detects some errors.

The styling follows these rules.

  1. Content is grouped by type of line (Access grant, include, etc).

  2. Each grouping is sorted alphabetically.

  3. Groups are placed the following order with a blank line separating each grouping.

    • “set noparent” line

    • “include” lines

    • “file:” lines

    • user grants (some examples: “*”, “”)

    • “per-file:” lines

This plugin will, by default, act upon any file named “OWNERS”.

Python style#

Pigweed uses the standard Python style: PEP8, which is available on the web at All Pigweed Python code should pass pw format, which invokes black with a couple options.

Python versions#

Pigweed officially supports a few Python versions. Upstream Pigweed code must support those Python versions. The only exception is pw_env_setup, which must also support Python 2 and 3.6.

Build files: GN#

Each Pigweed source module requires a GN build file named This encapsulates the build targets and specifies their sources and dependencies. GN build files use a format similar to Bazel’s BUILD files (see the Bazel style guide).

C/C++ build targets include a list of fields. The primary fields are:

  • <public> – public header files

  • <sources> – source files and private header files

  • <public_configs> – public build configuration

  • <configs> – private build configuration

  • <public_deps> – public dependencies

  • <deps> – private dependencies

Assets within each field must be listed in alphabetical order.

# Here is a brief example of a GN build file.


config("public_include_path") {
  include_dirs = [ "public" ]
  visibility = [":*"]

pw_source_set("pw_sample_module") {
  public = [ "public/pw_sample_module/sample_module.h" ]
  sources = [
  public_configs = [ ":public_include_path" ]
  public_deps = [ dir_pw_status ]
  deps = [ dir_pw_varint ]

pw_test_group("tests") {
  tests = [ ":sample_module_test" ]

pw_test("sample_module_test") {
  sources = [ "" ]
  deps = [ ":sample_module" ]

pw_doc_group("docs") {
  sources = [ "docs.rst" ]

Build files: Bazel#

Build files for the Bazel build system must be named BUILD.bazel. Bazel can interpret files named just BUILD, but Pigweed uses BUILD.bazel to avoid ambiguity with other build systems or tooling.

Pigweed’s Bazel files follow the Bazel style guide.



Pigweed’s documentation style guide came after much of the documentation was written, so Pigweed’s docs don’t yet 100% conform to this style guide. When updating docs, please update them to match the style guide.

Pigweed documentation is written using the reStructuredText markup language and processed by Sphinx. We use the Furo theme along with the sphinx-design extension.


Use headings according to the following hierarchy, with the shown characters for the ReST heading syntax.

Document Title: Two Bars of Equals
Document titles use equals ("====="), above and below. Capitalize the words
in the title, except for 'a', 'of', and 'the'.

Major Sections Within a Doc
Major sections use hyphens ("----"), above and below. Capitalize the words in
the title, except for 'a', 'of', and 'the'.

Heading 1 - For Sections Within a Doc
These should be title cased. Use a single equals bar ("====").

Heading 2 - for subsections
Subsections use hyphens ("----"). In many cases, these headings may be
sentence-like. In those cases, only the first letter should be capitalized.
For example, FAQ subsections would have a title with "Why does the X do the
Y?"; note the sentence capitalization (but not title capitalization).

Heading 3 - for subsubsections
Use the caret symbol ("^^^^") for subsubsections.

Note: Generally don't go beyond heading 3.

Heading 4 - for subsubsubsections
Don't use this heading level, but if you must, use period characters
("....") for the heading.

Do not put blank lines after headings.#

Yes: No blank after heading

Here is a heading
Note that there is no blank line after the heading separator!

No: Unnecessary blank line

Here is a heading

There is a totally unnecessary blank line above this one. Don't do this.

Do not put multiple blank lines before a heading.#

Yes: Just one blank after section content before the next heading

There is some text here in the section before the next. It's just here to
illustrate the spacing standard. Note that there is just one blank line
after this paragraph.

Just one blank!
There is just one blank line before the heading.

No: Extra blank lines

There is some text here in the section before the next. It's just here to
illustrate the spacing standard. Note that there are too many blank lines
after this paragraph; there should be just one.

Too many blanks
There are too many blanks before the heading for this section.


Indent directives 3 spaces; and put a blank line between the directive and the content. This aligns the directive content with the directive name.

Yes: Three space indent for directives; and nested

Here is a paragraph that has some content. After this content is a

.. my_directive::

   Note that this line's start aligns with the "m" above. The 3-space
   alignment accounts for the ".. " prefix for directives, to vertically
   align the directive name with the content.

   This indentation must continue for nested directives.

   .. nested_directive::

      Here is some nested directive content.

No: One space, two spaces, four spaces, or other indents for directives

Here is a paragraph with some content.

.. my_directive::

  The indentation here is incorrect! It's one space short; doesn't align
  with the directive name above.

  .. nested_directive::

      This isn't indented correctly either; it's too much (4 spaces).

No: Missing blank between directive and content.

Here is a paragraph with some content.

.. my_directive::
   Note the lack of blank line above here.


Consider using .. list-table:: syntax, which is more maintainable and easier to edit for complex tables (details).

Code Snippets#

Use code blocks from actual source code files wherever possible. This helps keep documentation fresh and removes duplicate code examples. There are a few ways to do this with Sphinx.

The literalinclude directive creates a code blocks from source files. Entire files can be included or a just a subsection. The best way to do this is with the :start-after: and :end-before: options.


Documentation Source (.rst file)

.. literalinclude::
   :start-after: [doxygen-environment-variables]
   :end-before: [doxygen-environment-variables]

Source File

# DOCSTAG: [doxygen-environment-variables]
env = os.environ.copy()
env['PW_DOXYGEN_OUTPUT_DIRECTORY'] = str(output_dir.resolve())
env['PW_DOXYGEN_INPUT'] = ' '.join(pw_module_list)
env['PW_DOXYGEN_PROJECT_NAME'] = 'Pigweed'
# DOCSTAG: [doxygen-environment-variables]

Rendered Output

env = os.environ.copy()
env['PW_DOXYGEN_OUTPUT_DIRECTORY'] = str(output_dir.resolve())
env['PW_DOXYGEN_INPUT'] = ' '.join(pw_module_list)
env['PW_DOXYGEN_PROJECT_NAME'] = 'Pigweed'

Generating API documentation from source#

Whenever possible, document APIs in the source code and use Sphinx to generate documentation for them. This keeps the documentation in sync with the code and reduces duplication.


Include Python API documentation from docstrings with autodoc directives. Example:

.. automodule:: pw_cli.log

.. automodule:: pw_console.embed
   :members: PwConsoleEmbed

.. autoclass:: pw_console.log_store.LogStore
   :members: __init__

Include argparse command line help with the argparse directive. Example:

.. argparse::
   :func: get_parser
   :prog: pw watch


Doxygen comments in C, C++, and Java are surfaced in Sphinx using Breathe.


Sources with doxygen comment blocks must be added to the _doxygen_input_files list in //docs/ to be processed.

Breathe provides various directives for bringing Doxygen comments into Sphinx. These include the following:

  • doxygenfile – Documents a source file. May limit to specific types of symbols with :sections:.

    .. doxygenfile:: pw_rpc/internal/config.h
       :sections: define, func
  • doxygenclass – Documents a class and optionally its members.

    .. doxygenclass:: pw::sync::BinarySemaphore
  • doxygentypedef – Documents an alias (typedef or using statement).

    .. doxygentypedef:: pw::Function
  • doxygenfunction – Documents a source file. Can be filtered to limit to specific types of symbols.

    .. doxygenfunction:: pw::tokenizer::EncodeArgs
  • doxygendefine – Documents a preprocessor macro.

    .. doxygendefine:: PW_TOKENIZE_STRING
Example Doxygen Comment Block#

Start a Doxygen comment block using /// (three forward slashes).

/// This is the documentation comment for the `PW_LOCK_RETURNED()` macro. It
/// describes how to use the macro.
/// Doxygen comments can refer to other symbols using Sphinx cross
/// references. For example, @cpp_class{pw::InlineBasicString}, which is
/// shorthand for @crossref{cpp,class,pw::InlineBasicString}, links to a C++
/// class. @crossref{py,func,pw_tokenizer.proto.detokenize_fields} links to a
/// Python function.
/// @param[out] dest The memory area to copy to.
/// @param[in]  src  The memory area to copy from.
/// @param[in]  n    The number of bytes to copy
/// @retval OK         KVS successfully initialized.
/// @retval DATA_LOSS  KVS initialized and is usable, but contains corrupt data.
/// @retval UNKNOWN    Unknown error. KVS is not initialized.
/// @rst
/// The ``@rst`` and ``@endrst`` commands form a block block of
/// reStructuredText that is rendered in Sphinx.
/// .. warning::
///    this is a warning admonition
/// .. code-block:: cpp
///    void release(ptrdiff_t update = 1);
/// @endrst
/// Example code block using Doxygen markup below. To override the language
/// use `@code{.cpp}`
/// @code
///   class Foo {
///    public:
///     Mutex* mu() PW_LOCK_RETURNED(mu) { return &mu; }
///    private:
///     Mutex mu;
///   };
/// @endcode
/// @b The first word in this sentence is bold (The).
#define PW_LOCK_RETURNED(x) __attribute__((lock_returned(x)))
Doxygen Syntax#

Pigweed prefers to use RST wherever possible, but there are a few Doxygen syntatic elements that may be needed.

Common Doxygen commands for use within a comment block:

  • @rst To start a reStructuredText block. This is a custom alias for \verbatim embed:rst:leading-asterisk.

  • @code Start a code block.

  • @param Document parameters, this may be repeated.

  • @pre Starts a paragraph where the precondition of an entity can be described.

  • @post Starts a paragraph where the postcondition of an entity can be described.

  • @return Single paragraph to describe return value(s).

  • @retval Document return values by name. This is rendered as a definition list.

  • @note Add a note admonition to the end of documentation.

  • @b To bold one word.

Doxygen provides structural commands that associate a comment block with a particular symbol. Example of these include @class, @struct, @def, @fn, and @file. Do not use these unless necessary, since they are redundant with the declarations themselves.

One case where structural commands are necessary is when a single comment block describes multiple symbols. To group multiple symbols into a single comment block, include a structural commands for each symbol on its own line. For example, the following comment documents two macros:

/// Documents functions that dynamically check to see if a lock is held, and
/// fail if it is not held.

Pigweed provides Doxygen aliases for creating Sphinx cross references within Doxygen comments. These should be used when referring to other symbols, such as functions, classes, or macros.

The basic alias is @crossref, which supports any Sphinx domain. For readability, aliases for commonnly used types are provided.

  • @crossref{domain,type,identifier} Inserts a cross reference of any type in any Sphinx domain. For example, @crossref{c,func,foo} is equivalent to :c:func:`foo` and links to the documentation for the C function foo, if it exists.

  • @c_macro{identifier} Equivalent to :c:macro:`identifier`.

  • @cpp_func{identifier} Equivalent to :cpp:func:`identifier`.

  • @cpp_class{identifier} Equivalent to :cpp:class:`identifier`.

  • @cpp_type{identifier} Equivalent to :cpp:type:`identifier`.


Use the @ aliases described above for all cross references. Doxygen provides other methods for cross references, but Sphinx cross references offer several advantages:

  • Sphinx cross references work for all identifiers known to Sphinx, including those documented with e.g. .. cpp:class:: or extracted from Python. Doxygen references can only refer to identifiers known to Doxygen.

  • Sphinx cross references always use consistent formatting. Doxygen cross references sometimes render as plain text instead of code-style monospace, or include () in macros that shouldn’t have them.

  • Sphinx cross references can refer to symbols that have not yet been documented. They will be formatted correctly and become links once the symbols are documented.

  • Using Sphinx cross references in Doxygen comments makes cross reference syntax more consistent within Doxygen comments and between RST and Doxygen.

Create cross reference links elsewhere in the document to symbols documented with Doxygen using standard Sphinx cross references, such as :cpp:class: and :cpp:func:.

- :cpp:class:`pw::sync::BinarySemaphore::BinarySemaphore`
- :cpp:func:`pw::sync::BinarySemaphore::try_acquire`

Commit message#

Pigweed commit message bodies and summaries are limited to 72 characters wide to improve readability. Commit summaries should also be prefixed with the name of the module that the commit is affecting. Examples of well and ill-formed commit messages are provided below.

Consider the following when writing a commit message:

  1. Documentation and comments are better - Consider whether the commit message contents would be better expressed in the documentation or code comments. Docs and code comments are durable and readable later; commit messages are rarely read after the change lands.

  2. Include why the change is made, not just what the change is - It is important to include a “why” component in most commits. Sometimes, why is evident - for example, reducing memory usage, or optimizing. But it is often not. Err on the side of over-explaining why, not under-explaining why.

Pigweed commit messages should conform to the following style:


pw_some_module: Short capitalized description

Details about the change here. Include a summary of the what, and a clear
description of why the change is needed for future maintainers.

Consider what parts of the commit message are better suited for

Yes: Small number of modules affected; use {} syntax.

pw_{foo, bar, baz}: Change something in a few places

When changes cross a few modules, include them with the syntax shown

Yes: Targets are effectively modules, even though they’re nested, so they get a / character.

targets/xyz123: Tweak support for XYZ's PQR

Yes: Uses imperative style for subject and text.

pw_something: Add foo and bar functions

This commit correctly uses imperative present-tense style.

No: Uses non-imperative style for subject and text.

pw_something: Adds more things

Use present tense imperative style for subjects and commit. The above
subject has a plural "Adds" which is incorrect; should be "Add".

Yes: Use bulleted lists when multiple changes are in a single CL. Prefer smaller CLs, but larger CLs are a practical reality.

pw_complicated_module: Pre-work for refactor

Prepare for a bigger refactor by reworking some arguments before the larger
change. This change must land in downstream projects before the refactor to
enable a smooth transition to the new API.

- Add arguments to MyImportantClass::MyFunction
- Update MyImportantClass to handle precondition Y
- Add stub functions to be used during the transition

No: Run on paragraph instead of bulleted list

pw_foo: Many things in a giant BWOT

This CL does A, B, and C. The commit message is a Big Wall Of Text
(BWOT), which we try to discourage in Pigweed. Also changes X and Y,
because Z and Q. Furthermore, in some cases, adds a new Foo (with Bar,
because we want to). Also refactors qux and quz.

No: Doesn’t capitalize the subject

pw_foo: do a thing

Above subject is incorrect, since it is a sentence style subject.

Yes: Doesn’t capitalize the subject when subject’s first word is a lowercase identifier.

pw_foo: std::unique_lock cleanup

This commit message demonstrates the subject when the subject has an
identifier for the first word. In that case, follow the identifier casing
instead of capitalizing.

However, imperative style subjects often have the identifier elsewhere in the subject; for example:

pw_foo: Improve use of std::unique_lock

No: Uses a non-standard [] to indicate module:

[pw_foo]: Do a thing

No: Has a period at the end of the subject

pw_bar: Do something great.

No: Puts extra stuff after the module which isn’t a module.

pw_bar/byte_builder: Add more stuff to builder

Copy-to-clipboard feature on code blocks#

The copy-to-clipboard feature on code blocks is powered by sphinx-copybutton.

sphinx-copybutton recognizes $ as an input prompt and automatically removes it.

There is a workflow for manually removing the copy-to-clipboard button for a particular code block but it has not been implemented yet. See Remove copybuttons using a CSS selector.