Module Structure

The Pigweed module structure is designed to keep as much code as possible for a particular slice of functionality in one place. That means including the code from multiple languages, as well as all the related documentation and tests.

Additionally, the structure is designed to limit the number of places a file could go, so that when reading callsites it is obvious where a header is from. That is where the duplicated <module> occurrences in file paths comes from.

Example module structure

pw_foo/...

  docs.rst   # If there is just 1 docs file, call it docs.rst
  README.md  # All modules must have a short README for gittiles

  BUILD.gn   # GN build required
  BUILD      # Bazel build required

  # C++ public headers; the repeated module name is required
  public/pw_foo/foo.h
  public/pw_foo/baz.h

  # Exposed private headers go under internal/
  public/pw_foo/internal/bar.h
  public/pw_foo/internal/qux.h

  # Public override headers must go in 'public_overrides'
  public_overrides/gtest/gtest.h
  public_overrides/string.h

  # Private headers go into <module>_*/...
  pw_foo_internal/zap.h
  pw_foo_private/zip.h
  pw_foo_secret/alxx.h

  # C++ implementations go in the root
  foo_impl.cc
  foo.cc
  baz.cc
  bar.cc
  zap.cc
  zip.cc
  alxx.cc

  # C++ tests also go in the root
  foo_test.cc
  bar_test.cc
  zip_test.cc

  # Python files go into 'py/<module>/...'
  py/BUILD.gn     # Python packages are declared in GN using pw_python_package
  py/setup.py     # Python files are structured as standard Python packages
  py/foo_test.py  # Tests go in py/ but outside of the Python package
  py/bar_test.py
  py/pw_foo/__init__.py
  py/pw_foo/__main__.py
  py/pw_foo/bar.py
  py/pw_foo/py.typed  # Indicates that this package has type annotations

  # Go files go into 'go/...'
  go/...

  # Examples go in examples/, mixing different languages
  examples/demo.py
  examples/demo.cc
  examples/demo.go
  examples/BUILD.gn
  examples/BUILD

  # Size reports go under size_report/
  size_report/BUILD.gn
  size_report/base.cc
  size_report/use_case_a.cc
  size_report/use_case_b.cc

  # Protobuf definition files go into <module>_protos/...
  pw_foo_protos/foo.proto
  pw_foo_protos/internal/zap.proto

  # Other directories are fine, but should be private.
  data/...
  graphics/...
  collection_of_tests/...
  code_relating_to_subfeature/...

Module name

Pigweed upstream modules are always named with a prefix pw_ to enforce namespacing. Projects using Pigweed that wish to make their own modules can use whatever name they like, but we suggest picking a short prefix to namespace your product (e.g. for an Internet of Toast project, perhaps the prefix could be it_).

C++ module structure

C++ public headers

Located {pw_module_dir}/public/<module>. These are headers that must be exposed due to C++ limitations (i.e. are included from the public interface, but are not intended for public use).

Public headers should take the form:

{pw_module_dir}/public/<module>/*.h

Exposed private headers should take the form:

{pw_module_dir}/public/<module>/internal/*.h

Examples:

pw_foo/...
  public/pw_foo/foo.h
  public/pw_foo/a_header.h
  public/pw_foo/baz.h

For headers that must be exposed due to C++ limitations (i.e. are included from the public interface, but are not intended for use), place the headers in a internal subfolder under the public headers directory; as {pw_module_dir}/public/<module>/internal/*.h. For example:

pw_foo/...
  public/pw_foo/internal/secret.h
  public/pw_foo/internal/business.h

Note

These headers must not override headers from other modules. For that, there is the public_overrides/ directory.

C++ public override headers

Located {pw_module_dir}/public_overrides/<module>. In general, the Pigweed philosophy is to avoid having “things hiding under rocks”, and having header files with the same name that can override each other is considered a rock where surprising things can hide. Additionally, a design goal of the Pigweed module structure is to make it so there is ideally exactly one obvious place to find a header based on an #include.

However, in some cases header overrides are necessary to enable flexibly combining modules. To make this as explicit as possible, headers which override other headers must go in

{pw_module_dir}/public_overrides/...`

For example, the pw_unit_test module provides a header override for gtest/gtest.h. The structure of the module is (omitting some files):

pw_unit_test/...

  public_overrides/gtest
  public_overrides/gtest/gtest.h

  public/pw_unit_test
  public/pw_unit_test/framework.h
  public/pw_unit_test/simple_printing_event_handler.h
  public/pw_unit_test/event_handler.h

Note that the overrides are in a separate directory public_overrides.

C++ implementation files

Located {pw_module_dir}/. C++ implementation files go at the top level of the module. Implementation files must always use “” style includes.

Example:

pw_unit_test/...
  main.cc
  framework.cc
  test.gni
  BUILD.gn
  README.md

Compile-time configuration

Pigweed modules are intended to be used in a wide variety of environments. In support of this, some modules expose compile-time configuration options. Pigweed has an established pattern for declaring and overriding module configuration.

Tip

Compile-time configuration provides flexibility, but also imposes restrictions. A module can only have one configuration in a given build. This makes testing modules with compile-time configuration more difficult. Where appropriate, consider alternatives such as C++ templates or runtime configuration.

Declaring configuration

Configuration options are declared in a header file as macros. If the macro is not already defined, a default definition is provided. Otherwise, nothing is done. Configuration headers may include static_assert statements to validate configuration values.

// Example configuration header

#ifndef PW_FOO_INPUT_BUFFER_SIZE_BYTES
#define PW_FOO_INPUT_BUFFER_SIZE_BYTES 128
#endif  // PW_FOO_INPUT_BUFFER_SIZE_BYTES

static_assert(PW_FOO_INPUT_BUFFER_SIZE_BYTES >= 64);

The configuration header may go in one of three places in the module, depending on whether the header should be exposed by the module or not.

pw_foo/...

  # Publicly accessible configuration header
  public/pw_foo/config.h

  # Internal configuration header that is included by other module headers
  public/pw_foo/internal/config.h

  # Internal configuration header
  pw_foo_private/config.h

The configuration header is provided by a build system library. This library acts as a facade. The facade uses a variable such as pw_foo_CONFIG. In upstream Pigweed, all config facades default to the pw_build_DEFAULT_MODULE_CONFIG backend. In the GN build system, the config facade is declared as follows:

declare_args() {
  # The build target that overrides the default configuration options for this
  # module. This should point to a source set that provides defines through a
  # public config (which may -include a file or add defines directly).
  pw_foo_CONFIG = pw_build_DEFAULT_MODULE_CONFIG
}

# An example source set for each potential config header location follows.

# Publicly accessible configuration header (most common)
pw_source_set("config") {
  public = [ "public/pw_foo/config.h" ]
  public_configs = [ ":public_include_path" ]
  public_deps = [ pw_foo_CONFIG ]
}

# Internal configuration header that is included by other module headers
pw_source_set("config") {
  sources = [ "public/pw_foo/internal/config.h" ]
  public_configs = [ ":public_include_path" ]
  public_deps = [ pw_foo_CONFIG ]
  visibility = [":*"]  # Only allow this module to depend on ":config"
  friend = [":*"]  # Allow this module to access the config.h header.
}

# Internal configuration header
pw_source_set("config") {
  public = [ "pw_foo_private/config.h" ]
  public_deps = [ pw_foo_CONFIG ]
  visibility = [":*"]  # Only allow this module to depend on ":config"
}

Overriding configuration

As noted above, all module configuration facades default to the same backend (pw_build_DEFAULT_MODULE_CONFIG). This allows projects to override configuration values for multiple modules from a single configuration backend, if desired. The configuration values may also be overridden individually by setting backends for the individual module configurations (e.g. in GN, pw_foo_CONFIG = "//configuration:my_foo_config").

Configurations options are overridden by setting macros in the config backend. These macro definitions can be provided through compilation options, such as -DPW_FOO_INPUT_BUFFER_SIZE_BYTES=256. Configuration macro definitions may also be set in a header file. The header file is included using the -include compilation option.

This example shows two ways to configure a module in the GN build system.

# In the toolchain, set either pw_build_DEFAULT_MODULE_CONFIG or pw_foo_CONFIG
pw_build_DEFAULT_MODULE_CONFIG = get_path_info(":define_overrides", "abspath")

# This configuration sets PW_FOO_INPUT_BUFFER_SIZE_BYTES using the -D flag.
pw_source_set("define_overrides") {
  public_configs = [ ":define_options" ]
}

config("define_options") {
  defines = [ "-DPW_FOO_INPUT_BUFFER_SIZE_BYTES=256" ]
}

# This configuration sets PW_FOO_INPUT_BUFFER_SIZE_BYTES in a header file.
pw_source_set("include_overrides") {
  public_configs = [ ":set_options_in_header_file" ]

  # Header file with #define PW_FOO_INPUT_BUFFER_SIZE_BYTES 256
  sources = [ "my_config_overrides.h" ]
}

config("set_options_in_header_file") {
  cflags = [
    "-include",
    rebase_path("my_config_overrides.h"),
  ]
}

Why this config pattern is preferred

Alternate patterns for configuring a module include overriding the module’s config header or having that header optionally include a header at a known path (e.g. pw_foo/config_overrides.h). There are a few downsides to these approaches:

  • The module needs its own config header that defines, provides defaults for, and validates the configuration options. Replacing this header with a user-defined header would require defining all options in the user’s header, which is cumbersome and brittle, and would bypass validation in the module’s header.

  • Including a config override header at a particular path would prevent multiple modules from sharing the same configuration file. Multiple headers could redirect to the same configuration file, but this would still require creating a separate header and setting the config backend variable for each module.

  • Optionally including a config override header requires boilerplate code that would have to be duplicated in every configurable module.

  • An optional config override header file would silently be excluded if the file path were accidentally misspelled.

Python module structure

Python code is structured as described in the Module structure section of Pigweed’s Python build.

Facades

In Pigweed, facades represent a dependency that can be swapped at compile time. Facades are similar in concept to a virtual interface, but the implementation is set by the build system. Runtime polymorphism with facades is not possible, and each facade may only have one implementation (backend) per toolchain compilation.

In the simplest sense, a facade is just a dependency represented by a variable. For example, the pw_log facade is represented by the pw_log_BACKEND build variable. Facades typically are bundled with a build system library that depends on the backend.

Facades are essential in some circumstances:

  • Low-level, platform-specific features (pw_cpu_exception).

  • Features that require a macro or non-virtual function interface (pw_log, pw_assert).

  • Highly leveraged code where a virtual interface or callback is too costly or cumbersome (pw_tokenizer).

Caution

Modules should only use facades when necessary. Facades are permanently locked to a particular implementation at compile time. Multpile backends cannot be used in one build, and runtime dependency injection is not possible, which makes testing difficult. Where appropriate, modules should use other mechanisms, such as virtual interfaces, callbacks, or templates, in place of facades.

The GN build system provides the pw_facade template as a convenient way to declare facades.

Documentation

Documentation should go in the root module folder, typically in the docs.rst file. There must be a docgen entry for the documentation in the BUILD.gn file with the target name docs; so the full target for the docs would be <module>:docs.

pw_example_module/...

  docs.rst

For modules with more involved documentation, create a separate directory called docs/ under the module root, and put the .rst files and other related files (like images and diagrams) there.

pw_example_module/...

  docs/docs.rst
  docs/bar.rst
  docs/foo.rst
  docs/image/screenshot.png
  docs/image/diagram.svg

Creating a new Pigweed module

To create a new Pigweed module, follow the below steps.

Tip

Connect with the Pigweed community (by mailing the Pigweed list or raising your idea in the Pigweed chat) to discuss your module idea before getting too far into the implementation. This can prevent accidentally duplicating work, or avoiding writing code that won’t get accepted.

  1. Create module folder following Module name guidelines

  2. Add C++ public headers files in {pw_module_dir}/public/{pw_module_name}/

  3. Add C++ implementation files files in {pw_module_dir}/

  4. Add module documentation

    • Add {pw_module_dir}/README.md that has a module summary

    • Add {pw_module_dir}/docs.rst that contains the main module documentation

  5. Add build support inside of new module

    • Add GN with {pw_module_dir}/BUILD.gn

    • Add Bazel with {pw_module_dir}/BUILD

    • Add CMake with {pw_module_dir}/CMakeLists.txt

  6. Add folder alias for new module variable in /modules.gni

    • dir_pw_new = get_path_info("pw_new", "abspath")

  7. Add new module to main GN build

    • in /BUILD.gn to group("pw_modules") using folder alias variable

  8. Add test target for new module in /BUILD.gn to pw_test_group("pw_module_tests")

  9. Add new module to CMake build

    • In /CMakeLists.txt add add_subdirectory(pw_new)

  10. Add the new module to docs module

    • Add in docs/BUILD.gn to pw_doc_gen("docs")

  11. Run pw module-check

    • $ pw module-check {pw_module_dir}

  12. Contribute your module to upstream Pigweed (optional but encouraged!)