GN / Ninja#

The GN / Ninja build system is the primary build system used for upstream Pigweed development, and is the most tested and feature-rich build system Pigweed offers.

This module’s build.gn file contains a number of C/C++ config declarations that are used by upstream Pigweed to set some architecture-agnostic compiler defaults. (See Pigweed’s //BUILDCONFIG.gn)

pw_build also provides several useful GN templates that are used throughout Pigweed.

Building upstream Pigweed#

See Pigweed’s Get started With upstream Pigweed development guide for a high-level introduction to using the GN build.

Pigweed’s root BUILD.gn file contains a variety of groups to help you control what parts of the project you’d like to build.

default: Same as just calling ninja -C out. Builds Pigweed’s documentation, recommended tests, and python linting, and static analysis.
extended_default: Everything in default, plus some other useful configurations that are tested in CQ.
all: Attempts to build everything in Pigweed. Note that pw package may need to be used to enable some branches of the build.
docs: Only build Pigweed’s documentation.
stm32f429i: Only build for the STMicroelectronics STM32F429I-DISC1 board.
host: Only build for the host.

There are a variety of other groups in the root BUILD.gn file that may be helpful for covering more areas of the build, or for reducing iteration time by only building a subset of the default build.

Some currently broken groups are gated behind the pw_BUILD_BROKEN_GROUPS build argument. You can set this to true using gn args out to try to build and debug known broken build configurations.

Build system philosophies#

While Pigweed’s GN build is not hermetic, it strives to adhere to principles of hermeticity. Some guidelines to move towards the ideal of hermeticity include:

Only rely on pre-compiled tools provided by CIPD (or some other versioned, pre-compiled binary distribution mechanism). This eliminates build artifact differences caused by different tool versions or variations (e.g. same tool version built with slightly different compilation flags).
Do not use absolute paths in Ninja commands. Typically, these appear when using rebase_path("//path/to/my_script.py"). Most of the time, Ninja steps should be passed paths rebased relative to the build directory (i.e. rebase_path("//path/to/my_script.py", root_build_dir)). This ensures build commands are the same across different machines.
Prevent produced artifacts from relying on or referencing system state. This includes time stamps, writing absolute paths to generated artifacts, or producing artifacts that reference system state in a way that prevents them from working the same way on a different machine.
Isolate build actions to the build directory. In general, the build system should not add or modify files outside of the build directory. This can cause confusion to users, and makes the concept of a clean build more ambiguous.

Target types#

import("$dir_pw_build/target_types.gni")

pw_source_set("my_library") {
  sources = [ "lib.cc" ]
}

Pigweed defines wrappers around the four basic GN binary types source_set, executable, static_library, and shared_library. These templates do several things:

Add default configs/deps

Rather than binding the majority of compiler flags related to C++ standard, cross-compilation, warning/error policy, etc. directly to toolchain invocations, these flags are applied as configs to all pw_* C/C++ target types. The primary motivations for this are to allow some targets to modify the default set of flags when needed by specifying remove_configs, and to reduce the complexity of building novel toolchains.

Pigweed’s global default configs are set in pw_build/default.gni, and individual platform-specific toolchains extend the list by appending to the default_configs build argument.

Default deps were added to support polyfill, which has since been deprecated. Default dependency functionality continues to exist for backwards compatibility.
Optionally add link-time binding

Some libraries like pw_assert and pw_log are borderline impossible to implement well without introducing circular dependencies. One solution for addressing this is to break apart the libraries into an interface with minimal dependencies, and an implementation with the bulk of the dependencies that would typically create dependency cycles. In order for the implementation to be linked in, it must be added to the dependency tree of linked artifacts (e.g. pw_executable, pw_static_library). Since there’s no way for the libraries themselves to just happily pull in the implementation if someone depends on the interface, the implementation is instead late-bound by adding it as a direct dependency of the final linked artifact. This is all managed through pw_build_LINK_DEPS, which is global for each toolchain and applied to every pw_executable, pw_static_library, and pw_shared_library.
Apply a default visibility policy

Projects can globally control the default visibility of pw_* target types by specifying pw_build_DEFAULT_VISIBILITY. This template applies that as the default visibility for any pw_* targets that do not explicitly specify a visibility.
Add source file names as metadata

All source file names are collected as GN metadata. This list can be writen to a file at build time using generated_file. The primary use case for this is to generate a token database containing all the source files. This allows PW_ASSERT to emit filename tokens even though it can’t add them to the elf file because of the reasons described at Assert API.

Note

pw_source_files, if not rebased will default to outputing module relative paths from a generated_file target. This is likely not useful. Adding a rebase argument to generated_file such as rebase = root_build_dir will result in usable paths. For an example, see //pw_tokenizer/database.gni’s pw_tokenizer_filename_database template.

The pw_executable template provides additional functionality around building complete binaries. As Pigweed is a collection of libraries, it does not know how its final targets are built. pw_executable solves this by letting each user of Pigweed specify a global executable template for their target, and have Pigweed build against it. This is controlled by the build variable pw_executable_config.target_type, specifying the name of the executable template for a project.

In some uncommon cases, a project’s pw_executable template definition may need to stamp out some pw_source_sets. Since a pw_executable template can’t import $dir_pw_build/target_types.gni due to circular imports, it should import $dir_pw_build/cc_library.gni instead.

Tip

Prefer to use pw_executable over plain executable targets to allow cleanly building the same code for multiple target configs.

Arguments#

All of the pw_* target type overrides accept any arguments supported by the underlying native types, as they simply forward them through to the underlying target.

Additionally, the following arguments are also supported:

remove_configs: (optional) A list of configs / config patterns to remove from the set of default configs specified by the current toolchain configuration.
remove_public_deps: (optional) A list of targets to remove from the set of default public_deps specified by the current toolchain configuration.

Link-only deps#

It may be necessary to specify additional link-time dependencies that may not be explicitly depended on elsewhere in the build. One example of this is a pw_assert backend, which may need to leave out dependencies to avoid circular dependencies. Its dependencies need to be linked for executables and libraries, even if they aren’t pulled in elsewhere.

The pw_build_LINK_DEPS build arg is a list of dependencies to add to all pw_executable, pw_static_library, and pw_shared_library targets. This should only be used as a last resort when dependencies cannot be properly expressed in the build.

Third party libraries#

Pigweed includes build files for a selection of third-party libraries. For a given library, these include:

third_party/<library>/library.gni: Declares build arguments like dir_pw_third_party_<library> that default to "" but can be set to the absolute path of the library in order to use it.
third_party/<library>/BUILD.gn: Describes how to build the library. This should import third_party/<library>/library.gni and refer to source paths relative to dir_pw_third_party_<library>.

To add or update GN build files for libraries that only offer Bazel build files, the Python script at pw_build/py/pw_build/generate_3p_gn.py may be used.

Note

The generate_3p_gn.py script is experimental, and may not work on an arbitrary Bazel library.

To generate or update the GN offered by Pigweed from an Bazel upstream project, first create a third_party/<library>/repo.json file. This file should describe a single JSON object, with the following fields:

name: String containg the project name.
```
"name": "FuzzTest"
```
repos: Object mapping Bazel repositories to library names.
```
"repos": { "com_google_absl": "abseil-cpp" }
```
aliases: Object mapping GN labels to other GN labels. In some cases, a third party library may have a dependency on another library already supported by Pigweed, but with a label that differs from what the script would generate. This field allows those labels to be rewritten.
```
"aliases": {
  "$dir_pw_third_party/googletest:gtest": "$dir_pw_third_party/googletest"
}
```
add: List of labels to existing GN configs. These will be added to every target in the library.
```
"add": [ "$dir_pw_third_party/re2/configs:disabled_warnings" ]
```
remove: List of labels to default GN configs. These will be removed from every target.
```
"remove" = [ "$dir_pw_fuzzer:instrumentation" ]
```
allow_testonly: Boolean indicating whether to generate GN for Bazel targets marked test-only. Defaults to false.
```
"allow_testonly": true
```
no_gn_check: List of Bazel targets that violate gn check’s rules. Third-party targets that do not conform can be excluded.
```
"no_gn_check": [ "//fuzztest:regexp_dfa" ]
```
extra_files: Object mapping additional files to create to Bazel targets that create them. These targets will be passed to bazel run and their output saved to the named file within third_party/<library>. For example:
```
"extra_files": {
  "fuzztest.bazelrc": "@com_google_fuzztest//bazel:setup_configs"
}
```

Python packages#

GN templates for Python build automation are described in Python GN Templates.

pw_cc_blob_library#

The pw_cc_blob_library template is useful for embedding binary data into a program. The template takes in a mapping of symbol names to file paths, and generates a set of C++ source and header files that embed the contents of the passed-in files as arrays of std::byte.

The blob byte arrays are constant initialized and are safe to access at any time, including before main().

pw_cc_blob_library is also available in the CMake build. It is provided by pw_build/cc_blob_library.cmake.

Arguments#

blobs: A list of GN scopes, where each scope corresponds to a binary blob to be transformed from file to byte array. This is a required field. Blob fields include:
- symbol_name: The C++ symbol for the byte array.
- file_path: The file path for the binary blob.
- linker_section: If present, places the byte array in the specified linker section.
- alignas: If present, uses the specified string or integer verbatim in the alignas() specifier for the byte array.
out_header: The header file to generate. Users will include this file exactly as it is written here to reference the byte arrays.
namespace: An optional (but highly recommended!) C++ namespace to place the generated blobs within.

Example#

BUILD.gn

pw_cc_blob_library("foo_bar_blobs") {
  blobs: [
    {
      symbol_name: "kFooBlob"
      file_path: "${target_out_dir}/stuff/bin/foo.bin"
    },
    {
      symbol_name: "kBarBlob"
      file_path: "//stuff/bin/bar.bin"
      linker_section: ".bar_section"
    },
  ]
  out_header: "my/stuff/foo_bar_blobs.h"
  namespace: "my::stuff"
  deps = [ ":generate_foo_bin" ]
}

Note

If the binary blobs are generated as part of the build, be sure to list them as deps to the pw_cc_blob_library target.

Generated Header

#pragma once

#include <array>
#include <cstddef>

namespace my::stuff {

extern const std::array<std::byte, 100> kFooBlob;

extern const std::array<std::byte, 50> kBarBlob;

}  // namespace my::stuff

Generated Source

#include "my/stuff/foo_bar_blobs.h"

#include <array>
#include <cstddef>

#include "pw_preprocessor/compiler.h"

namespace my::stuff {

const std::array<std::byte, 100> kFooBlob = { ... };

PW_PLACE_IN_SECTION(".bar_section")
const std::array<std::byte, 50> kBarBlob = { ... };

}  // namespace my::stuff

pw_facade#

In their simplest form, a facade is a GN build arg used to change a dependency at compile time. Pigweed targets configure these facades as needed.

The pw_facade template bundles a pw_source_set with a facade build arg. This allows the facade to provide header files, compilation options or anything else a GN source_set provides.

The pw_facade template declares two targets:

$target_name: the public-facing pw_source_set, with a public_dep on the backend
$target_name.facade: target used by the backend to avoid circular dependencies

# Declares ":foo" and ":foo.facade" GN targets
pw_facade("foo") {
  backend = pw_log_BACKEND
  public_configs = [ ":public_include_path" ]
  public = [ "public/pw_foo/foo.h" ]
}

Low-level facades like pw_assert cannot express all of their dependencies due to the potential for dependency cycles. Facades with this issue may require backends to place their implementations in a separate build target to be listed in pw_build_LINK_DEPS (see Link-only deps). The require_link_deps variable in pw_facade asserts that all specified build targets are present in pw_build_LINK_DEPS if the facade’s backend variable is set.

pw_python_action#

Arguments#

pw_python_action accepts all of the arguments of a regular action target. Additionally, it has some of its own arguments:

module: Run the specified Python module instead of a script. Either script or module must be specified, but not both.
capture_output: Optional boolean. If true, script output is hidden unless the script fails with an error. Defaults to true.
stamp: Optional variable indicating whether to automatically create a placeholder output file for the script. This allows running scripts without specifying outputs. If stamp is true, a generic output file is used. If stamp is a file path, that file is used as a stamp file. Like any output file, stamp must be in the build directory. Defaults to false.
environment: Optional list of strings. Environment variables to set, passed as NAME=VALUE strings.
working_directory: Optional file path. When provided the current working directory will be set to this location before the Python module or script is run.
command_launcher: Optional string. Arguments to prepend to the Python command, e.g. '/usr/bin/fakeroot --' will run the Python script within a fakeroot environment.
venv: Optional gn target of the pw_python_venv that should be used to run this action.
python_deps: Extra dependencies that are required for running the Python script for the action. This must be used with module to specify the build dependency of the module if it is user-defined code.
python_metadata_deps: Extra dependencies that are ensured completed before generating a Python package metadata manifest, not the overall Python script action. This should rarely be used by non-Pigweed code.

pw_python_action_test#

The pw_python_action_test extends pw_python_action to create a test that is run by a Python script, and its associated test metadata.

Include action tests in the pw_test_group to produce the JSON metadata that pw_test_info adds.

This template derives several additional targets:

<target_name>.metadata produces the test metadata when included in a pw_test_group. This metadata includes the Ninja target that runs the test.
If``action`` is not provided as a label, <target_name>.script wraps a pw_python_action to run the test as a standalone pw_python_package.
<target_name>.group creates a pw_python_group in order to apply tools, e.g. linters, to the standalone package.
<target_name>.lib is an empty group for compatibility with pw_test_group.
<target_name>.run invokes the test.

Targets defined using this template will produce test metadata with a test_type of “action_test” and a ninja_target value that will invoke the test when passed to Ninja, i.e. ninja -C out <ninja_target>.

Arguments#

pw_python_action_test accepts the following arguments:

All of the arguments of pw_test.
action: An optional string or scope. If a string, this should be a label to a pw_python_action target that performs the test. If a scope, this has the same meaning as for pw_python_script.
Optionally, the test_type and extra_metadata arguments of the pw_test_info template.
Optionally, all of the arguments of the pw_python_action template except module, capture_output, stamp, and python_metadata_deps.
Optionally, all of the arguments of the pw_python_package template except setup, generate_setup, tests, python_test_deps, and proto_library.

pw_test_info#

pw_test_info generates metadata describing tests. To produce a JSON file containing this metadata:

For new modules, add a pw_test_group to the BUILD.gn file. All modules are required to have a tests target.
Include one or more tests or test groups via tests or group_deps, respectively, in the pw_test_group.
Set output_metadata to true in the pw_test_group definition.

This template does not typically need to be used directly, unless adding new types of tests. It is typically used by other templates, such as the pw_test and the pw_test_group.

Arguments#

test_type: One of “test_group”, “unit_test”, “action_test”, “perf_test”, or “fuzz_test”.
test_name: Name of the test. Defaults to the target name.
build_label: GN label for the test. Defaults to the test name.
extra_metadata: Additional variables to add to the metadata.

Specific test templates add additional details using extra_metadata. For example:

The pw_test_group includes its collected list of tests and test groups as deps.
The pw_test and the Build Your Test includes the test_directory that contains the test executable.
The pw_python_action_test includes the Ninja target that can be used to invoke the Python action and run the test.

Example#

Let //my_module/BUILD.gn contain the following:

import("$dir_pw_build/python_action_test.gni")
import("$dir_pw_perf_test/perf_test.gni")
import("$dir_pw_unit_test/test.gni")

pw_test("my_unit_test") {
  sources = [ ... ]
  deps = [ ... ]
}

pw_python_action_test("my_action_test") {
  script = [ ... ]
  args = [ ... ]
  deps = [ ... ]
}

pw_python_action_test("my_integration_test") {
  script = [ ... ]
  args = [ ... ]
  deps = [ ... ]
  tags = [ "integration" ]
}

pw_perf_test("my_perf_test") {
  sources = [ ... ]
  deps = [ ... ]
}

pw_test_group("tests") {
  tests = [
   ":my_unit_test",
   ":my_action_test",
   ":my_integration_test",
  ]
}

Let //BUILD.gn` contain the following:

import("$dir_pw_unit_test/test.gni")

group("run_tests") {
  deps = [ ":my_module_tests(//targets/my_targets:my_toolchain)" ]
}

pw_test_group("my_module_tests") {
  group_deps = [ "//my_module:tests" ]
  output_metadata = true
}

Then running gn gen out will produce the following JSON file at out/my_toolchain/my_module_tests.testinfo.json:

[
  {
    "build_label": "//my_module:my_unit_test",
    "test_directory": "my_toolchain/obj/my_module/test",
    "test_name": "my_unit_test",
    "test_type": "unit_test"
  },
  {
    "build_label": "//my_module:my_action_test",
    "ninja_target": "my_toolchain/obj/my_module/my_action_test.run.stamp",
    "test_name": "my_action_test",
    "test_type": "action_test"
  },
  {
    "build_label": "//my_module:my_integration_test",
    "ninja_target": "my_toolchain/obj/my_module/my_integration_test.run.stamp",
    "tags": [
      "integration"
    ],
    "test_name": "my_integration_test",
    "test_type": "action_test"
  },
  {
    "build_label": "//my_module:my_perf_test",
    "test_directory": "my_toolchain/obj/my_module/test",
    "test_name": "my_perf_test",
    "test_type": "perf_test"
  },
  {
    "build_label": "//my_module:tests",
    "deps": [
      "//my_module:my_unit_test",
      "//my_module:my_action_test",
      "//my_module:my_integration_test",
    ],
    "test_name": "my_module/tests",
    "test_type": "test_group"
  },
  {
    "build_label": "//:my_module_tests",
    "deps": [
      "//my_module:tests",
    ],
    "test_name": "my_module_tests",
    "test_type": "test_group"
  }
]

Expressions#

pw_python_action evaluates expressions in args, the arguments passed to the script. These expressions function similarly to generator expressions in CMake. Expressions may be passed as a standalone argument or as part of another argument. A single argument may contain multiple expressions.

Generally, these expressions are used within templates rather than directly in BUILD.gn files. This allows build code to use GN labels without having to worry about converting them to files.

Note

We intend to replace these expressions with native GN features when possible. See b/234886742.

The following expressions are supported:

<TARGET_FILE(gn_target)>

Evaluates to the output file of the provided GN target. For example, the expression

"<TARGET_FILE(//foo/bar:static_lib)>"

might expand to

"/home/User/project_root/out/obj/foo/bar/static_lib.a"

TARGET_FILE parses the .ninja file for the GN target, so it should always find the correct output file, regardless of the toolchain’s or target’s configuration. Some targets, such as source_set and group targets, do not have an output file, and attempting to use TARGET_FILE with them results in an error.

TARGET_FILE only resolves GN target labels to their outputs. To resolve paths generally, use the standard GN approach of applying the rebase_path(path, root_build_dir) function. This function converts the provided GN path or list of paths to be relative to the build directory, from which all build commands and scripts are executed.

<TARGET_FILE_IF_EXISTS(gn_target)>

TARGET_FILE_IF_EXISTS evaluates to the output file of the provided GN target, if the output file exists. If the output file does not exist, the entire argument that includes this expression is omitted, even if there is other text or another expression.

For example, consider this expression:

"--database=<TARGET_FILE_IF_EXISTS(//alpha/bravo)>"

If the //alpha/bravo target file exists, this might expand to the following:

"--database=/home/User/project/out/obj/alpha/bravo/bravo.elf"

If the //alpha/bravo target file does not exist, the entire --database= argument is omitted from the script arguments.

<TARGET_OBJECTS(gn_target)>

Evaluates to the object files of the provided GN target. Expands to a separate argument for each object file. If the target has no object files, the argument is omitted entirely. Because it does not expand to a single expression, the <TARGET_OBJECTS(...)> expression may not have leading or trailing text.

For example, the expression

"<TARGET_OBJECTS(//foo/bar:a_source_set)>"

might expand to multiple separate arguments:

"/home/User/project_root/out/obj/foo/bar/a_source_set.file_a.cc.o"
"/home/User/project_root/out/obj/foo/bar/a_source_set.file_b.cc.o"
"/home/User/project_root/out/obj/foo/bar/a_source_set.file_c.cc.o"

Example#

import("$dir_pw_build/python_action.gni")

pw_python_action("postprocess_main_image") {
  script = "py/postprocess_binary.py"
  args = [
    "--database",
    rebase_path("my/database.csv", root_build_dir),
    "--binary=<TARGET_FILE(//firmware/images:main)>",
  ]
  stamp = true
}

pw_evaluate_path_expressions#

It is not always feasible to pass information to a script through command line arguments. If a script requires a large amount of input data, writing to a file is often more convenient. However, doing so bypasses pw_python_action’s GN target label resolution, preventing such scripts from working with build artifacts in a build system-agnostic manner.

pw_evaluate_path_expressions is designed to address this use case. It takes a list of input files and resolves target expressions within them, modifying the files in-place.

Refer to pw_python_action’s Expressions section for the list of supported expressions.

Note

pw_evaluate_path_expressions is typically used as an intermediate sub-target of a larger template, rather than a standalone build target.

Arguments#

files: A list of scopes, each containing a source file to process and a dest file to which to write the result.

Example#

The following template defines an executable target which additionally outputs the list of object files from which it was compiled, making use of pw_evaluate_path_expressions to resolve their paths.

import("$dir_pw_build/evaluate_path_expressions.gni")

template("executable_with_artifacts") {
  executable("${target_name}.exe") {
    sources = invoker.sources
    if defined(invoker.deps) {
      deps = invoker.deps
    }
  }

  _artifacts_input = "$target_gen_dir/${target_name}_artifacts.json.in"
  _artifacts_output = "$target_gen_dir/${target_name}_artifacts.json"
  _artifacts = {
    binary = "<TARGET_FILE(:${target_name}.exe)>"
    objects = "<TARGET_OBJECTS(:${target_name}.exe)>"
  }
  write_file(_artifacts_input, _artifacts, "json")

  pw_evaluate_path_expressions("${target_name}.evaluate") {
    files = [
      {
        source = _artifacts_input
        dest = _artifacts_output
      },
    ]
  }

  group(target_name) {
    deps = [
      ":${target_name}.exe",
      ":${target_name}.evaluate",
    ]
  }
}

pw_exec#

pw_exec allows for execution of arbitrary programs. It is a wrapper around pw_python_action but allows for specifying the program to execute.

Note

Prefer to use pw_python_action instead of calling out to shell scripts, as the Python will be more portable. pw_exec should generally only be used for interacting with legacy/existing scripts.

Arguments#

program: The program to run. Can be a full path or just a name (in which case $PATH is searched).
args: Optional list of arguments to the program.
deps: Dependencies for this target.
public_deps: Public dependencies for this target. In addition to outputs from this target, outputs generated by public dependencies can be used as inputs from targets that depend on this one. This is not the case for private deps.
inputs: Optional list of build inputs to the program.
outputs: Optional list of artifacts produced by the program’s execution.
env: Optional list of key-value pairs defining environment variables for the program.
env_file: Optional path to a file containing a list of newline-separated key-value pairs defining environment variables for the program.
args_file: Optional path to a file containing additional positional arguments to the program. Each line of the file is appended to the invocation. Useful for specifying arguments from GN metadata.
skip_empty_args: If args_file is provided, boolean indicating whether to skip running the program if the file is empty. Used to avoid running commands which error when called without arguments.
capture_output: If true, output from the program is hidden unless the program exits with an error. Defaults to true.
working_directory: The working directory to execute the subprocess with. If not specified it will not be set and the subprocess will have whatever the parent current working directory is.
venv: Python virtualenv to pass along to the underlying pw_python_action.
visibility: GN visibility to apply to the underlying target.

Example#

import("$dir_pw_build/exec.gni")

pw_exec("hello_world") {
  program = "/bin/sh"
  args = [
    "-c",
    "echo hello \$WORLD",
  ]
  env = [
    "WORLD=world",
  ]
}

pw_input_group#

pw_input_group defines a group of input files which are not directly processed by the build but are still important dependencies of later build steps. This is commonly used alongside metadata to propagate file dependencies through the build graph and force rebuilds on file modifications.

For example pw_docgen defines a pw_doc_group template which outputs metadata from a list of input files. The metadata file is not actually part of the build, and so changes to any of the input files do not trigger a rebuild. This is problematic, as targets that depend on the metadata should rebuild when the inputs are modified but GN cannot express this dependency.

pw_input_group solves this problem by allowing a list of files to be listed in a target that does not output any build artifacts, causing all dependent targets to correctly rebuild.

Arguments#

pw_input_group accepts all arguments that can be passed to a group target, as well as requiring one extra:

inputs: List of input files.

Example#

import("$dir_pw_build/input_group.gni")

pw_input_group("foo_metadata") {
  metadata = {
    files = [
      "x.foo",
      "y.foo",
      "z.foo",
    ]
  }
  inputs = metadata.files
}

Targets depending on foo_metadata will rebuild when any of the .foo files are modified.

pw_zip#

pw_zip is a target that allows users to zip up a set of input files and directories into a single output .zip file—a simple automation of a potentially repetitive task.

Arguments#

inputs: List of source files as well as the desired relative zip destination. See below for the input syntax.
dirs: List of entire directories to be zipped as well as the desired relative zip destination. See below for the input syntax.
output: Filename of output .zip file.
deps: List of dependencies for the target.

Input Syntax#

Inputs all need to follow the correct syntax:

Path to source file or directory. Directories must end with a /.
The delimiter (defaults to >).
The desired destination of the contents within the .zip. Must start with / to indicate the zip root. Any number of subdirectories are allowed. If the source is a file it can be put into any subdirectory of the root. If the source is a file, the zip copy can also be renamed by ending the zip destination with a filename (no trailing /).

Thus, it should look like the following: "[source file or dir] > /".

Example#

Let’s say we have the following structure for a //source/ directory:

source/
├── file1.txt
├── file2.txt
├── file3.txt
└── some_dir/
    ├── file4.txt
    └── some_other_dir/
        └── file5.txt

And we create the following build target:

import("$dir_pw_build/zip.gni")

pw_zip("target_name") {
  inputs = [
    "//source/file1.txt > /",             # Copied to the zip root dir.
    "//source/file2.txt > /renamed.txt",  # File renamed.
    "//source/file3.txt > /bar/",         # File moved to the /bar/ dir.
  ]

  dirs = [
    "//source/some_dir/ > /bar/some_dir/",  # All /some_dir/ contents copied
                                            # as /bar/some_dir/.
  ]

  # Note on output: if the specific output directory isn't defined
  # (such as output = "zoo.zip") then the .zip will output to the
  # same directory as the BUILD.gn file that called the target.
  output = "//$target_out_dir/foo.zip"  # Where the foo.zip will end up
}

This will result in a .zip file called foo.zip stored in //$target_out_dir with the following structure:

foo.zip
├── bar/
│   ├── file3.txt
│   └── some_dir/
│       ├── file4.txt
│       └── some_other_dir/
│           └── file5.txt
├── file1.txt
└── renamed.txt

pw_relative_source_file_names#

This template recursively walks the listed dependencies and collects the names of all the headers and source files required by the targets, and then transforms them such that they reflect the __FILE__ when pw_build’s relative_paths config is applied. This is primarily intended for side-band generation of pw_tokenizer tokens so file name tokens can be utilized in places where pw_tokenizer is unable to embed token information as part of C/C++ compilation.

This template produces a JSON file containing an array of strings (file paths with -ffile-prefix-map-like transformations applied) that can be used to generate a token database.

Arguments#

deps: A required list of targets to recursively extract file names from.
outputs: A required array with a single element: the path to write the final JSON file to.

Example#

Let’s say we have the following project structure:

project root
├── foo/
│   ├── foo.h
│   └── foo.cc
├── bar/
│   ├── bar.h
│   └── bar.cc
├── unused/
│   ├── unused.h
│   └── unused.cc
└── main.cc

And a BUILD.gn at the root:

pw_source_set("bar") {
  public_configs = [ ":bar_headers" ]
  public = [ "bar/bar.h" ]
  sources = [ "bar/bar.cc" ]
}

pw_source_set("foo") {
  public_configs = [ ":foo_headers" ]
  public = [ "foo/foo.h" ]
  sources = [ "foo/foo.cc" ]
  deps = [ ":bar" ]
}


pw_source_set("unused") {
  public_configs = [ ":unused_headers" ]
  public = [ "unused/unused.h" ]
  sources = [ "unused/unused.cc" ]
  deps = [ ":bar" ]
}

pw_executable("main") {
  sources = [ "main.cc" ]
  deps = [ ":foo" ]
}

pw_relative_source_file_names("main_source_files") {
  deps = [ ":main" ]
  outputs = [ "$target_gen_dir/main_source_files.json" ]
}

The json file written to out/gen/main_source_files.json will contain:

[
  "bar/bar.cc",
  "bar/bar.h",
  "foo/foo.cc",
  "foo/foo.h",
  "main.cc"
]

Since unused isn’t a transitive dependency of main, its source files are not included. Similarly, even though bar is not a direct dependency of main, its source files are included because foo brings in bar as a transitive dependency.

Note how the file paths in this example are relative to the project root rather than being absolute paths (e.g. /home/user/ralph/coding/my_proj/main.cc). This is a result of transformations applied to strip absolute pathing prefixes, matching the behavior of pw_build’s $dir_pw_build:relative_paths config.

Build time errors: pw_error and pw_build_assert#

In Pigweed’s complex, multi-toolchain GN build it is not possible to build every target in every configuration. GN’s assert statement is not ideal for enforcing the correct configuration because it may prevent the GN build files or targets from being referred to at all, even if they aren’t used.

The pw_error GN template results in an error if it is executed during the build. These error targets can exist in the build graph, but cannot be depended on without an error.

pw_build_assert evaluates to a pw_error if a condition fails or nothing (an empty group) if the condition passes. Targets can add a dependency on a pw_build_assert to enforce a condition at build time.

The templates for build time errors are defined in pw_build/error.gni.

Generate code coverage reports: `pw_coverage_report`#

Pigweed supports generating coverage reports, in a variety of formats, for C/C++ code using the pw_coverage_report GN template.

Coverage Caveats#

There are currently two code coverage caveats when enabled:

Coverage reports are only populated based on host tests that use a clang toolchain.
Coverage reports will only show coverage information for headers included in a test binary.

These two caveats mean that all device-specific code that cannot be compiled for and run on the host will not be able to have reports generated for them, and that the existence of these files will not appear in any coverage report.

Try to ensure that your code can be written in a way that it can be compiled into a host test for the purpose of coverage reporting, although this is sometimes impossible due to requiring hardware-specific APIs to be available.

Coverage Instrumentation#

For the pw_coverage_report to generate meaningful output, you must ensure that it is invoked by a toolchain that instruments tests for code coverage collection and output.

Instrumentation is controlled by two GN build arguments:

pw_toolchain_COVERAGE_ENABLED being set to true.
pw_toolchain_PROFILE_SOURCE_FILES is an optional argument that provides a list of source files to selectively collect coverage.

Note

It is possible to also instrument binaries for UBSAN, ASAN, or TSAN at the same time as coverage. However, TSAN will find issues in the coverage instrumentation code and fail to properly build.

This can most easily be done by using the host_clang_coverage toolchain provided in pw_toolchain, but you can also create custom toolchains that manually set these GN build arguments as well.

`pw_coverage_report`#

pw_coverage_report is basically a GN frontend to the llvm-cov tool that can be integrated into the normal build.

It can be found at pw_build/coverage_report.gni and is available through import("$dir_pw_build/coverage_report.gni").

The supported report formats are:

text: A text representation of the code coverage report. This format is not suitable for further machine manipulation and is instead only useful for cases where a human needs to interpret the report. The text format provides a nice summary, but if you desire to drill down into the coverage details more, please consider using html instead.
- This is equivalent to llvm-cov show --format text and similar to llvm-cov report.
html: A static HTML site that provides an overall coverage summary and per-file information. This format is not suitable for further machine manipulation and is instead only useful for cases where a human needs to interpret the report.
- This is equivalent to llvm-cov show --format html.
lcov: A machine-friendly coverage report format. This format is not human- friendly. If that is necessary, use text or html instead.
- This is equivalent to llvm-cov export --format lcov.
json: A machine-friendly coverage report format. This format is not human- friendly. If that is necessary, use text or html instead.
- This is equivalent to llvm-cov export --format json.

Arguments#

There are three classes of template arguments: build, coverage, and test.

Build Arguments:

enable_if (optional): Conditionally activates coverage report generation when set to a boolean expression that evaluates to true. This can be used to allow project builds to conditionally enable or disable coverage reports to minimize work needed for certain build configurations.
failure_mode (optional/unstable): Specify the failure mode for llvm-profdata (used to merge inidividual profraw files from pw_test runs). Available options are "any" (default) or "all".
- This should be considered an unstable/deprecated argument that should only be used as a last resort to get a build working again. Using failure_mode = "all" usually indicates that there are underlying problems in the build or test infrastructure that should be independently resolved. Please reach out to the Pigweed team for assistance.

Coverage Arguments:

filter_paths (optional): List of file paths to include when generating the coverage report. These cannot be regular expressions, but can be concrete file or folder paths. Folder paths will allow all files in that directory or any recursive child directory.
- These are passed to llvm-cov by the optional trailing positional [SOURCES] arguments.
ignore_filename_patterns (optional): List of file path regular expressions to ignore when generating the coverage report.
- These are passed to llvm-cov via --ignore-filename-regex named parameters.

Test Arguments (one of these is required to be provided):

tests: A list of pw_test targets.
group_deps: A list of pw_test_group targets.

Note

tests and group_deps are treated exactly the same by pw_coverage_report, so it is not that important to ensure they are used properly.

Target Expansion#

pw_coverage_report(<target_name>) expands to one concrete target for each report format.

<target_name>.text: Generates the text coverage report.
<target_name>.html: Generates the html coverage report.
<target_name>.lcov: Generates the lcov coverage report.
<target_name>.json: Generates the json coverage report.

To use any of these targets, you need only to add a dependency on the desired target somewhere in your build.

There is also a <target_name> target generated that is a group that adds a dependency on all of the format-specific targets listed above.

Note

These targets are always available, even when the toolchain executing the target does not support coverage or coverage is not enabled. In these cases, the targets are set to empty groups.

Coverage Output#

Coverage reports are currently generated and placed into the build output directory associated with the path to the GN file where the pw_coverage_report is used in a subfolder named <target_name>.<report_type>.

Note

Due to limitations with telling GN the entire output of coverage reports (stemming from per-source-file generation for HTML and text representations), it is not as simple as using GN’s built-in copy to be able to move these coverage reports to another output location. However, it seems possible to add a target that can use Python to copy the entire output directory.

Improved Ninja interface#

Ninja includes a basic progress display, showing in a single line the number of targets finished, the total number of targets, and the name of the most recent target it has either started or finished.

For additional insight into the status of the build, Pigweed includes a Ninja wrapper, pw-wrap-ninja, that displays additional real-time information about the progress of the build. The wrapper is invoked the same way you’d normally invoke Ninja:

pw-wrap-ninja -C out

The script lists the progress of the build, as well as the list of targets that Ninja is currently building, along with a timer that measures how long each target has been building for:

[51.3s] Building [8924/10690] ...
  [10.4s] c++ pw_strict_host_clang_debug/obj/pw_string/string_test.lib.string_test.cc.o
  [ 9.5s] ACTION //pw_console/py:py.lint.mypy(//pw_build/python_toolchain:python)
  [ 9.4s] ACTION //pw_console/py:py.lint.pylint(//pw_build/python_toolchain:python)
  [ 6.1s] clang-tidy ../pw_log_rpc/log_service.cc
  [ 6.1s] clang-tidy ../pw_log_rpc/log_service_test.cc
  [ 6.1s] clang-tidy ../pw_log_rpc/rpc_log_drain.cc
  [ 6.1s] clang-tidy ../pw_log_rpc/rpc_log_drain_test.cc
  [ 5.4s] c++ pw_strict_host_clang_debug/obj/BUILD_DIR/pw_strict_host_clang_debug/gen/pw...
  ... and 109 more

This allows you to, at a glance, know what Ninja’s currently building, which targets are bottlenecking the rest of the build, and which targets are taking an unusually long time to complete.

pw-wrap-ninja includes other useful functionality as well. The --write-trace option writes a build trace to the specified path, which can be viewed in the Perfetto UI, or via Chrome’s built-in chrome://tracing tool.

GN / Ninja#

Building upstream Pigweed#

Build system philosophies#

Target types#

Arguments#

Link-only deps#

Third party libraries#

Python packages#

pw_cc_blob_library#

Arguments#

Example#

pw_facade#

pw_python_action#

Arguments#

pw_python_action_test#

Arguments#

pw_test_info#

Arguments#

Example#

Expressions#

Example#

pw_evaluate_path_expressions#

Arguments#

Example#

pw_exec#

Arguments#

Example#

pw_input_group#

Arguments#

Example#

pw_zip#

Arguments#

Input Syntax#

Example#

pw_relative_source_file_names#

Arguments#

Example#

Build time errors: pw_error and pw_build_assert#

Generate code coverage reports: pw_coverage_report#

Coverage Caveats#

Coverage Instrumentation#

pw_coverage_report#

Arguments#

Target Expansion#

Coverage Output#

Improved Ninja interface#

Generate code coverage reports: `pw_coverage_report`#

`pw_coverage_report`#