This page is automatically generated from gn help --markdown all
.
Analyze which targets are affected by a list of files. This command takes three arguments: out_dir is the path to the build directory. input_path is a path to a file containing a JSON object with three fields: - "files": A list of the filenames to check. - "test_targets": A list of the labels for targets that are needed to run the tests we wish to run. - "additional_compile_targets": A list of the labels for targets that we wish to rebuild, but aren't necessarily needed for testing. The important difference between this field and "test_targets" is that if an item in the additional_compile_targets list refers to a group, then any dependencies of that group will be returned if they are out of date, but the group itself does not need to be. If the dependencies themselves are groups, the same filtering is repeated. This filtering can be used to avoid rebuilding dependencies of a group that are unaffected by the input files. The list may also contain the string "all" to refer to a pseudo-group that contains every root target in the build graph. This filtering behavior is also known as "pruning" the list of compile targets. If input_path is -, input is read from stdin. output_path is a path indicating where the results of the command are to be written. The results will be a file containing a JSON object with one or more of following fields: - "compile_targets": A list of the labels derived from the input compile_targets list that are affected by the input files. Due to the way the filtering works for compile targets as described above, this list may contain targets that do not appear in the input list. - "test_targets": A list of the labels from the input test_targets list that are affected by the input files. This list will be a proper subset of the input list. - "invalid_targets": A list of any names from the input that do not exist in the build graph. If this list is non-empty, the "error" field will also be set to "Invalid targets". - "status": A string containing one of three values: - "Found dependency" - "No dependency" - "Found dependency (all)" In the first case, the lists returned in compile_targets and test_targets should be passed to ninja to build. In the second case, nothing was affected and no build is necessary. In the third case, GN could not determine the correct answer and returned the input as the output in order to be safe. - "error": This will only be present if an error occurred, and will contain a string describing the error. This includes cases where the input file is not in the right format, or contains invalid targets. If output_path is -, output is written to stdout. The command returns 1 if it is unable to read the input file or write the output file, or if there is something wrong with the build such that gen would also fail, and 0 otherwise. In particular, it returns 0 even if the "error" key is non-empty and a non-fatal error occurred. In other words, it tries really hard to always write something to the output JSON and convey errors that way rather than via return codes.
Display or configure arguments declared by the build. gn args <out_dir> [--list] [--short] [--args] [--overrides-only] See also "gn help buildargs" for a more high-level overview of how build arguments work.
gn args <out_dir> Open the arguments for the given build directory in an editor. If the given build directory doesn't exist, it will be created and an empty args file will be opened in the editor. You would type something like this into that file: enable_doom_melon=false os="android" To find your editor on Posix, GN will search the environment variables in order: GN_EDITOR, VISUAL, and EDITOR. On Windows GN will open the command associated with .txt files. Note: you can edit the build args manually by editing the file "args.gn" in the build directory and then running "gn gen <out_dir>". gn args <out_dir> --list[=<exact_arg>] [--short] [--overrides-only] [--json] Lists all build arguments available in the current configuration, or, if an exact_arg is specified for the list flag, just that one build argument. The output will list the declaration location, current value for the build, default value (if different than the current value), and comment preceding the declaration. If --short is specified, only the names and current values will be printed. If --overrides-only is specified, only the names and current values of arguments that have been overridden (i.e. non-default arguments) will be printed. Overrides come from the <out_dir>/args.gn file and //.gn If --json is specified, the output will be emitted in json format. JSON schema for output: [ { "name": variable_name, "current": { "value": overridden_value, "file": file_name, "line": line_no }, "default": { "value": default_value, "file": file_name, "line": line_no }, "comment": comment_string }, ... ]
gn args out/Debug Opens an editor with the args for out/Debug. gn args out/Debug --list --short Prints all arguments with their default values for the out/Debug build. gn args out/Debug --list --short --overrides-only Prints overridden arguments for the out/Debug build. gn args out/Debug --list=target_cpu Prints information about the "target_cpu" argument for the " "out/Debug build. gn args --list --args="os=\"android\" enable_doom_melon=true" Prints all arguments with the default values for a build with the given arguments set (which may affect the values of other arguments).
GN's include header checker validates that the includes for C-like source files match the build dependency graph. "gn check" is the same thing as "gn gen" with the "--check" flag except that this command does not write out any build files. It's intended to be an easy way to manually trigger include file checking. The <label_pattern> can take exact labels or patterns that match more than one (although not general regular expressions). If specified, only those matching targets will be checked. See "gn help label_pattern" for details.
--check-generated Generated files are normally not checked since they do not exist until after a build. With this flag, those generated files that can be found on disk are also checked. --check-system Check system style includes (using <angle brackets>) in addition to "double quote" includes. --default-toolchain Normally wildcard targets are matched in all toolchains. This switch makes wildcard labels with no explicit toolchain reference only match targets in the default toolchain. Non-wildcard inputs with no explicit toolchain specification will always match only a target in the default toolchain if one exists. --force Ignores specifications of "check_includes = false" and checks all target's files that match the target label.
The .gn file may specify a list of targets to be checked in the list check_targets (see "gn help dotfile"). Alternatively, the .gn file may specify a list of targets not to be checked in no_check_targets. If a label pattern is specified on the command line, neither check_targets or no_check_targets is used. Targets can opt-out from checking with "check_includes = false" (see "gn help check_includes"). For targets being checked: - GN opens all C-like source files in the targets to be checked and scans the top for includes. - Generated files (that might not exist yet) are ignored unless the --check-generated flag is provided. - Includes with a "nogncheck" annotation are skipped (see "gn help nogncheck"). - Includes using "quotes" are always checked. If system style checking is enabled, includes using <angle brackets> are also checked. - Include paths are assumed to be relative to any of the "include_dirs" for the target (including the implicit current dir). - GN does not run the preprocessor so will not understand conditional includes. - Only includes matching known files in the build are checked: includes matching unknown paths are ignored. For an include to be valid: - The included file must be in the current target, or there must be a path following only public dependencies to a target with the file in it ("gn path" is a good way to diagnose problems). - There can be multiple targets with an included file: only one needs to be valid for the include to be allowed. - If there are only "sources" in a target, all are considered to be public and can be included by other targets with a valid public dependency path. - If a target lists files as "public", only those files are able to be included by other targets. Anything in the sources will be considered private and will not be includable regardless of dependency paths. - Outputs from actions are treated like public sources on that target. - A target can include headers from a target that depends on it if the other target is annotated accordingly. See "gn help allow_circular_includes_from".
If you have a third party project that is difficult to fix or doesn't care about include checks it's generally best to exclude that target from checking altogether via "check_includes = false". If you have conditional includes, make sure the build conditions and the preprocessor conditions match, and annotate the line with "nogncheck" (see "gn help nogncheck" for an example). If two targets are hopelessly intertwined, use the "allow_circular_includes_from" annotation. Ideally each should have identical dependencies so configs inherited from those dependencies are consistent (see "gn help allow_circular_includes_from"). If you have a standalone header file or files that need to be shared between a few targets, you can consider making a source_set listing only those headers as public sources. With only header files, the source set will be a no-op from a build perspective, but will give a central place to refer to those headers. That source set's files will still need to pass "gn check" in isolation. In rare cases it makes sense to list a header in more than one target if it could be considered conceptually a member of both.
gn check out/Debug Check everything. gn check out/Default //foo:bar Check only the files in the //foo:bar target. gn check out/Default "//foo/* Check only the files in targets in the //foo directory tree.
Deletes the contents of the output directory except for args.gn and creates a Ninja build environment sufficient to regenerate the build.
gn desc <out_dir> <label or pattern> [<what to show>] [--blame] [--format=json] Displays information about a given target or config. The build parameters will be taken for the build in the given <out_dir>. The <label or pattern> can be a target label, a config label, or a label pattern (see "gn help label_pattern"). A label pattern will only match targets.
(If unspecified an overall summary will be displayed.) all_dependent_configs allow_circular_includes_from arflags [--blame] args cflags [--blame] cflags_c [--blame] cflags_cc [--blame] check_includes configs [--tree] (see below) data_keys defines [--blame] depfile deps [--all] [--tree] (see below) framework_dirs frameworks include_dirs [--blame] inputs ldflags [--blame] lib_dirs libs metadata output_conversion outputs public_configs public rebase script sources testonly visibility walk_keys weak_frameworks runtime_deps Compute all runtime deps for the given target. This is a computed list and does not correspond to any GN variable, unlike most other values here. The output is a list of file names relative to the build directory. See "gn help runtime_deps" for how this is computed. This also works with "--blame" to see the source of the dependency.
--default-toolchain Normally wildcard targets are matched in all toolchains. This switch makes wildcard labels with no explicit toolchain reference only match targets in the default toolchain. Non-wildcard inputs with no explicit toolchain specification will always match only a target in the default toolchain if one exists. --format=json Format the output as JSON instead of text.
--blame Used with any value specified on a config, this will name the config that causes that target to get the flag. This doesn't currently work for libs, lib_dirs, frameworks, weak_frameworks and framework_dirs because those are inherited and are more complicated to figure out the blame (patches welcome).
The "configs" section will list all configs that apply. For targets this will include configs specified in the "configs" variable of the target, and also configs pushed onto this target via public or "all dependent" configs. Configs can have child configs. Specifying --tree will show the hierarchy.
The "outputs" section will list all outputs that apply, including the outputs computed from the tool definition (eg for "executable", "static_library", ... targets).
Deps will include all public, private, and data deps (TODO this could be clarified and enhanced) sorted in order applying. The following may be used: --all Collects all recursive dependencies and prints a sorted flat list. Also usable with --tree (see below). --as=(buildfile|label|output) How to print targets. buildfile Prints the build files where the given target was declared as file names. label (default) Prints the label of the target. output Prints the first output file for the target relative to the root build directory. --testonly=(true|false) Restrict outputs to targets with the testonly flag set accordingly. When unspecified, the target's testonly flags are ignored. --tree Print a dependency tree. By default, duplicates will be elided with "..." but when --all and -tree are used together, no eliding will be performed. The "deps", "public_deps", and "data_deps" will all be included in the tree. Tree output can not be used with the filtering or output flags: --as, --type, --testonly. --type=(action|copy|executable|group|loadable_module|shared_library| source_set|static_library) Restrict outputs to targets matching the given type. If unspecified, no filtering will be performed.
This command will show the full name of directories and source files, but when directories and source paths are written to the build file, they will be adjusted to be relative to the build directory. So the values for paths displayed by this command won't match (but should mean the same thing).
gn desc out/Debug //base:base Summarizes the given target. gn desc out/Foo :base_unittests deps --tree Shows a dependency tree of the "base_unittests" project in the current directory. gn desc out/Debug //base defines --blame Shows defines set for the //base:base target, annotated by where each one was set from.
Formats .gn file to a standard format. The contents of some lists ('sources', 'deps', etc.) will be sorted to a canonical order. To suppress this, you can add a comment of the form "# NOSORT" immediately preceding the assignment. e.g. # NOSORT sources = [ "z.cc", "a.cc", ]
--dry-run Prints the list of files that would be reformatted but does not write anything to disk. This is useful for presubmit/lint-type checks. - Exit code 0: successful format, matches on disk. - Exit code 1: general failure (parse error, etc.) - Exit code 2: successful format, but differs from on disk. --dump-tree[=( text | json )] Dumps the parse tree to stdout and does not update the file or print formatted output. If no format is specified, text format will be used. --stdin Read input from stdin and write to stdout rather than update a file in-place.
gn format //some/BUILD.gn //some/other/BUILD.gn //and/another/BUILD.gn gn format some\\BUILD.gn gn format /abspath/some/BUILD.gn gn format --stdin
Generates ninja files from the current tree and puts them in the given output directory. The output directory can be a source-repo-absolute path name such as: //out/foo Or it can be a directory relative to the current directory such as: out/foo "gn gen --check" is the same as running "gn check". "gn gen --check=system" is the same as running "gn check --check-system". See "gn help check" for documentation on that mode. See "gn help switches" for the common command-line switches.
--ninja-executable=<string> Can be used to specify the ninja executable to use. This executable will be used as an IDE option to indicate which ninja to use for building. This executable will also be used as part of the gen process for triggering a restat on generated ninja files.
GN optionally generates files for IDE. Files won't be overwritten if their contents don't change. Possibilities for <ide options> --ide=<ide_name> Generate files for an IDE. Currently supported values: "eclipse" - Eclipse CDT settings file. "vs" - Visual Studio project/solution files. (default Visual Studio version: 2019) "vs2013" - Visual Studio 2013 project/solution files. "vs2015" - Visual Studio 2015 project/solution files. "vs2017" - Visual Studio 2017 project/solution files. "vs2019" - Visual Studio 2019 project/solution files. "xcode" - Xcode workspace/solution files. "qtcreator" - QtCreator project files. "json" - JSON file containing target information --filters=<path_prefixes> Semicolon-separated list of label patterns used to limit the set of generated projects (see "gn help label_pattern"). Only matching targets and their dependencies will be included in the solution. Only used for Visual Studio, Xcode and JSON.
--sln=<file_name> Override default sln file name ("all"). Solution file is written to the root build directory. --no-deps Don't include targets dependencies to the solution. Changes the way how --filters option works. Only directly matching targets are included. --winsdk=<sdk_version> Use the specified Windows 10 SDK version to generate project files. As an example, "10.0.15063.0" can be specified to use Creators Update SDK instead of the default one. --ninja-extra-args=<string> This string is passed without any quoting to the ninja invocation command-line. Can be used to configure ninja flags, like "-j".
--xcode-project=<file_name> Override defaut Xcode project file name ("all"). The project file is written to the root build directory. --xcode-build-system=<value> Configure the build system to use for the Xcode project. Supported values are (default to "legacy"): "legacy" - Legacy Build system "new" - New Build System --ninja-executable=<string> Can be used to specify the ninja executable to use when building. --ninja-extra-args=<string> This string is passed without any quoting to the ninja invocation command-line. Can be used to configure ninja flags, like "-j". --ide-root-target=<target_name> Name of the target corresponding to "All" target in Xcode. If unset, "All" invokes ninja without any target and builds everything.
--ide-root-target=<target_name> Name of the root target for which the QtCreator project will be generated to contain files of it and its dependencies. If unset, the whole build graph will be emitted.
GN DOES NOT generate Eclipse CDT projects. Instead, it generates a settings file which can be imported into an Eclipse CDT project. The XML file contains a list of include paths and defines. Because GN does not generate a full .cproject definition, it is not possible to properly define includes/defines for each file individually. Instead, one set of includes/defines is generated for the entire project. This works fairly well but may still result in a few indexer issues here and there.
Dumps target information to a JSON file and optionally invokes a python script on the generated file. See the comments at the beginning of json_project_writer.cc and desc_builder.cc for an overview of the JSON file format. --json-file-name=<json_file_name> Overrides default file name (project.json) of generated JSON file. --json-ide-script=<path_to_python_script> Executes python script after the JSON file is generated or updated with new content. Path can be project absolute (//), system absolute (/) or relative, in which case the output directory will be base. Path to generated JSON file will be first argument when invoking script. --json-ide-script-args=<argument> Optional second argument that will passed to executed script.
--export-rust-project Produces a rust-project.json file in the root of the build directory This is used for various tools in the Rust ecosystem allowing for the replay of individual compilations independent of the build system. This is an unstable format and likely to change without warning. --export-compile-commands[=<target_name1,target_name2...>] Produces a compile_commands.json file in the root of the build directory containing an array of “command objects”, where each command object specifies one way a translation unit is compiled in the project. If a list of target_name is supplied, only targets that are reachable from any target in any build file whose name is target_name will be used for “command objects” generation, otherwise all available targets will be used. This is used for various Clang-based tooling, allowing for the replay of individual compilations independent of the build system. e.g. "foo" will match: - "//path/to/src:foo" - "//other/path:foo" - "//foo:foo" and not match: - "//foo:bar"
Yo dawg, I heard you like help on your help so I put help on the help in the help. You can also use "all" as the parameter to get all help at once.
--markdown Format output in markdown syntax.
gn help --markdown all Dump all help to stdout in markdown format.
[--type=...] [--testonly=...] Lists all targets matching the given pattern for the given build directory. By default, only targets in the default toolchain will be matched unless a toolchain is explicitly supplied. If the label pattern is unspecified, list all targets. The label pattern is not a general regular expression (see "gn help label_pattern"). If you need more complex expressions, pipe the result through grep.
--as=(buildfile|label|output) How to print targets. buildfile Prints the build files where the given target was declared as file names. label (default) Prints the label of the target. output Prints the first output file for the target relative to the root build directory. --default-toolchain Normally wildcard targets are matched in all toolchains. This switch makes wildcard labels with no explicit toolchain reference only match targets in the default toolchain. Non-wildcard inputs with no explicit toolchain specification will always match only a target in the default toolchain if one exists. --testonly=(true|false) Restrict outputs to targets with the testonly flag set accordingly. When unspecified, the target's testonly flags are ignored. --type=(action|copy|executable|group|loadable_module|shared_library| source_set|static_library) Restrict outputs to targets matching the given type. If unspecified, no filtering will be performed.
gn ls out/Debug Lists all targets in the default toolchain. gn ls out/Debug "//base/*" Lists all targets in the directory base and all subdirectories. gn ls out/Debug "//base:*" Lists all targets defined in //base/BUILD.gn. gn ls out/Debug //base --as=output Lists the build output file for //base:base gn ls out/Debug --type=executable Lists all executables produced by the build. gn ls out/Debug "//base/*" --as=output | xargs ninja -C out/Debug Builds all targets in //base and all subdirectories.
gn meta <out_dir> <target>* --data=<key>[,<key>*]* [--walk=<key>[,<key>*]*] [--rebase=<dest dir>] Lists collected metaresults of all given targets for the given data key(s), collecting metadata dependencies as specified by the given walk key(s). See `gn help generated_file` for more information on the walk.
<target(s)> A list of target labels from which to initiate the walk. --data A list of keys from which to extract data. In each target walked, its metadata scope is checked for the presence of these keys. If present, the contents of those variable in the scope are appended to the results list. --walk (optional) A list of keys from which to control the walk. In each target walked, its metadata scope is checked for the presence of any of these keys. If present, the contents of those variables is checked to ensure that it is a label of a valid dependency of the target and then added to the set of targets to walk. If the empty string ("") is present in any of these keys, all deps and data_deps are added to the walk set. --rebase (optional) A destination directory onto which to rebase any paths found. If set, all collected metadata will be rebased onto this path. This option will throw errors if collected metadata is not a list of strings.
gn meta out/Debug "//base/foo" --data=files Lists collected metaresults for the `files` key in the //base/foo:foo target and all of its dependency tree. gn meta out/Debug "//base/foo" --data=files --data=other Lists collected metaresults for the `files` and `other` keys in the //base/foo:foo target and all of its dependency tree. gn meta out/Debug "//base/foo" --data=files --walk=stop Lists collected metaresults for the `files` key in the //base/foo:foo target and all of the dependencies listed in the `stop` key (and so on). gn meta out/Debug "//base/foo" --data=files --rebase="/" Lists collected metaresults for the `files` key in the //base/foo:foo target and all of its dependency tree, rebasing the strings in the `files` key onto the source directory of the target's declaration relative to "/".
Lists the output files corresponding to the given target(s) or file name(s). There can be multiple outputs because there can be more than one output generated by a build step, and there can be more than one toolchain matched. You can also list multiple inputs which will generate a union of all the outputs from those inputs. - The input target/file names are relative to the current directory. - The output file names are relative to the root build directory. This command is useful for finding a ninja command that will build only a portion of the build.
If the parameter is a target name that includes a toolchain, it will match only that target in that toolchain. If no toolchain is specified, it will match all targets with that name in any toolchain. The result will be the outputs specified by that target which could be a library, executable, output of an action, a stamp file, etc.
If the parameter is a file name it will compute the output for that compile step for all targets in all toolchains that contain that file as a source file. If the source is not compiled (e.g. a header or text file), the command will produce no output. If the source is listed as an "input" to a binary target or action will resolve to that target's outputs.
gn outputs out/debug some/directory:some_target Find the outputs of a given target. gn outputs out/debug src/project/my_file.cc | xargs ninja -C out/debug Compiles just the given source file in all toolchains it's referenced in. git diff --name-only | xargs gn outputs out/x64 | xargs ninja -C out/x64 Compiles all files changed in git.
Finds paths of dependencies between two targets. Each unique path will be printed in one group, and groups will be separate by newlines. The two targets can appear in either order (paths will be found going in either direction). By default, a single path will be printed. If there is a path with only public dependencies, the shortest public path will be printed. Otherwise, the shortest path using either public or private dependencies will be printed. If --with-data is specified, data deps will also be considered. If there are multiple shortest paths, an arbitrary one will be selected.
In a large project, there can be 100's of millions of unique paths between a very high level and a common low-level target. To make the output more useful (and terminate in a reasonable time), GN will not revisit sub-paths previously known to lead to the target.
--all Prints all "interesting" paths found rather than just the first one. Public paths will be printed first in order of increasing length, followed by non-public paths in order of increasing length. --public Considers only public paths. Can't be used with --with-data. --with-data Additionally follows data deps. Without this flag, only public and private linked deps will be followed. Can't be used with --public.
gn path out/Default //base //gn
gn refs <out_dir> (<label_pattern>|<label>|<file>|@<response_file>)* [--all] [--default-toolchain] [--as=...] [--testonly=...] [--type=...] Finds reverse dependencies (which targets reference something). The input is a list containing: - Target label: The result will be which targets depend on it. - Config label: The result will be which targets list the given config in its "configs" or "public_configs" list. - Label pattern: The result will be which targets depend on any target matching the given pattern. Patterns will not match configs. These are not general regular expressions, see "gn help label_pattern" for details. - File name: The result will be which targets list the given file in its "inputs", "sources", "public", "data", or "outputs". Any input that does not contain wildcards and does not match a target or a config will be treated as a file. - Response file: If the input starts with an "@", it will be interpreted as a path to a file containing a list of labels or file names, one per line. This allows us to handle long lists of inputs without worrying about command line limits.
--all When used without --tree, will recurse and display all unique dependencies of the given targets. For example, if the input is a target, this will output all targets that depend directly or indirectly on the input. If the input is a file, this will output all targets that depend directly or indirectly on that file. When used with --tree, turns off eliding to show a complete tree. --as=(buildfile|label|output) How to print targets. buildfile Prints the build files where the given target was declared as file names. label (default) Prints the label of the target. output Prints the first output file for the target relative to the root build directory. --default-toolchain Normally wildcard targets are matched in all toolchains. This switch makes wildcard labels with no explicit toolchain reference only match targets in the default toolchain. Non-wildcard inputs with no explicit toolchain specification will always match only a target in the default toolchain if one exists. -q Quiet. If nothing matches, don't print any output. Without this option, if there are no matches there will be an informational message printed which might interfere with scripts processing the output. --testonly=(true|false) Restrict outputs to targets with the testonly flag set accordingly. When unspecified, the target's testonly flags are ignored. --tree Outputs a reverse dependency tree from the given target. Duplicates will be elided. Combine with --all to see a full dependency tree. Tree output can not be used with the filtering or output flags: --as, --type, --testonly. --type=(action|copy|executable|group|loadable_module|shared_library| source_set|static_library) Restrict outputs to targets matching the given type. If unspecified, no filtering will be performed.
gn refs out/Debug //gn:gn Find all targets depending on the given exact target name. gn refs out/Debug //base:i18n --as=buildfiles | xargs gvim Edit all .gn files containing references to //base:i18n gn refs out/Debug //base --all List all targets depending directly or indirectly on //base:base. gn refs out/Debug "//base/*" List all targets depending directly on any target in //base or its subdirectories. gn refs out/Debug "//base:*" List all targets depending directly on any target in //base/BUILD.gn. gn refs out/Debug //base --tree Print a reverse dependency tree of //base:base
gn refs out/Debug //base/macros.h Print target(s) listing //base/macros.h as a source. gn refs out/Debug //base/macros.h --tree Display a reverse dependency tree to get to the given file. This will show how dependencies will reference that file. gn refs out/Debug //base/macros.h //base/at_exit.h --all Display all unique targets with some dependency path to a target containing either of the given files as a source. gn refs out/Debug //base/macros.h --testonly=true --type=executable --all --as=output Display the executable file names of all test executables potentially affected by a change to the given file.
This target type allows you to run a script a single time to produce one or more output files. If you want to run a script once for each of a set of input files, see "gn help action_foreach".
In an action the "sources" and "inputs" are treated the same: they're both input dependencies on script execution with no special handling. If you want to pass the sources to your script, you must do so explicitly by including them in the "args". Note also that this means there is no special handling of paths since GN doesn't know which of the args are paths and not. You will want to use rebase_path() to convert paths to be relative to the root_build_dir. You can dynamically write input dependencies (for incremental rebuilds if an input file changes) by writing a depfile when the script is run (see "gn help depfile"). This is more flexible than "inputs". If the command line length is very long, you can use response files to pass args to your script. See "gn help response_file_contents". It is recommended you put inputs to your script in the "sources" variable, and stuff like other Python files required to run your script in the "inputs" variable. The "deps" and "public_deps" for an action will always be completed before any part of the action is run so it can depend on the output of previous steps. The "data_deps" will be built if the action is built, but may not have completed before all steps of the action are started. This can give additional parallelism in the build for runtime-only dependencies.
You should specify files created by your script by specifying them in the "outputs". The script will be executed with the given arguments with the current directory being that of the root build directory. If you pass files to your script, see "gn help rebase_path" for how to convert file names to be relative to the build directory (file names in the sources, outputs, and inputs will be all treated as relative to the current build file and converted as needed automatically). GN sets Ninja's flag 'restat = 1` for all action commands. This means that Ninja will check the timestamp of the output after the action completes. If output timestamp is unchanged, the step will be treated as if it never needed to be rebuilt, potentially eliminating some downstream steps for incremental builds. Scripts can improve build performance by taking care not to change the timstamp of the output file(s) if the contents have not changed.
All output files must be inside the output directory of the build. You would generally use |$target_out_dir| or |$target_gen_dir| to reference the output or generated intermediate file directories, respectively.
args, data, data_deps, depfile, deps, inputs, metadata, outputs*, pool, response_file_contents, script*, sources * = required
action("run_this_guy_once") { script = "doprocessing.py" sources = [ "my_configuration.txt" ] outputs = [ "$target_gen_dir/insightful_output.txt" ] # Our script imports this Python file so we want to rebuild if it changes. inputs = [ "helper_library.py" ] # Note that we have to manually pass the sources to our script if the # script needs them as inputs. args = [ "--out", rebase_path(target_gen_dir, root_build_dir) ] + rebase_path(sources, root_build_dir) }
This target type allows you to run a script once-per-file over a set of sources. If you want to run a script once that takes many files as input, see "gn help action".
The script will be run once per file in the "sources" variable. The "outputs" variable should specify one or more files with a source expansion pattern in it (see "gn help source_expansion"). The output file(s) for each script invocation should be unique. Normally you use "{{source_name_part}}" in each output file. If your script takes additional data as input, such as a shared configuration file or a Python module it uses, those files should be listed in the "inputs" variable. These files are treated as dependencies of each script invocation. If the command line length is very long, you can use response files to pass args to your script. See "gn help response_file_contents". You can dynamically write input dependencies (for incremental rebuilds if an input file changes) by writing a depfile when the script is run (see "gn help depfile"). This is more flexible than "inputs". The "deps" and "public_deps" for an action will always be completed before any part of the action is run so it can depend on the output of previous steps. The "data_deps" will be built if the action is built, but may not have completed before all steps of the action are started. This can give additional parallelism in the build for runtime-only dependencies.
The script will be executed with the given arguments with the current directory being that of the root build directory. If you pass files to your script, see "gn help rebase_path" for how to convert file names to be relative to the build directory (file names in the sources, outputs, and inputs will be all treated as relative to the current build file and converted as needed automatically). GN sets Ninja's flag 'restat = 1` for all action commands. This means that Ninja will check the timestamp of the output after the action completes. If output timestamp is unchanged, the step will be treated as if it never needed to be rebuilt, potentially eliminating some downstream steps for incremental builds. Scripts can improve build performance by taking care not to change the timstamp of the output file(s) if the contents have not changed.
All output files must be inside the output directory of the build. You would generally use |$target_out_dir| or |$target_gen_dir| to reference the output or generated intermediate file directories, respectively.
args, data, data_deps, depfile, deps, inputs, metadata, outputs*, pool, response_file_contents, script*, sources* * = required
# Runs the script over each IDL file. The IDL script will generate both a .cc # and a .h file for each input. action_foreach("my_idl") { script = "idl_processor.py" sources = [ "foo.idl", "bar.idl" ] # Our script reads this file each time, so we need to list it as a # dependency so we can rebuild if it changes. inputs = [ "my_configuration.txt" ] # Transformation from source file name to output file names. outputs = [ "$target_gen_dir/{{source_name_part}}.h", "$target_gen_dir/{{source_name_part}}.cc" ] # Note that since "args" is opaque to GN, if you specify paths here, you # will need to convert it to be relative to the build directory using # rebase_path(). args = [ "{{source}}", "-o", rebase_path(relative_target_gen_dir, root_build_dir) + "/{{source_name_part}}.h" ] }
This target type allows to declare data that is required at runtime. It is used to inform "create_bundle" targets of the files to copy into generated bundle, see "gn help create_bundle" for help. The target must define a list of files as "sources" and a single "outputs". If there are multiple files, source expansions must be used to express the output. The output must reference a file inside of {{bundle_root_dir}}. This target can be used on all platforms though it is designed only to generate iOS/macOS bundle. In cross-platform projects, it is advised to put it behind iOS/macOS conditionals. See "gn help create_bundle" for more information.
sources*, outputs*, deps, data_deps, metadata, public_deps, visibility * = required
bundle_data("icudata") { sources = [ "sources/data/in/icudtl.dat" ] outputs = [ "{{bundle_resources_dir}}/{{source_file_part}}" ] } bundle_data("base_unittests_bundle_data]") { sources = [ "test/data" ] outputs = [ "{{bundle_resources_dir}}/{{source_root_relative_dir}}/" + "{{source_file_part}}" ] } bundle_data("material_typography_bundle_data") { sources = [ "src/MaterialTypography.bundle/Roboto-Bold.ttf", "src/MaterialTypography.bundle/Roboto-Italic.ttf", "src/MaterialTypography.bundle/Roboto-Regular.ttf", "src/MaterialTypography.bundle/Roboto-Thin.ttf", ] outputs = [ "{{bundle_resources_dir}}/MaterialTypography.bundle/" "{{source_file_part}}" ] }
All output files must be inside the output directory of the build. You would generally use |$target_out_dir| or |$target_gen_dir| to reference the output or generated intermediate file directories, respectively. Both "sources" and "outputs" must be specified. Sources can include as many files as you want, but there can only be one item in the outputs list (plural is used for the name for consistency with other target types). If there is more than one source file, your output name should specify a mapping from each source file to an output file name using source expansion (see "gn help source_expansion"). The placeholders will look like "{{source_name_part}}", for example.
# Write a rule that copies a checked-in DLL to the output directory. copy("mydll") { sources = [ "mydll.dll" ] outputs = [ "$target_out_dir/mydll.dll" ] } # Write a rule to copy several files to the target generated files directory. copy("myfiles") { sources = [ "data1.dat", "data2.dat", "data3.dat" ] # Use source expansion to generate output files with the corresponding file # names in the gen dir. This will just copy each file. outputs = [ "$target_gen_dir/{{source_file_part}}" ] }
This target generates an iOS or macOS bundle (which is a directory with a well-know structure). This target does not define any sources, instead they are computed from all "bundle_data" target this one depends on transitively (the recursion stops at "create_bundle" targets). The "bundle_*_dir" are be used for the expansion of {{bundle_*_dir}} rules in "bundle_data" outputs. The properties are optional but must be defined if any of the "bundle_data" target use them. This target can be used on all platforms though it is designed only to generate iOS or macOS bundle. In cross-platform projects, it is advised to put it behind iOS/macOS conditionals. If a create_bundle is specified as a data_deps for another target, the bundle is considered a leaf, and its public and private dependencies will not contribute to any data or data_deps. Required runtime dependencies should be placed in the bundle. A create_bundle can declare its own explicit data and data_deps, however.
Some bundle needs to be code signed as part of the build (on iOS all application needs to be code signed to run on a device). The code signature can be configured via the code_signing_script variable. If set, code_signing_script is the path of a script that invoked after all files have been moved into the bundle. The script must not change any file in the bundle, but may add new files. If code_signing_script is defined, then code_signing_outputs must also be defined and non-empty to inform when the script needs to be re-run. The code_signing_args will be passed as is to the script (so path have to be rebased) and additional inputs may be listed with the variable code_signing_sources.
bundle_root_dir, bundle_contents_dir, bundle_resources_dir, bundle_executable_dir, bundle_deps_filter, deps, data_deps, public_deps, visibility, product_type, code_signing_args, code_signing_script, code_signing_sources, code_signing_outputs, xcode_extra_attributes, xcode_test_application_name, partial_info_plist, metadata
# Defines a template to create an application. On most platform, this is just # an alias for an "executable" target, but on iOS/macOS, it builds an # application bundle. template("app") { if (!is_ios && !is_mac) { executable(target_name) { forward_variables_from(invoker, "*") } } else { app_name = target_name gen_path = target_gen_dir action("${app_name}_generate_info_plist") { script = [ "//build/ios/ios_gen_plist.py" ] sources = [ "templates/Info.plist" ] outputs = [ "$gen_path/Info.plist" ] args = rebase_path(sources, root_build_dir) + rebase_path(outputs, root_build_dir) } bundle_data("${app_name}_bundle_info_plist") { public_deps = [ ":${app_name}_generate_info_plist" ] sources = [ "$gen_path/Info.plist" ] outputs = [ "{{bundle_contents_dir}}/Info.plist" ] } executable("${app_name}_generate_executable") { forward_variables_from(invoker, "*", [ "output_name", "visibility", ]) output_name = rebase_path("$gen_path/$app_name", root_build_dir) } code_signing = defined(invoker.code_signing) && invoker.code_signing if (!is_ios || !code_signing) { bundle_data("${app_name}_bundle_executable") { public_deps = [ ":${app_name}_generate_executable" ] sources = [ "$gen_path/$app_name" ] outputs = [ "{{bundle_executable_dir}}/$app_name" ] } } create_bundle("$app_name.app") { product_type = "com.apple.product-type.application" if (is_ios) { bundle_root_dir = "$root_build_dir/$target_name" bundle_contents_dir = bundle_root_dir bundle_resources_dir = bundle_contents_dir bundle_executable_dir = bundle_contents_dir xcode_extra_attributes = { ONLY_ACTIVE_ARCH = "YES" DEBUG_INFORMATION_FORMAT = "dwarf" } } else { bundle_root_dir = "$root_build_dir/$target_name" bundle_contents_dir = "$bundle_root_dir/Contents" bundle_resources_dir = "$bundle_contents_dir/Resources" bundle_executable_dir = "$bundle_contents_dir/MacOS" } deps = [ ":${app_name}_bundle_info_plist" ] if (is_ios && code_signing) { deps += [ ":${app_name}_generate_executable" ] code_signing_script = "//build/config/ios/codesign.py" code_signing_sources = [ invoker.entitlements_path, "$target_gen_dir/$app_name", ] code_signing_outputs = [ "$bundle_root_dir/$app_name", "$bundle_root_dir/_CodeSignature/CodeResources", "$bundle_root_dir/embedded.mobileprovision", "$target_gen_dir/$app_name.xcent", ] code_signing_args = [ "-i=" + ios_code_signing_identity, "-b=" + rebase_path( "$target_gen_dir/$app_name", root_build_dir), "-e=" + rebase_path( invoker.entitlements_path, root_build_dir), "-e=" + rebase_path( "$target_gen_dir/$app_name.xcent", root_build_dir), rebase_path(bundle_root_dir, root_build_dir), ] } else { deps += [ ":${app_name}_bundle_executable" ] } } } }
The tools and commands used to create this target type will be determined by the source files in its sources. Targets containing multiple compiler-incompatible languages are not allowed (e.g. a target containing both C and C++ sources is acceptable, but a target containing C and Rust sources is not).
Flags: cflags, cflags_c, cflags_cc, cflags_objc, cflags_objcc, asmflags, defines, include_dirs, inputs, ldflags, lib_dirs, libs, precompiled_header, precompiled_source, rustflags, rustenv, swiftflags Deps: data_deps, deps, public_deps Dependent configs: all_dependent_configs, public_configs General: check_includes, configs, data, friend, inputs, metadata, output_name, output_extension, public, sources, testonly, visibility Rust variables: aliased_deps, crate_root, crate_name
Writes data value(s) to disk on resolution. This target type mirrors some functionality of the write_file() function, but also provides the ability to collect metadata from its dependencies on resolution rather than writing out at parse time. The `outputs` variable is required to be a list with a single element, specifying the intended location of the output file. The `output_conversion` variable specified the format to write the value. See `gn help output_conversion`. One of `contents` or `data_keys` must be specified; use of `data` will write the contents of that value to file, while use of `data_keys` will trigger a metadata collection walk based on the dependencies of the target and the optional values of the `rebase` and `walk_keys` variables. See `gn help metadata`. Collected metadata, if specified, will be returned in postorder of dependencies. See the example for details.
Given the following targets defined in //base/BUILD.gn, where A depends on B and B depends on C and D: group("a") { metadata = { doom_melon = [ "enable" ] my_files = [ "foo.cpp" ] # Note: this is functionally equivalent to not defining `my_barrier` # at all in this target's metadata. my_barrier = [ "" ] } deps = [ ":b" ] } group("b") { metadata = { my_files = [ "bar.cpp" ] my_barrier = [ ":c" ] } deps = [ ":c", ":d" ] } group("c") { metadata = { doom_melon = [ "disable" ] my_files = [ "baz.cpp" ] } } group("d") { metadata = { my_files = [ "missing.cpp" ] } } If the following generated_file target is defined: generated_file("my_files_metadata") { outputs = [ "$root_build_dir/my_files.json" ] data_keys = [ "my_files" ] deps = [ "//base:a" ] } The following will be written to "$root_build_dir/my_files.json" (less the comments): [ "baz.cpp", // from //base:c via //base:b "missing.cpp" // from //base:d via //base:b "bar.cpp", // from //base:b via //base:a "foo.cpp", // from //base:a ] Alternatively, as an example of using walk_keys, if the following generated_file target is defined: generated_file("my_files_metadata") { outputs = [ "$root_build_dir/my_files.json" ] data_keys = [ "my_files" ] walk_keys = [ "my_barrier" ] deps = [ "//base:a" ] } The following will be written to "$root_build_dir/my_files.json" (again less the comments): [ "baz.cpp", // from //base:c via //base:b "bar.cpp", // from //base:b via //base:a "foo.cpp", // from //base:a ] If `rebase` is used in the following generated_file target: generated_file("my_files_metadata") { outputs = [ "$root_build_dir/my_files.json" ] data_keys = [ "my_files" ] walk_keys = [ "my_barrier" ] rebase = root_build_dir deps = [ "//base:a" ] } The following will be written to "$root_build_dir/my_files.json" (again less the comments) (assuming root_build_dir = "//out"): [ "../base/baz.cpp", // from //base:c via //base:b "../base/bar.cpp", // from //base:b via //base:a "../base/foo.cpp", // from //base:a ]
contents data_keys rebase walk_keys output_conversion Deps: data_deps, deps, public_deps Dependent configs: all_dependent_configs, public_configs
This target type allows you to create meta-targets that just collect a set of dependencies into one named target. Groups can additionally specify configs that apply to their dependents.
Deps: data_deps, deps, public_deps Dependent configs: all_dependent_configs, public_configs
group("all") { deps = [ "//project:runner", "//project:unit_tests", ] }
This target type allows you to create an object file that is (and can only be) loaded and unloaded at runtime. A loadable module will be specified on the linker line for targets listing the loadable module in its "deps". If you don't want this (if you don't need to dynamically load the library at runtime), then you should use a "shared_library" target type instead.
The tools and commands used to create this target type will be determined by the source files in its sources. Targets containing multiple compiler-incompatible languages are not allowed (e.g. a target containing both C and C++ sources is acceptable, but a target containing C and Rust sources is not).
Flags: cflags, cflags_c, cflags_cc, cflags_objc, cflags_objcc, asmflags, defines, include_dirs, inputs, ldflags, lib_dirs, libs, precompiled_header, precompiled_source, rustflags, rustenv, swiftflags Deps: data_deps, deps, public_deps Dependent configs: all_dependent_configs, public_configs General: check_includes, configs, data, friend, inputs, metadata, output_name, output_extension, public, sources, testonly, visibility Rust variables: aliased_deps, crate_root, crate_name, crate_type
A Rust library is an archive containing additional rust-c provided metadata. These are the files produced by the rustc compiler with the `.rlib` extension, and are the intermediate step for most Rust-based binaries.
The tools and commands used to create this target type will be determined by the source files in its sources. Targets containing multiple compiler-incompatible languages are not allowed (e.g. a target containing both C and C++ sources is acceptable, but a target containing C and Rust sources is not).
Flags: cflags, cflags_c, cflags_cc, cflags_objc, cflags_objcc, asmflags, defines, include_dirs, inputs, ldflags, lib_dirs, libs, precompiled_header, precompiled_source, rustflags, rustenv, swiftflags Deps: data_deps, deps, public_deps Dependent configs: all_dependent_configs, public_configs General: check_includes, configs, data, friend, inputs, metadata, output_name, output_extension, public, sources, testonly, visibility Rust variables: aliased_deps, crate_root, crate_name
A Rust procedural macro allows creating syntax extensions as execution of a function. They are compiled as dynamic libraries and used by the compiler at runtime. Their use is the same as of other Rust libraries, but their build has some additional restrictions in terms of supported flags.
The tools and commands used to create this target type will be determined by the source files in its sources. Targets containing multiple compiler-incompatible languages are not allowed (e.g. a target containing both C and C++ sources is acceptable, but a target containing C and Rust sources is not).
Flags: cflags, cflags_c, cflags_cc, cflags_objc, cflags_objcc, asmflags, defines, include_dirs, inputs, ldflags, lib_dirs, libs, precompiled_header, precompiled_source, rustflags, rustenv, swiftflags Deps: data_deps, deps, public_deps Dependent configs: all_dependent_configs, public_configs General: check_includes, configs, data, friend, inputs, metadata, output_name, output_extension, public, sources, testonly, visibility Rust variables: aliased_deps, crate_root, crate_name
A shared library will be specified on the linker line for targets listing the shared library in its "deps". If you don't want this (say you dynamically load the library at runtime), then you should depend on the shared library via "data_deps" or, on Darwin platforms, use a "loadable_module" target type instead.
The tools and commands used to create this target type will be determined by the source files in its sources. Targets containing multiple compiler-incompatible languages are not allowed (e.g. a target containing both C and C++ sources is acceptable, but a target containing C and Rust sources is not).
Flags: cflags, cflags_c, cflags_cc, cflags_objc, cflags_objcc, asmflags, defines, include_dirs, inputs, ldflags, lib_dirs, libs, precompiled_header, precompiled_source, rustflags, rustenv, swiftflags Deps: data_deps, deps, public_deps Dependent configs: all_dependent_configs, public_configs General: check_includes, configs, data, friend, inputs, metadata, output_name, output_extension, public, sources, testonly, visibility Rust variables: aliased_deps, crate_root, crate_name, crate_type
Only C-language source sets are supported at the moment.
A source set is a collection of sources that get compiled, but are not linked to produce any kind of library. Instead, the resulting object files are implicitly added to the linker line of all targets that depend on the source set. In most cases, a source set will behave like a static library, except no actual library file will be produced. This will make the build go a little faster by skipping creation of a large static library, while maintaining the organizational benefits of focused build targets. The main difference between a source set and a static library is around handling of exported symbols. Most linkers assume declaring a function exported means exported from the static library. The linker can then do dead code elimination to delete code not reachable from exported functions. A source set will not do this code elimination since there is no link step. This allows you to link many source sets into a shared library and have the "exported symbol" notation indicate "export from the final shared library and not from the intermediate targets." There is no way to express this concept when linking multiple static libraries into a shared library.
Flags: cflags, cflags_c, cflags_cc, cflags_objc, cflags_objcc, asmflags, defines, include_dirs, inputs, ldflags, lib_dirs, libs, precompiled_header, precompiled_source, rustflags, rustenv, swiftflags Deps: data_deps, deps, public_deps Dependent configs: all_dependent_configs, public_configs General: check_includes, configs, data, friend, inputs, metadata, output_name, output_extension, public, sources, testonly, visibility
Make a ".a" / ".lib" file. If you only need the static library for intermediate results in the build, you should consider a source_set instead since it will skip the (potentially slow) step of creating the intermediate library file.
complete_static_lib Flags: cflags, cflags_c, cflags_cc, cflags_objc, cflags_objcc, asmflags, defines, include_dirs, inputs, ldflags, lib_dirs, libs, precompiled_header, precompiled_source, rustflags, rustenv, swiftflags Deps: data_deps, deps, public_deps Dependent configs: all_dependent_configs, public_configs General: check_includes, configs, data, friend, inputs, metadata, output_name, output_extension, public, sources, testonly, visibility Rust variables: aliased_deps, crate_root, crate_name The tools and commands used to create this target type will be determined by the source files in its sources. Targets containing multiple compiler-incompatible languages are not allowed (e.g. a target containing both C and C++ sources is acceptable, but a target containing C and Rust sources is not).
target(target_type_string, target_name_string) { ... } The target() function is a way to invoke a built-in target or template with a type determined at runtime. This is useful for cases where the type of a target might not be known statically. Only templates and built-in target functions are supported for the target_type_string parameter. Arbitrary functions, configs, and toolchains are not supported. The call: target("source_set", "doom_melon") { Is equivalent to: source_set("doom_melon") {
if (foo_build_as_shared) { my_type = "shared_library" } else { my_type = "source_set" } target(my_type, "foo") { ... }
assert(<condition> [, <error string>]) If the condition is false, the build will fail with an error. If the optional second argument is provided, that string will be printed with the error message.
assert(is_win) assert(defined(sources), "Sources must be defined");
Configuration objects can be applied to targets and specify sets of compiler flags, includes, defines, etc. They provide a way to conveniently group sets of this configuration information. A config is referenced by its label just like a target. The values in a config are additive only. If you want to remove a flag you need to remove the corresponding config that sets it. The final set of flags, defines, etc. for a target is generated in this order: 1. The values specified directly on the target (rather than using a config. 2. The configs specified in the target's "configs" list, in order. 3. Public_configs from a breadth-first traversal of the dependency tree in the order that the targets appear in "deps". 4. All dependent configs from a breadth-first traversal of the dependency tree in the order that the targets appear in "deps".
Configs solve a problem where the build system needs to have a higher-level understanding of various compiler settings. For example, some compiler flags have to appear in a certain order relative to each other, some settings like defines and flags logically go together, and the build system needs to de-duplicate flags even though raw command-line parameters can't always be operated on in that way. The config gives a name to a group of settings that can then be reasoned about by GN. GN can know that configs with the same label are the same thing so can be de-duplicated. It allows related settings to be grouped so they are added or removed as a unit. And it allows targets to refer to settings with conceptual names ("no_rtti", "enable_exceptions", etc.) rather than having to hard-coding every compiler's flags each time they are referred to.
Flags: cflags, cflags_c, cflags_cc, cflags_objc, cflags_objcc, asmflags, defines, include_dirs, inputs, ldflags, lib_dirs, libs, precompiled_header, precompiled_source, rustflags, rustenv, swiftflags Nested configs: configs
all_dependent_configs, configs, public_configs
config("myconfig") { include_dirs = [ "include/common" ] defines = [ "ENABLE_DOOM_MELON" ] } executable("mything") { configs = [ ":myconfig" ] }
Introduces the given arguments into the current scope. If they are not specified on the command line or in a toolchain's arguments, the default values given in the declare_args block will be used. However, these defaults will not override command-line values. See also "gn help buildargs" for an overview. The precise behavior of declare args is: 1. The declare_args() block executes. Any variable defined in the enclosing scope is available for reading, but any variable defined earlier in the current scope is not (since the overrides haven't been applied yet). 2. At the end of executing the block, any variables set within that scope are saved, with the values specified in the block used as the "default value" for that argument. Once saved, these variables are available for override via args.gn. 3. User-defined overrides are applied. Anything set in "gn args" now overrides any default values. The resulting set of variables is promoted to be readable from the following code in the file. This has some ramifications that may not be obvious: - You should not perform difficult work inside a declare_args block since this only sets a default value that may be discarded. In particular, don't use the result of exec_script() to set the default value. If you want to have a script-defined default, set some default "undefined" value like [], "", or -1, and after the declare_args block, call exec_script if the value is unset by the user. - Because you cannot read the value of a variable defined in the same block, if you need to make the default value of one arg depend on the possibly-overridden value of another, write two separate declare_args() blocks: declare_args() { enable_foo = true } declare_args() { # Bar defaults to same user-overridden state as foo. enable_bar = enable_foo }
declare_args() { enable_teleporter = true enable_doom_melon = false } If you want to override the (default disabled) Doom Melon: gn --args="enable_doom_melon=true enable_teleporter=true" This also sets the teleporter, but it's already defaulted to on so it will have no effect.
Returns true if the given argument is defined. This is most useful in templates to assert that the caller set things up properly. You can pass an identifier: defined(foo) which will return true or false depending on whether foo is defined in the current scope. You can also check a named scope: defined(foo.bar) which will return true or false depending on whether bar is defined in the named scope foo. It will throw an error if foo is not defined or is not a scope.
template("mytemplate") { # To help users call this template properly... assert(defined(invoker.sources), "Sources must be defined") # If we want to accept an optional "values" argument, we don't # want to dereference something that may not be defined. if (defined(invoker.values)) { values = invoker.values } else { values = "some default value" } }
exec_script(filename, arguments = [], input_conversion = "", file_dependencies = []) Runs the given script, returning the stdout of the script. The build generation will fail if the script does not exist or returns a nonzero exit code. The current directory when executing the script will be the root build directory. If you are passing file names, you will want to use the rebase_path() function to make file names relative to this path (see "gn help rebase_path"). The default script interpreter is Python ("python" on POSIX, "python.exe" or "python.bat" on Windows). This can be configured by the script_executable variable, see "gn help dotfile".
filename: File name of script to execute. Non-absolute names will be treated as relative to the current build file. arguments: A list of strings to be passed to the script as arguments. May be unspecified or the empty list which means no arguments. input_conversion: Controls how the file is read and parsed. See "gn help io_conversion". If unspecified, defaults to the empty string which causes the script result to be discarded. exec script will return None. dependencies: (Optional) A list of files that this script reads or otherwise depends on. These dependencies will be added to the build result such that if any of them change, the build will be regenerated and the script will be re-run. The script itself will be an implicit dependency so you do not need to list it.
all_lines = exec_script( "myscript.py", [some_input], "list lines", [ rebase_path("data_file.txt", root_build_dir) ]) # This example just calls the script with no arguments and discards the # result. exec_script("//foo/bar/myscript.py")
filter_exclude(values, exclude_patterns) The argument values must be a list of strings. The argument exclude_patterns must be a list of file patterns (see "gn help file_pattern"). Any elements in values matching at least one of those patterns will be excluded.
values = [ "foo.cc", "foo.h", "foo.proto" ] result = filter_exclude(values, [ "*.proto" ]) # result will be [ "foo.cc", "foo.h" ]
filter_include(values, include_patterns) The argument values must be a list of strings. The argument include_patterns must be a list of file patterns (see "gn help file_pattern"). Only elements from values matching at least one of the pattern will be included.
values = [ "foo.cc", "foo.h", "foo.proto" ] result = filter_include(values, [ "*.proto" ]) # result will be [ "foo.proto" ]
foreach(<loop_var>, <list>) { <loop contents> } Executes the loop contents block over each item in the list, assigning the loop_var to each item in sequence. The <loop_var> will be a copy so assigning to it will not mutate the list. The loop will iterate over a copy of <list> so mutating it inside the loop will not affect iteration. The block does not introduce a new scope, so that variable assignments inside the loop will be visible once the loop terminates. The loop variable will temporarily shadow any existing variables with the same name for the duration of the loop. After the loop terminates the loop variable will no longer be in scope, and the previous value (if any) will be restored.
mylist = [ "a", "b", "c" ] foreach(i, mylist) { print(i) } Prints: a b c
forward_variables_from(from_scope, variable_list_or_star, variable_to_not_forward_list = []) Copies the given variables from the given scope to the local scope if they exist. This is normally used in the context of templates to use the values of variables defined in the template invocation to a template-defined target. The variables in the given variable_list will be copied if they exist in the given scope or any enclosing scope. If they do not exist, nothing will happen and they be left undefined in the current scope. As a special case, if the variable_list is a string with the value of "*", all variables from the given scope will be copied. "*" only copies variables set directly on the from_scope, not enclosing ones. Otherwise it would duplicate all global variables. When an explicit list of variables is supplied, if the variable exists in the current (destination) scope already, an error will be thrown. If "*" is specified, variables in the current scope will be clobbered (the latter is important because most targets have an implicit configs list, which means it wouldn't work at all if it didn't clobber). If variables_to_not_forward_list is non-empty, then it must contains a list of variable names that will not be forwarded. This is mostly useful when variable_list_or_star has a value of "*".
# forward_variables_from(invoker, ["foo"]) # is equivalent to: assert(!defined(foo)) if (defined(invoker.foo)) { foo = invoker.foo } # This is a common action template. It would invoke a script with some given # parameters, and wants to use the various types of deps and the visibility # from the invoker if it's defined. It also injects an additional dependency # to all targets. template("my_test") { action(target_name) { forward_variables_from(invoker, [ "data_deps", "deps", "public_deps", "visibility"]) # Add our test code to the dependencies. # "deps" may or may not be defined at this point. if (defined(deps)) { deps += [ "//tools/doom_melon" ] } else { deps = [ "//tools/doom_melon" ] } } } # This is a template around a target whose type depends on a global variable. # It forwards all values from the invoker. template("my_wrapper") { target(my_wrapper_target_type, target_name) { forward_variables_from(invoker, "*") } } # A template that wraps another. It adds behavior based on one # variable, and forwards all others to the nested target. template("my_ios_test_app") { ios_test_app(target_name) { forward_variables_from(invoker, "*", ["test_bundle_name"]) if (!defined(extra_substitutions)) { extra_substitutions = [] } extra_substitutions += [ "BUNDLE_ID_TEST_NAME=$test_bundle_name" ] } }
get_label_info(target_label, what) Given the label of a target, returns some attribute of that target. The target need not have been previously defined in the same file, since none of the attributes depend on the actual target definition, only the label itself. See also "gn help get_target_outputs".
"name" The short name of the target. This will match the value of the "target_name" variable inside that target's declaration. For the label "//foo/bar:baz" this will return "baz". "dir" The directory containing the target's definition, with no slash at the end. For the label "//foo/bar:baz" this will return "//foo/bar". "target_gen_dir" The generated file directory for the target. This will match the value of the "target_gen_dir" variable when inside that target's declaration. "root_gen_dir" The root of the generated file tree for the target. This will match the value of the "root_gen_dir" variable when inside that target's declaration. "target_out_dir The output directory for the target. This will match the value of the "target_out_dir" variable when inside that target's declaration. "root_out_dir" The root of the output file tree for the target. This will match the value of the "root_out_dir" variable when inside that target's declaration. "label_no_toolchain" The fully qualified version of this label, not including the toolchain. For the input ":bar" it might return "//foo:bar". "label_with_toolchain" The fully qualified version of this label, including the toolchain. For the input ":bar" it might return "//foo:bar(//toolchain:x64)". "toolchain" The label of the toolchain. This will match the value of the "current_toolchain" variable when inside that target's declaration.
get_label_info(":foo", "name") # Returns string "foo". get_label_info("//foo/bar:baz", "target_gen_dir") # Returns string "//out/Debug/gen/foo/bar".
get_path_info(input, what) The first argument is either a string representing a file or directory name, or a list of such strings. If the input is a list the return value will be a list containing the result of applying the rule to each item in the input.
"file" The substring after the last slash in the path, including the name and extension. If the input ends in a slash, the empty string will be returned. "foo/bar.txt" => "bar.txt" "bar.txt" => "bar.txt" "foo/" => "" "" => "" "name" The substring of the file name not including the extension. "foo/bar.txt" => "bar" "foo/bar" => "bar" "foo/" => "" "extension" The substring following the last period following the last slash, or the empty string if not found. The period is not included. "foo/bar.txt" => "txt" "foo/bar" => "" "dir" The directory portion of the name, not including the slash. "foo/bar.txt" => "foo" "//foo/bar" => "//foo" "foo" => "." The result will never end in a slash, so if the resulting is empty, the system ("/") or source ("//") roots, a "." will be appended such that it is always legal to append a slash and a filename and get a valid path. "out_dir" The output file directory corresponding to the path of the given file, not including a trailing slash. "//foo/bar/baz.txt" => "//out/Default/obj/foo/bar" "gen_dir" The generated file directory corresponding to the path of the given file, not including a trailing slash. "//foo/bar/baz.txt" => "//out/Default/gen/foo/bar" "abspath" The full absolute path name to the file or directory. It will be resolved relative to the current directory, and then the source- absolute version will be returned. If the input is system- absolute, the same input will be returned. "foo/bar.txt" => "//mydir/foo/bar.txt" "foo/" => "//mydir/foo/" "//foo/bar" => "//foo/bar" (already absolute) "/usr/include" => "/usr/include" (already absolute) If you want to make the path relative to another directory, or to be system-absolute, see rebase_path().
sources = [ "foo.cc", "foo.h" ] result = get_path_info(source, "abspath") # result will be [ "//mydir/foo.cc", "//mydir/foo.h" ] result = get_path_info("//foo/bar/baz.cc", "dir") # result will be "//foo/bar" # Extract the source-absolute directory name, result = get_path_info(get_path_info(path, "dir"), "abspath")
get_target_outputs(target_label) Returns a list of output files for the named target. The named target must have been previously defined in the current file before this function is called (it can't reference targets in other files because there isn't a defined execution order, and it obviously can't reference targets that are defined after the function call). Only copy, generated_file, and action targets are supported. The outputs from binary targets will depend on the toolchain definition which won't necessarily have been loaded by the time a given line of code has run, and source sets and groups have no useful output file.
The names in the resulting list will be absolute file paths (normally like "//out/Debug/bar.exe", depending on the build directory). action, copy, and generated_file targets: this will just return the files specified in the "outputs" variable of the target. action_foreach targets: this will return the result of applying the output template to the sources (see "gn help source_expansion"). This will be the same result (though with guaranteed absolute file paths), as process_file_template will return for those inputs (see "gn help process_file_template"). source sets and groups: this will return a list containing the path of the "stamp" file that Ninja will produce once all outputs are generated. This probably isn't very useful.
# Say this action generates a bunch of C source files. action_foreach("my_action") { sources = [ ... ] outputs = [ ... ] } # Compile the resulting source files into a source set. source_set("my_lib") { sources = get_target_outputs(":my_action") }
value = getenv(env_var_name) Returns the value of the given environment variable. If the value is not found, it will try to look up the variable with the "opposite" case (based on the case of the first letter of the variable), but is otherwise case-sensitive. If the environment variable is not found, the empty string will be returned. Note: it might be nice to extend this if we had the concept of "none" in the language to indicate lookup failure.
home_dir = getenv("HOME")
The import command loads the rules and variables resulting from executing the given file into the current scope. By convention, imported files are named with a .gni extension. An import is different than a C++ "include". The imported file is executed in a standalone environment from the caller of the import command. The results of this execution are cached for other files that import the same .gni file. Note that you can not import a BUILD.gn file that's otherwise used in the build. Files must either be imported or implicitly loaded as a result of deps rules, but not both. The imported file's scope will be merged with the scope at the point import was called. If there is a conflict (both the current scope and the imported file define some variable or rule with the same name but different value), a runtime error will be thrown. Therefore, it's good practice to minimize the stuff that an imported file defines. Variables and templates beginning with an underscore '_' are considered private and will not be imported. Imported files can use such variables for internal computation without affecting other files.
import("//build/rules/idl_compilation_rule.gni") # Looks in the current directory. import("my_vars.gni")
not_needed(variable_list_or_star, variable_to_ignore_list = []) not_needed(from_scope, variable_list_or_star, variable_to_ignore_list = []) Mark the variables in the current or given scope as not needed, which means you will not get an error about unused variables for these. The variable_to_ignore_list allows excluding variables from "all matches" if variable_list_or_star is "*".
not_needed("*", [ "config" ]) not_needed([ "data_deps", "deps" ]) not_needed(invoker, "*", [ "config" ]) not_needed(invoker, [ "data_deps", "deps" ])
Pool objects can be applied to a tool to limit the parallelism of the build. This object has a single property "depth" corresponding to the number of tasks that may run simultaneously. As the file containing the pool definition may be executed in the context of more than one toolchain it is recommended to specify an explicit toolchain when defining and referencing a pool. A pool named "console" defined in the root build file represents Ninja's console pool. Targets using this pool will have access to the console's stdin and stdout, and output will not be buffered. This special pool must have a depth of 1. Pools not defined in the root must not be named "console". The console pool can only be defined for the default toolchain. Refer to the Ninja documentation on the console pool for more info. A pool is referenced by its label just like a target.
depth* * = required
if (current_toolchain == default_toolchain) { pool("link_pool") { depth = 1 } } toolchain("toolchain") { tool("link") { command = "..." pool = ":link_pool($default_toolchain)" } }
Prints all arguments to the console separated by spaces. A newline is automatically appended to the end. This function is intended for debugging. Note that build files are run in parallel so you may get interleaved prints. A buildfile may also be executed more than once in parallel in the context of different toolchains so the prints from one file may be duplicated or interleaved with itself.
print("Hello world") print(sources, deps)
process_file_template(source_list, template) process_file_template applies a template list to a source file list, returning the result of applying each template to each source. This is typically used for computing output file names from input files. In most cases, get_target_outputs() will give the same result with shorter, more maintainable code. This function should only be used when that function can't be used (like there's no target or the target is defined in another build file).
The source_list is a list of file names. The template can be a string or a list. If it is a list, multiple output strings are generated for each input. The template should contain source expansions to which each name in the source list is applied. See "gn help source_expansion".
sources = [ "foo.idl", "bar.idl", ] myoutputs = process_file_template( sources, [ "$target_gen_dir/{{source_name_part}}.cc", "$target_gen_dir/{{source_name_part}}.h" ]) The result in this case will be: [ "//out/Debug/foo.cc" "//out/Debug/foo.h" "//out/Debug/bar.cc" "//out/Debug/bar.h" ]
read_file(filename, input_conversion) Whitespace will be trimmed from the end of the file. Throws an error if the file can not be opened.
filename Filename to read, relative to the build file. input_conversion Controls how the file is read and parsed. See "gn help io_conversion".
lines = read_file("foo.txt", "list lines")
converted = rebase_path(input, new_base = "", current_base = ".") Takes a string argument representing a file name, or a list of such strings and converts it/them to be relative to a different base directory. When invoking the compiler or scripts, GN will automatically convert sources and include directories to be relative to the build directory. However, if you're passing files directly in the "args" array or doing other manual manipulations where GN doesn't know something is a file name, you will need to convert paths to be relative to what your tool is expecting. The common case is to use this to convert paths relative to the current directory to be relative to the build directory (which will be the current directory when executing scripts). If you want to convert a file path to be source-absolute (that is, beginning with a double slash like "//foo/bar"), you should use the get_path_info() function. This function won't work because it will always make relative paths, and it needs to support making paths relative to the source root, so it can't also generate source-absolute paths without more special-cases.
input A string or list of strings representing file or directory names. These can be relative paths ("foo/bar.txt"), system absolute paths ("/foo/bar.txt"), or source absolute paths ("//foo/bar.txt"). new_base The directory to convert the paths to be relative to. This can be an absolute path or a relative path (which will be treated as being relative to the current BUILD-file's directory). As a special case, if new_base is the empty string (the default), all paths will be converted to system-absolute native style paths with system path separators. This is useful for invoking external programs. current_base Directory representing the base for relative paths in the input. If this is not an absolute path, it will be treated as being relative to the current build file. Use "." (the default) to convert paths from the current BUILD-file's directory.
The return value will be the same type as the input value (either a string or a list of strings). All relative and source-absolute file names will be converted to be relative to the requested output System-absolute paths will be unchanged. Whether an output path will end in a slash will match whether the corresponding input path ends in a slash. It will return "." or "./" (depending on whether the input ends in a slash) to avoid returning empty strings. This means if you want a root path ("//" or "/") not ending in a slash, you can add a dot ("//.").
# Convert a file in the current directory to be relative to the build # directory (the current dir when executing compilers and scripts). foo = rebase_path("myfile.txt", root_build_dir) # might produce "../../project/myfile.txt". # Convert a file to be system absolute: foo = rebase_path("myfile.txt") # Might produce "D:\\source\\project\\myfile.txt" on Windows or # "/home/you/source/project/myfile.txt" on Linux. # Typical usage for converting to the build directory for a script. action("myscript") { # Don't convert sources, GN will automatically convert these to be relative # to the build directory when it constructs the command line for your # script. sources = [ "foo.txt", "bar.txt" ] # Extra file args passed manually need to be explicitly converted # to be relative to the build directory: args = [ "--data", rebase_path("//mything/data/input.dat", root_build_dir), "--rel", rebase_path("relative_path.txt", root_build_dir) ] + rebase_path(sources, root_build_dir) }
set_default_toolchain(toolchain_label) The given label should identify a toolchain definition (see "gn help toolchain"). This toolchain will be used for all targets unless otherwise specified. This function is only valid to call during the processing of the build configuration file. Since the build configuration file is processed separately for each toolchain, this function will be a no-op when called under any non-default toolchains. For example, the default toolchain should be appropriate for the current environment. If the current environment is 32-bit and somebody references a target with a 64-bit toolchain, we wouldn't want processing of the build config file for the 64-bit toolchain to reset the default toolchain to 64-bit, we want to keep it 32-bits.
toolchain_label Toolchain name.
# Set default toolchain only has an effect when run in the context of the # default toolchain. Pick the right one according to the current CPU # architecture. if (target_cpu == "x64") { set_default_toolchain("//toolchains:64") } else if (target_cpu == "x86") { set_default_toolchain("//toolchains:32") }
set_defaults(<target_type_name>) { <values...> } Sets the default values for a given target type. Whenever target_type_name is seen in the future, the values specified in set_default's block will be copied into the current scope. When the target type is used, the variable copying is very strict. If a variable with that name is already in scope, the build will fail with an error. set_defaults can be used for built-in target types ("executable", "shared_library", etc.) and custom ones defined via the "template" command. It can be called more than once and the most recent call in any scope will apply, but there is no way to refer to the previous defaults and modify them (each call to set_defaults must supply a complete list of all defaults it wants). If you want to share defaults, store them in a separate variable.
set_defaults("static_library") { configs = [ "//tools/mything:settings" ] } static_library("mylib") { # The configs will be auto-populated as above. You can remove it if # you don't want the default for a particular default: configs -= [ "//tools/mything:settings" ] }
This feature is deprecated. It will be removed once all usages have been removed. Only supports a single argument that needs to be an empty list.
result = split_list(input, n) Given a list and a number N, splits the list into N sub-lists of approximately equal size. The return value is a list of the sub-lists. The result will always be a list of size N. If N is greater than the number of elements in the input, it will be padded with empty lists. The expected use is to divide source files into smaller uniform chunks.
The code: mylist = [1, 2, 3, 4, 5, 6] print(split_list(mylist, 3)) Will print: [[1, 2], [3, 4], [5, 6]
result = string_join(separator, strings) Concatenate a list of strings with intervening occurrences of separator.
string_join("", ["a", "b", "c"]) --> "abc" string_join("|", ["a", "b", "c"]) --> "a|b|c" string_join(", ", ["a", "b", "c"]) --> "a, b, c" string_join("s", ["", ""]) --> "s"
result = string_replace(str, old, new[, max]) Returns a copy of the string str in which the occurrences of old have been replaced with new, optionally restricting the number of replacements. The replacement is performed sequentially, so if new contains old, it won't be replaced.
The code: mystr = "Hello, world!" print(string_replace(mystr, "world", "GN")) Will print: Hello, GN!
result = string_split(str[, sep]) Split string into all substrings separated by separator and returns a list of the substrings between those separators. If the separator argument is omitted, the split is by any whitespace, and any leading/trailing whitespace is ignored; similar to Python's str.split().
string_split("") --> [] string_split("a") --> ["a"] string_split(" aa bb") --> ["aa", "bb"]
string_split("", "|") --> [""] string_split(" a b ", " ") --> ["", "", "a", "b", "", ""] string_split("aa+-bb+-c", "+-") --> ["aa", "bb", "c"]
A template defines a custom name that acts like a function. It provides a way to add to the built-in target types. The template() function is used to declare a template. To invoke the template, just use the name of the template like any other target type. Often you will want to declare your template in a special file that other files will import (see "gn help import") so your template rule can be shared across build files.
When you call template() it creates a closure around all variables currently in scope with the code in the template block. When the template is invoked, the closure will be executed. When the template is invoked, the code in the caller is executed and passed to the template code as an implicit "invoker" variable. The template uses this to read state out of the invoking code. One thing explicitly excluded from the closure is the "current directory" against which relative file names are resolved. The current directory will be that of the invoking code, since typically that code specifies the file names. This means all files internal to the template should use absolute names. A template will typically forward some or all variables from the invoking scope to a target that it defines. Often, such variables might be optional. Use the pattern: if (defined(invoker.deps)) { deps = invoker.deps } The function forward_variables_from() provides a shortcut to forward one or more or possibly all variables in this manner: forward_variables_from(invoker, ["deps", "public_deps"])
Your template should almost always define a built-in target with the name the template invoker specified. For example, if you have an IDL template and somebody does: idl("foo") {... you will normally want this to expand to something defining a source_set or static_library named "foo" (among other things you may need). This way, when another target specifies a dependency on "foo", the static_library or source_set will be linked. It is also important that any other targets your template expands to have unique names, or you will get collisions. Access the invoking name in your template via the implicit "target_name" variable. This should also be the basis for how other targets that a template expands to ensure uniqueness. A typical example would be a template that defines an action to generate some source files, and a source_set to compile that source. Your template would name the source_set "target_name" because that's what you want external targets to depend on to link your code. And you would name the action something like "${target_name}_action" to make it unique. The source set would have a dependency on the action to make it run.
You can use template to redefine a built-in target in which case your template takes a precedence over the built-in one. All uses of the target from within the template definition will refer to the built-in target which makes it possible to extend the behavior of the built-in target: template("shared_library") { shared_library(shlib) { forward_variables_from(invoker, "*") ... } }
template("my_idl") { # Be nice and help callers debug problems by checking that the variables # the template requires are defined. This gives a nice message rather than # giving the user an error about an undefined variable in the file defining # the template # # You can also use defined() to give default values to variables # unspecified by the invoker. assert(defined(invoker.sources), "Need sources in $target_name listing the idl files.") # Name of the intermediate target that does the code gen. This must # incorporate the target name so it's unique across template # instantiations. code_gen_target_name = target_name + "_code_gen" # Intermediate target to convert IDL to C source. Note that the name is # based on the name the invoker of the template specified. This way, each # time the template is invoked we get a unique intermediate action name # (since all target names are in the global scope). action_foreach(code_gen_target_name) { # Access the scope defined by the invoker via the implicit "invoker" # variable. sources = invoker.sources # Note that we need an absolute path for our script file name. The # current directory when executing this code will be that of the invoker # (this is why we can use the "sources" directly above without having to # rebase all of the paths). But if we need to reference a script relative # to the template file, we'll need to use an absolute path instead. script = "//tools/idl/idl_code_generator.py" # Tell GN how to expand output names given the sources. # See "gn help source_expansion" for more. outputs = [ "$target_gen_dir/{{source_name_part}}.cc", "$target_gen_dir/{{source_name_part}}.h" ] } # Name the source set the same as the template invocation so instancing # this template produces something that other targets can link to in their # deps. source_set(target_name) { # Generates the list of sources, we get these from the action_foreach # above. sources = get_target_outputs(":$code_gen_target_name") # This target depends on the files produced by the above code gen target. deps = [ ":$code_gen_target_name" ] } }
# This calls the template code above, defining target_name to be # "foo_idl_files" and "invoker" to be the set of stuff defined in the curly # brackets. my_idl("foo_idl_files") { # Goes into the template as "invoker.sources". sources = [ "foo.idl", "bar.idl" ] } # Here is a target that depends on our template. executable("my_exe") { # Depend on the name we gave the template call above. Internally, this will # produce a dependency from executable to the source_set inside the # template (since it has this name), which will in turn depend on the code # gen action. deps = [ ":foo_idl_files" ] }
tool(<tool type>) { <tool variables...> }
Compiler tools: "cc": C compiler "cxx": C++ compiler "cxx_module": C++ compiler used for Clang .modulemap files "objc": Objective C compiler "objcxx": Objective C++ compiler "rc": Resource compiler (Windows .rc files) "asm": Assembler "swift": Swift compiler driver Linker tools: "alink": Linker for static libraries (archives) "solink": Linker for shared libraries "link": Linker for executables Other tools: "stamp": Tool for creating stamp files "copy": Tool to copy files. "action": Defaults for actions Platform specific tools: "copy_bundle_data": [iOS, macOS] Tool to copy files in a bundle. "compile_xcassets": [iOS, macOS] Tool to compile asset catalogs. Rust tools: "rust_bin": Tool for compiling Rust binaries "rust_cdylib": Tool for compiling C-compatible dynamic libraries. "rust_dylib": Tool for compiling Rust dynamic libraries. "rust_macro": Tool for compiling Rust procedural macros. "rust_rlib": Tool for compiling Rust libraries. "rust_staticlib": Tool for compiling Rust static libraries.
command [string with substitutions] Valid for: all tools except "action" (required) The command to run. command_launcher [string] Valid for: all tools except "action" (optional) The prefix with which to launch the command (e.g. the path to a Goma or CCache compiler launcher). Note that this prefix will not be included in the compilation database or IDE files generated from the build. default_output_dir [string with substitutions] Valid for: linker tools Default directory name for the output file relative to the root_build_dir. It can contain other substitution patterns. This will be the default value for the {{output_dir}} expansion (discussed below) but will be overridden by the "output_dir" variable in a target, if one is specified. GN doesn't do anything with this string other than pass it along, potentially with target-specific overrides. It is the tool's job to use the expansion so that the files will be in the right place. default_output_extension [string] Valid for: linker tools Extension for the main output of a linkable tool. It includes the leading dot. This will be the default value for the {{output_extension}} expansion (discussed below) but will be overridden by by the "output extension" variable in a target, if one is specified. Empty string means no extension. GN doesn't actually do anything with this extension other than pass it along, potentially with target-specific overrides. One would typically use the {{output_extension}} value in the "outputs" to read this value. Example: default_output_extension = ".exe" depfile [string with substitutions] Valid for: compiler tools (optional) If the tool can write ".d" files, this specifies the name of the resulting file. These files are used to list header file dependencies (or other implicit input dependencies) that are discovered at build time. See also "depsformat". Example: depfile = "{{output}}.d" depsformat [string] Valid for: compiler tools (when depfile is specified) Format for the deps outputs. This is either "gcc" or "msvc". See the ninja documentation for "deps" for more information. Example: depsformat = "gcc" description [string with substitutions, optional] Valid for: all tools What to print when the command is run. Example: description = "Compiling {{source}}" exe_output_extension [string, optional, rust tools only] rlib_output_extension [string, optional, rust tools only] dylib_output_extension [string, optional, rust tools only] cdylib_output_extension [string, optional, rust tools only] rust_proc_macro_output_extension [string, optional, rust tools only] Valid for: Rust tools These specify the default tool output for each of the crate types. The default is empty for executables, shared, and static libraries and ".rlib" for rlibs. Note that the Rust compiler complains with an error if external crates do not take the form `lib<name>.rlib` or `lib<name>.<shared_extension>`, where `<shared_extension>` is `.so`, `.dylib`, or `.dll` as appropriate for the platform. lib_switch [string, optional, link tools only] lib_dir_switch [string, optional, link tools only] Valid for: Linker tools except "alink" These strings will be prepended to the libraries and library search directories, respectively, because linkers differ on how to specify them. If you specified: lib_switch = "-l" lib_dir_switch = "-L" then the "{{libs}}" expansion for [ "freetype", "expat" ] would be "-lfreetype -lexpat". framework_switch [string, optional, link tools only] weak_framework_switch [string, optional, link tools only] framework_dir_switch [string, optional, link tools only] Valid for: Linker tools These strings will be prepended to the frameworks and framework search path directories, respectively, because linkers differ on how to specify them. If you specified: framework_switch = "-framework " weak_framework_switch = "-weak_framework " framework_dir_switch = "-F" and: framework_dirs = [ "$root_out_dir" ] frameworks = [ "UIKit.framework", "Foo.framework" ] weak_frameworks = [ "MediaPlayer.framework" ] would be: "-F. -framework UIKit -framework Foo -weak_framework MediaPlayer" swiftmodule_switch [string, optional, link tools only] Valid for: Linker tools except "alink" The string will be prependend to the path to the .swiftmodule files that are embedded in the linker output. If you specified: swiftmodule_swift = "-Wl,-add_ast_path," then the "{{swiftmodules}}" expansion for [ "obj/foo/Foo.swiftmodule" ] would be "-Wl,-add_ast_path,obj/foo/Foo.swiftmodule" outputs [list of strings with substitutions] Valid for: Linker and compiler tools (required) An array of names for the output files the tool produces. These are relative to the build output directory. There must always be at least one output file. There can be more than one output (a linker might produce a library and an import library, for example). This array just declares to GN what files the tool will produce. It is your responsibility to specify the tool command that actually produces these files. If you specify more than one output for shared library links, you should consider setting link_output, depend_output, and runtime_outputs. Example for a compiler tool that produces .obj files: outputs = [ "{{source_out_dir}}/{{source_name_part}}.obj" ] Example for a linker tool that produces a .dll and a .lib. The use of {{target_output_name}}, {{output_extension}} and {{output_dir}} allows the target to override these values. outputs = [ "{{output_dir}}/{{target_output_name}}{{output_extension}}", "{{output_dir}}/{{target_output_name}}.lib", ] partial_outputs [list of strings with substitutions] Valid for: "swift" only An array of names for the partial outputs the tool produces. These are relative to the build output directory. The expansion will be evaluated for each file listed in the "sources" of the target. This is used to deal with whole module optimization, allowing to list one object file per source file when whole module optimization is disabled. pool [label, optional] Valid for: all tools (optional) Label of the pool to use for the tool. Pools are used to limit the number of tasks that can execute concurrently during the build. See also "gn help pool". link_output [string with substitutions] depend_output [string with substitutions] Valid for: "solink" only (optional) These two files specify which of the outputs from the solink tool should be used for linking and dependency tracking. These should match entries in the "outputs". If unspecified, the first item in the "outputs" array will be used for all. See "Separate linking and dependencies for shared libraries" below for more. On Windows, where the tools produce a .dll shared library and a .lib import library, you will want the first two to be the import library and the third one to be the .dll file. On Linux, if you're not doing the separate linking/dependency optimization, all of these should be the .so output. output_prefix [string] Valid for: Linker tools (optional) Prefix to use for the output name. Defaults to empty. This prefix will be prepended to the name of the target (or the output_name if one is manually specified for it) if the prefix is not already there. The result will show up in the {{output_name}} substitution pattern. Individual targets can opt-out of the output prefix by setting: output_prefix_override = true (see "gn help output_prefix_override"). This is typically used to prepend "lib" to libraries on Posix systems: output_prefix = "lib" precompiled_header_type [string] Valid for: "cc", "cxx", "objc", "objcxx" Type of precompiled headers. If undefined or the empty string, precompiled headers will not be used for this tool. Otherwise use "gcc" or "msvc". For precompiled headers to be used for a given target, the target (or a config applied to it) must also specify a "precompiled_header" and, for "msvc"-style headers, a "precompiled_source" value. If the type is "gcc", then both "precompiled_header" and "precompiled_source" must resolve to the same file, despite the different formats required for each." See "gn help precompiled_header" for more. restat [boolean] Valid for: all tools (optional, defaults to false) Requests that Ninja check the file timestamp after this tool has run to determine if anything changed. Set this if your tool has the ability to skip writing output if the output file has not changed. Normally, Ninja will assume that when a tool runs the output be new and downstream dependents must be rebuild. When this is set to trye, Ninja can skip rebuilding downstream dependents for input changes that don't actually affect the output. Example: restat = true rspfile [string with substitutions] Valid for: all tools except "action" (optional) Name of the response file. If empty, no response file will be used. See "rspfile_content". rspfile_content [string with substitutions] Valid for: all tools except "action" (required when "rspfile" is used) The contents to be written to the response file. This may include all or part of the command to send to the tool which allows you to get around OS command-line length limits. This example adds the inputs and libraries to a response file, but passes the linker flags directly on the command line: tool("link") { command = "link -o {{output}} {{ldflags}} @{{output}}.rsp" rspfile = "{{output}}.rsp" rspfile_content = "{{inputs}} {{solibs}} {{libs}} {{rlibs}}" } runtime_outputs [string list with substitutions] Valid for: linker tools If specified, this list is the subset of the outputs that should be added to runtime deps (see "gn help runtime_deps"). By default (if runtime_outputs is empty or unspecified), it will be the link_output.
All paths are relative to the root build directory, which is the current directory for running all tools. These expansions are available to all tools: {{label}} The label of the current target. This is typically used in the "description" field for link tools. The toolchain will be omitted from the label for targets in the default toolchain, and will be included for targets in other toolchains. {{label_name}} The short name of the label of the target. This is the part after the colon. For "//foo/bar:baz" this will be "baz". Unlike {{target_output_name}}, this is not affected by the "output_prefix" in the tool or the "output_name" set on the target. {{label_no_toolchain}} The label of the current target, never including the toolchain (otherwise, this is identical to {{label}}). This is used as the module name when using .modulemap files. {{output}} The relative path and name of the output(s) of the current build step. If there is more than one output, this will expand to a list of all of them. Example: "out/base/my_file.o" {{target_gen_dir}} {{target_out_dir}} The directory of the generated file and output directories, respectively, for the current target. There is no trailing slash. See also {{output_dir}} for linker tools. Example: "out/base/test" {{target_output_name}} The short name of the current target with no path information, or the value of the "output_name" variable if one is specified in the target. This will include the "output_prefix" if any. See also {{label_name}}. Example: "libfoo" for the target named "foo" and an output prefix for the linker tool of "lib". Compiler tools have the notion of a single input and a single output, along with a set of compiler-specific flags. The following expansions are available: {{asmflags}} {{cflags}} {{cflags_c}} {{cflags_cc}} {{cflags_objc}} {{cflags_objcc}} {{defines}} {{include_dirs}} Strings correspond that to the processed flags/defines/include directories specified for the target. Example: "--enable-foo --enable-bar" Defines will be prefixed by "-D" and include directories will be prefixed by "-I" (these work with Posix tools as well as Microsoft ones). {{module_deps}} {{module_deps_no_self}} Strings that correspond to the flags necessary to depend upon the Clang modules referenced by the current target. The "_no_self" version doesn't include the module for the current target, and can be used to compile the pcm itself. {{source}} The relative path and name of the current input file. Example: "../../base/my_file.cc" {{source_file_part}} The file part of the source including the extension (with no directory information). Example: "foo.cc" {{source_name_part}} The filename part of the source file with no directory or extension. Example: "foo" {{source_gen_dir}} {{source_out_dir}} The directory in the generated file and output directories, respectively, for the current input file. If the source file is in the same directory as the target is declared in, they will will be the same as the "target" versions above. Example: "gen/base/test" Linker tools have multiple inputs and (potentially) multiple outputs. The static library tool ("alink") is not considered a linker tool. The following expansions are available: {{inputs}} {{inputs_newline}} Expands to the inputs to the link step. This will be a list of object files and static libraries. Example: "obj/foo.o obj/bar.o obj/somelibrary.a" The "_newline" version will separate the input files with newlines instead of spaces. This is useful in response files: some linkers can take a "-filelist" flag which expects newline separated files, and some Microsoft tools have a fixed-sized buffer for parsing each line of a response file. {{ldflags}} Expands to the processed set of ldflags and library search paths specified for the target. Example: "-m64 -fPIC -pthread -L/usr/local/mylib" {{libs}} Expands to the list of system libraries to link to. Each will be prefixed by the "lib_switch". Example: "-lfoo -lbar" {{output_dir}} The value of the "output_dir" variable in the target, or the the value of the "default_output_dir" value in the tool if the target does not override the output directory. This will be relative to the root_build_dir and will not end in a slash. Will be "." for output to the root_build_dir. This is subtly different than {{target_out_dir}} which is defined by GN based on the target's path and not overridable. {{output_dir}} is for the final output, {{target_out_dir}} is generally for object files and other outputs. Usually {{output_dir}} would be defined in terms of either {{target_out_dir}} or {{root_out_dir}} {{output_extension}} The value of the "output_extension" variable in the target, or the value of the "default_output_extension" value in the tool if the target does not specify an output extension. Example: ".so" {{solibs}} Extra libraries from shared library dependencies not specified in the {{inputs}}. This is the list of link_output files from shared libraries (if the solink tool specifies a "link_output" variable separate from the "depend_output"). These should generally be treated the same as libs by your tool. Example: "libfoo.so libbar.so" {{rlibs}} Any Rust .rlibs which need to be linked into a final C++ target. These should be treated as {{inputs}} except that sometimes they might have different linker directives applied. Example: "obj/foo/libfoo.rlib" {{frameworks}} Shared libraries packaged as framework bundle. This is principally used on Apple's platforms (macOS and iOS). All name must be ending with ".framework" suffix; the suffix will be stripped when expanding {{frameworks}} and each item will be preceded by "-framework" or "-weak_framework". {{swiftmodules}} Swift .swiftmodule files that needs to be embedded into the binary. This is necessary to correctly link with object generated by the Swift compiler (the .swiftmodule file cannot be embedded in object files directly). Those will be prefixed with "swiftmodule_switch" value. The static library ("alink") tool allows {{arflags}} plus the common tool substitutions. The copy tool allows the common compiler/linker substitutions, plus {{source}} which is the source of the copy. The stamp tool allows only the common tool substitutions. The copy_bundle_data and compile_xcassets tools only allows the common tool substitutions. Both tools are required to create iOS/macOS bundles and need only be defined on those platforms. The copy_bundle_data tool will be called with one source and needs to copy (optionally optimizing the data representation) to its output. It may be called with a directory as input and it needs to be recursively copied. The compile_xcassets tool will be called with one or more source (each an asset catalog) that needs to be compiled to a single output. The following substitutions are available: {{inputs}} Expands to the list of .xcassets to use as input to compile the asset catalog. {{bundle_product_type}} Expands to the product_type of the bundle that will contain the compiled asset catalog. Usually corresponds to the product_type property of the corresponding create_bundle target. {{bundle_partial_info_plist}} Expands to the path to the partial Info.plist generated by the assets catalog compiler. Usually based on the target_name of the create_bundle target. {{xcasset_compiler_flags}} Expands to the list of flags specified in corresponding create_bundle target. The Swift tool has multiple input and outputs. It must have exactly one output of .swiftmodule type, but can have one or more object file outputs, in addition to other type of ouputs. The following expansions are available: {{module_name}} Expands to the string representing the module name of target under compilation (see "module_name" variable). {{module_dirs}} Expands to the list of -I<path> for the target Swift module search path computed from target dependencies. {{swiftflags}} Expands to the list of strings representing Swift compiler flags. Rust tools have the notion of a single input and a single output, along with a set of compiler-specific flags. The following expansions are available: {{crate_name}} Expands to the string representing the crate name of target under compilation. {{crate_type}} Expands to the string representing the type of crate for the target under compilation. {{externs}} Expands to the list of --extern flags needed to include addition Rust libraries in this target. Includes any specified renamed dependencies. {{rustdeps}} Expands to the list of -Ldependency=<path> strings needed to compile this target. {{rustenv}} Expands to the list of environment variables. {{rustflags}} Expands to the list of strings representing Rust compiler flags.
Shared libraries are special in that not all changes to them require that dependent targets be re-linked. If the shared library is changed but no imports or exports are different, dependent code needn't be relinked, which can speed up the build. If your link step can output a list of exports from a shared library and writes the file only if the new one is different, the timestamp of this file can be used for triggering re-links, while the actual shared library would be used for linking. You will need to specify restat = true in the linker tool to make this work, so Ninja will detect if the timestamp of the dependency file has changed after linking (otherwise it will always assume that running a command updates the output): tool("solink") { command = "..." outputs = [ "{{output_dir}}/{{target_output_name}}{{output_extension}}", "{{output_dir}}/{{target_output_name}}{{output_extension}}.TOC", ] link_output = "{{output_dir}}/{{target_output_name}}{{output_extension}}" depend_output = "{{output_dir}}/{{target_output_name}}{{output_extension}}.TOC" restat = true }
toolchain("my_toolchain") { # Put these at the top to apply to all tools below. lib_switch = "-l" lib_dir_switch = "-L" tool("cc") { command = "gcc {{source}} -o {{output}}" outputs = [ "{{source_out_dir}}/{{source_name_part}}.o" ] description = "GCC {{source}}" } tool("cxx") { command = "g++ {{source}} -o {{output}}" outputs = [ "{{source_out_dir}}/{{source_name_part}}.o" ] description = "G++ {{source}}" } };
A toolchain is a set of commands and build flags used to compile the source code. The toolchain() function defines these commands.
You can have more than one toolchain in use at once in a build and a target can exist simultaneously in multiple toolchains. A build file is executed once for each toolchain it is referenced in so the GN code can vary all parameters of each target (or which targets exist) on a per-toolchain basis. When you have a simple build with only one toolchain, the build config file is loaded only once at the beginning of the build. It must call set_default_toolchain() (see "gn help set_default_toolchain") to tell GN the label of the toolchain definition to use. The "toolchain_args" section of the toolchain definition is ignored. When a target has a dependency on a target using different toolchain (see "gn help labels" for how to specify this), GN will start a build using that secondary toolchain to resolve the target. GN will load the build config file with the build arguments overridden as specified in the toolchain_args. Because the default toolchain is already known, calls to set_default_toolchain() are ignored. To load a file in an alternate toolchain, GN does the following: 1. Loads the file with the toolchain definition in it (as determined by the toolchain label). 2. Re-runs the master build configuration file, applying the arguments specified by the toolchain_args section of the toolchain definition. 3. Loads the destination build file in the context of the configuration file in the previous step. The toolchain configuration is two-way. In the default toolchain (i.e. the main build target) the configuration flows from the build config file to the toolchain. The build config file looks at the state of the build (OS type, CPU architecture, etc.) and decides which toolchain to use (via set_default_toolchain()). In secondary toolchains, the configuration flows from the toolchain to the build config file: the "toolchain_args" in the toolchain definition specifies the arguments to re-invoke the build.
tool() The tool() function call specifies the commands to run for a given step. See "gn help tool". toolchain_args [scope] Overrides for build arguments to pass to the toolchain when invoking it. This is a variable of type "scope" where the variable names correspond to variables in declare_args() blocks. When you specify a target using an alternate toolchain, the master build configuration file is re-interpreted in the context of that toolchain. toolchain_args allows you to control the arguments passed into this alternate invocation of the build. Any default system arguments or arguments passed in via "gn args" will also be passed to the alternate invocation unless explicitly overridden by toolchain_args. The toolchain_args will be ignored when the toolchain being defined is the default. In this case, it's expected you want the default argument values. See also "gn help buildargs" for an overview of these arguments. propagates_configs [boolean, default=false] Determines whether public_configs and all_dependent_configs in this toolchain propagate to targets in other toolchains. When false (the default), this toolchain will not propagate any configs to targets in other toolchains that depend on it targets inside this toolchain. This matches the most common usage of toolchains where they represent different architectures or compilers and the settings that apply to one won't necessarily apply to others. When true, configs (public and all-dependent) will cross the boundary out of this toolchain as if the toolchain boundary wasn't there. This only affects one direction of dependencies: a toolchain can't control whether it accepts such configs, only whether it pushes them. The build is responsible for ensuring that any external targets depending on targets in this toolchain are compatible with the compiler flags, etc. that may be propagated. deps [string list] Dependencies of this toolchain. These dependencies will be resolved before any target in the toolchain is compiled. To avoid circular dependencies these must be targets defined in another toolchain. This is expressed as a list of targets, and generally these targets will always specify a toolchain: deps = [ "//foo/bar:baz(//build/toolchain:bootstrap)" ] This concept is somewhat inefficient to express in Ninja (it requires a lot of duplicate of rules) so should only be used when absolutely necessary.
toolchain("32") { tool("cc") { command = "gcc {{source}}" ... } toolchain_args = { use_doom_melon = true # Doom melon always required for 32-bit builds. current_cpu = "x86" } } toolchain("64") { tool("cc") { command = "gcc {{source}}" ... } toolchain_args = { # use_doom_melon is not overridden here, it will take the default. current_cpu = "x64" } }
If a 64-bit target wants to depend on a 32-bit binary, it would specify a dependency using data_deps (data deps are like deps that are only needed at runtime and aren't linked, since you can't link a 32-bit and a 64-bit library). executable("my_program") { ... if (target_cpu == "x64") { # The 64-bit build needs this 32-bit helper. data_deps = [ ":helper(//toolchains:32)" ] } } if (target_cpu == "x86") { # Our helper library is only compiled in 32-bits. shared_library("helper") { ... } }
write_file(filename, data, output_conversion = "") If data is a list, the list will be written one-item-per-line with no quoting or brackets. If the file exists and the contents are identical to that being written, the file will not be updated. This will prevent unnecessary rebuilds of targets that depend on this file. One use for write_file is to write a list of inputs to an script that might be too long for the command line. However, it is preferable to use response files for this purpose. See "gn help response_file_contents".
filename Filename to write. This must be within the output directory. data The list or string to write. output_conversion Controls how the output is written. See "gn help io_conversion".
The build configuration usually sets this value based on the value of "host_cpu" (see "gn help host_cpu") and then threads this through the toolchain definitions to ensure that it always reflects the appropriate value. This value is not used internally by GN for any purpose. It is set to the empty string ("") by default but is declared so that it can be overridden on the command line if so desired. See "gn help target_cpu" for a list of common values returned.
The build configuration usually sets this value based on the value of "target_os" (see "gn help target_os"), and then threads this through the toolchain definitions to ensure that it always reflects the appropriate value. This value is not used internally by GN for any purpose. It is set to the empty string ("") by default but is declared so that it can be overridden on the command line if so desired. See "gn help target_os" for a list of common values returned.
A fully-qualified label representing the current toolchain. You can use this to make toolchain-related decisions in the build. See also "default_toolchain".
if (current_toolchain == "//build:64_bit_toolchain") { executable("output_thats_64_bit_only") { ...
A fully-qualified label representing the default toolchain, which may not necessarily be the current one (see "current_toolchain").
Corresponds to the number printed by `gn --version`.
assert(gn_version >= 1700, "need GN version 1700 for the frobulate feature")
This is value is exposed so that cross-compile toolchains can access the host architecture when needed. The value should generally be considered read-only, but it can be overriden in order to handle unusual cases where there might be multiple plausible values for the host architecture (e.g., if you can do either 32-bit or 64-bit builds). The value is not used internally by GN for any purpose.
- "x64" - "x86"
This value is exposed so that cross-compiles can access the host build system's settings. This value should generally be treated as read-only. It, however, is not used internally by GN for any purpose.
- "linux" - "mac" - "win"
Inside a template invocation, this variable refers to the scope of the invoker of the template. Outside of template invocations, this variable is undefined. All of the variables defined inside the template invocation are accessible as members of the "invoker" scope. This is the way that templates read values set by the callers. This is often used with "defined" to see if a value is set on the invoking scope. See "gn help template" for more examples.
template("my_template") { print(invoker.sources) # Prints [ "a.cc", "b.cc" ] print(defined(invoker.foo)) # Prints false. print(defined(invoker.bar)) # Prints true. } my_template("doom_melon") { sources = [ "a.cc", "b.cc" ] bar = 123 }
Normally used in toolchain definitions if running some command requires Python. You will normally not need this when invoking scripts since GN automatically finds it for you.
This is the root build output directory which will be the current directory when executing all compilers and scripts. Most often this is used with rebase_path (see "gn help rebase_path") to convert arguments to be relative to a script's current directory.
Absolute path to the root of the generated output directory tree for the current toolchain. An example would be "//out/Debug/gen" for the default toolchain, or "//out/Debug/arm/gen" for the "arm" toolchain. This is primarily useful for setting up include paths for generated files. If you are passing this to a script, you will want to pass it through rebase_path() (see "gn help rebase_path") to convert it to be relative to the build directory. See also "target_gen_dir" which is usually a better location for generated files. It will be inside the root generated dir.
Absolute path to the root of the output directory tree for the current toolchain. It will not have a trailing slash. For the default toolchain this will be the same as the root_build_dir. An example would be "//out/Debug" for the default toolchain, or "//out/Debug/arm" for the "arm" toolchain. This is primarily useful for setting up script calls. If you are passing this to a script, you will want to pass it through rebase_path() (see "gn help rebase_path") to convert it to be relative to the build directory. See also "target_out_dir" which is usually a better location for output files. It will be inside the root output dir.
action("myscript") { # Pass the output dir to the script. args = [ "-o", rebase_path(root_out_dir, root_build_dir) ] }
This value should be used to indicate the desired architecture for the primary objects of the build. It will match the cpu architecture of the default toolchain, but not necessarily the current toolchain. In many cases, this is the same as "host_cpu", but in the case of cross-compiles, this can be set to something different. This value is different from "current_cpu" in that it does not change based on the current toolchain. When writing rules, "current_cpu" should be used rather than "target_cpu" most of the time. This value is not used internally by GN for any purpose, so it may be set to whatever value is needed for the build. GN defaults this value to the empty string ("") and the configuration files should set it to an appropriate value (e.g., setting it to the value of "host_cpu") if it is not overridden on the command line or in the args.gn file.
- "x86" - "x64" - "arm" - "arm64" - "mipsel"
Absolute path to the target's generated file directory. This will be the "root_gen_dir" followed by the relative path to the current build file. If your file is in "//tools/doom_melon" then target_gen_dir would be "//out/Debug/gen/tools/doom_melon". It will not have a trailing slash. This is primarily useful for setting up include paths for generated files. If you are passing this to a script, you will want to pass it through rebase_path() (see "gn help rebase_path") to convert it to be relative to the build directory. See also "gn help root_gen_dir".
action("myscript") { # Pass the generated output dir to the script. args = [ "-o", rebase_path(target_gen_dir, root_build_dir) ] }
Inside a target or template invocation, this variable refers to the name given to the target or template invocation. Outside of these, this variable is undefined. This is most often used in template definitions to name targets defined in the template based on the name of the invocation. This is necessary both to ensure generated targets have unique names and to generate a target with the exact name of the invocation that other targets can depend on. Be aware that this value will always reflect the innermost scope. So when defining a target inside a template, target_name will refer to the target rather than the template invocation. To get the name of the template invocation in this case, you should save target_name to a temporary variable outside of any target definitions. See "gn help template" for more examples.
executable("doom_melon") { print(target_name) # Prints "doom_melon". } template("my_template") { print(target_name) # Prints "space_ray" when invoked below. executable(target_name + "_impl") { print(target_name) # Prints "space_ray_impl". } } my_template("space_ray") { }
This value should be used to indicate the desired operating system for the primary object(s) of the build. It will match the OS of the default toolchain. In many cases, this is the same as "host_os", but in the case of cross-compiles, it may be different. This variable differs from "current_os" in that it can be referenced from inside any toolchain and will always return the initial value. This should be set to the most specific value possible. So, "android" or "chromeos" should be used instead of "linux" where applicable, even though Android and ChromeOS are both Linux variants. This can mean that one needs to write if (target_os == "android" || target_os == "linux") { # ... } and so forth. This value is not used internally by GN for any purpose, so it may be set to whatever value is needed for the build. GN defaults this value to the empty string ("") and the configuration files should set it to an appropriate value (e.g., setting it to the value of "host_os") if it is not set via the command line or in the args.gn file.
- "android" - "chromeos" - "ios" - "linux" - "nacl" - "mac" - "win"
Absolute path to the target's generated file directory. If your current target is in "//tools/doom_melon" then this value might be "//out/Debug/obj/tools/doom_melon". It will not have a trailing slash. This is primarily useful for setting up arguments for calling scripts. If you are passing this to a script, you will want to pass it through rebase_path() (see "gn help rebase_path") to convert it to be relative to the build directory. See also "gn help root_out_dir".
action("myscript") { # Pass the output dir to the script. args = [ "-o", rebase_path(target_out_dir, root_build_dir) ] }
Valid for `rust_library` targets and `executable`, `static_library`, and `shared_library` targets that contain Rust sources. A scope, each key indicating the renamed crate and the corresponding value specifying the label of the dependency producing the relevant binary. All dependencies listed in this field *must* be listed as deps of the target. executable("foo") { sources = [ "main.rs" ] deps = [ "//bar" ] } This target would compile the `foo` crate with the following `extern` flag: `rustc ...command... --extern bar=<build_out_dir>/obj/bar` executable("foo") { sources = [ "main.rs" ] deps = [ ":bar" ] aliased_deps = { bar_renamed = ":bar" } } With the addition of `aliased_deps`, above target would instead compile with: `rustc ...command... --extern bar_renamed=<build_out_dir>/obj/bar`
A list of config labels. All targets depending on this one, and recursively, all targets depending on those, will have the configs listed in this variable added to them. These configs will also apply to the current target. This addition happens in a second phase once a target and all of its dependencies have been resolved. Therefore, a target will not see these force-added configs in their "configs" variable while the script is running, and they can not be removed. As a result, this capability should generally only be used to add defines and include directories necessary to compile a target's headers. See also "public_configs".
1. Those set on the current target (not in a config). 2. Those set on the "configs" on the target in order that the configs appear in the list. 3. Those set on the "all_dependent_configs" on the target in order that the configs appear in the list. 4. Those set on the "public_configs" on the target in order that those configs appear in the list. 5. all_dependent_configs pulled from dependencies, in the order of the "deps" list. This is done recursively. If a config appears more than once, only the first occurence will be used. 6. public_configs pulled from dependencies, in the order of the "deps" list. If a dependency is public, they will be applied recursively.
A list of target labels. Must be a subset of the target's "deps". These targets will be permitted to include headers from the current target despite the dependency going in the opposite direction. When you use this, both targets must be included in a final binary for it to link. To keep linker errors from happening, it is good practice to have all external dependencies depend only on one of the two targets, and to set the visibility on the other to enforce this. Thus the targets will always be linked together in any output.
Normally, for a file in target A to include a file from target B, A must list B as a dependency. This invariant is enforced by the "gn check" command (and the --check flag to "gn gen" -- see "gn help check"). Sometimes, two targets might be the same unit for linking purposes (two source sets or static libraries that would always be linked together in a final executable or shared library) and they each include headers from the other: you want A to be able to include B's headers, and B to include A's headers. This is not an ideal situation but is sometimes unavoidable. This list, if specified, lists which of the dependencies of the current target can include header files from the current target. That is, if A depends on B, B can only include headers from A if it is in A's allow_circular_includes_from list. Normally includes must follow the direction of dependencies, this flag allows them to go in the opposite direction.
In the above example, A's headers are likely to include headers from A's dependencies. Those dependencies may have public_configs that apply flags, defines, and include paths that make those headers work properly. With allow_circular_includes_from, B can include A's headers, and transitively from A's dependencies, without having the dependencies that would bring in the public_configs those headers need. The result may be errors or inconsistent builds. So when you use allow_circular_includes_from, make sure that any compiler settings, flags, and include directories are the same between both targets (consider putting such things in a shared config they can both reference). Make sure the dependencies are also the same (you might consider a group to collect such dependencies they both depend on).
source_set("a") { deps = [ ":b", ":a_b_shared_deps" ] allow_circular_includes_from = [ ":b" ] ... } source_set("b") { deps = [ ":a_b_shared_deps" ] # Sources here can include headers from a despite lack of deps. ... } group("a_b_shared_deps") { public_deps = [ ":c" ] }
A list of flags passed to the archive/lib command that creates static libraries. arflags are NOT pushed to dependents, so applying arflags to source sets or any other target type will be a no-op. As with ldflags, you could put the arflags in a config and set that as a public or "all dependent" config, but that will likely not be what you want. If you have a chain of static libraries dependent on each other, this can cause the flags to propagate up to other static libraries. Due to the nature of how arflags are typically used, you will normally want to apply them directly on static_library targets themselves.
1. Those set on the current target (not in a config). 2. Those set on the "configs" on the target in order that the configs appear in the list. 3. Those set on the "all_dependent_configs" on the target in order that the configs appear in the list. 4. Those set on the "public_configs" on the target in order that those configs appear in the list. 5. all_dependent_configs pulled from dependencies, in the order of the "deps" list. This is done recursively. If a config appears more than once, only the first occurence will be used. 6. public_configs pulled from dependencies, in the order of the "deps" list. If a dependency is public, they will be applied recursively.
For action and action_foreach targets, args is the list of arguments to pass to the script. Typically you would use source expansion (see "gn help source_expansion") to insert the source file names. See also "gn help action" and "gn help action_foreach".
A list of strings. "asmflags" are passed to any invocation of a tool that takes an .asm or .S file as input.
1. Those set on the current target (not in a config). 2. Those set on the "configs" on the target in order that the configs appear in the list. 3. Those set on the "all_dependent_configs" on the target in order that the configs appear in the list. 4. Those set on the "public_configs" on the target in order that those configs appear in the list. 5. all_dependent_configs pulled from dependencies, in the order of the "deps" list. This is done recursively. If a config appears more than once, only the first occurence will be used. 6. public_configs pulled from dependencies, in the order of the "deps" list. If a dependency is public, they will be applied recursively.
A list of label patterns. This list is a list of patterns that must not match any of the transitive dependencies of the target. These include all public, private, and data dependencies, and cross shared library boundaries. This allows you to express that undesirable code isn't accidentally added to downstream dependencies in a way that might otherwise be difficult to notice. Checking does not cross executable boundaries. If a target depends on an executable, it's assumed that the executable is a tool that is producing part of the build rather than something that is linked and distributed. This allows assert_no_deps to express what is distributed in the final target rather than depend on the internal build steps (which may include non-distributable code). See "gn help label_pattern" for the format of the entries in the list. These patterns allow blacklisting individual targets or whole directory hierarchies. Sometimes it is desirable to enforce that many targets have no dependencies on a target or set of targets. One efficient way to express this is to create a group with the assert_no_deps rule on it, and make that group depend on all targets you want to apply that assertion to.
executable("doom_melon") { deps = [ "//foo:bar" ] ... assert_no_deps = [ "//evil/*", # Don't link any code from the evil directory. "//foo:test_support", # This target is also disallowed. ] }
Valid for binary targets that contain Swift sources. Path to an header that includes C/Objective-C functions and types that needs to be made available to the Swift module.
create_bundle. A string corresponding to a path in $root_build_dir. This string is used by the "create_bundle" target to expand the {{bundle_contents_dir}} of the "bundle_data" target it depends on. This must correspond to a path under "bundle_root_dir". See "gn help bundle_root_dir" for examples.
A list of target labels. This list contains target label patterns that should be filtered out when creating the bundle. Any target matching one of those label will be removed from the dependencies of the create_bundle target. This is mostly useful when creating application extension bundle as the application extension has access to runtime resources from the application bundle and thus do not require a second copy. See "gn help create_bundle" for more information.
create_bundle("today_extension") { deps = [ "//base" ] bundle_root_dir = "$root_out_dir/today_extension.appex" bundle_deps_filter = [ # The extension uses //base but does not use any function calling into # third_party/icu and thus does not need the icudtl.dat file. "//third_party/icu:icudata", ] }
bundle_executable_dir: Expansion of {{bundle_executable_dir}} in create_bundle. A string corresponding to a path in $root_build_dir. This string is used by the "create_bundle" target to expand the {{bundle_executable_dir}} of the "bundle_data" target it depends on. This must correspond to a path under "bundle_root_dir". See "gn help bundle_root_dir" for examples.
bundle_resources_dir: Expansion of {{bundle_resources_dir}} in create_bundle. A string corresponding to a path in $root_build_dir. This string is used by the "create_bundle" target to expand the {{bundle_resources_dir}} of the "bundle_data" target it depends on. This must correspond to a path under "bundle_root_dir". See "gn help bundle_root_dir" for examples.
A string corresponding to a path in root_build_dir. This string is used by the "create_bundle" target to expand the {{bundle_root_dir}} of the "bundle_data" target it depends on. This must correspond to a path under root_build_dir.
bundle_data("info_plist") { sources = [ "Info.plist" ] outputs = [ "{{bundle_contents_dir}}/Info.plist" ] } create_bundle("doom_melon.app") { deps = [ ":info_plist" ] bundle_root_dir = "${root_build_dir}/doom_melon.app" bundle_contents_dir = "${bundle_root_dir}/Contents" bundle_resources_dir = "${bundle_contents_dir}/Resources" bundle_executable_dir = "${bundle_contents_dir}/MacOS" }
A list of strings. "cflags" are passed to all invocations of the C, C++, Objective C, and Objective C++ compilers. To target one of these variants individually, use "cflags_c", "cflags_cc", "cflags_objc", and "cflags_objcc", respectively. These variant-specific versions of cflags* will be appended on the compiler command line after "cflags". See also "asmflags" for flags for assembly-language files and "swiftflags" for swift files.
1. Those set on the current target (not in a config). 2. Those set on the "configs" on the target in order that the configs appear in the list. 3. Those set on the "all_dependent_configs" on the target in order that the configs appear in the list. 4. Those set on the "public_configs" on the target in order that those configs appear in the list. 5. all_dependent_configs pulled from dependencies, in the order of the "deps" list. This is done recursively. If a config appears more than once, only the first occurence will be used. 6. public_configs pulled from dependencies, in the order of the "deps" list. If a dependency is public, they will be applied recursively.
A list of strings. "cflags" are passed to all invocations of the C, C++, Objective C, and Objective C++ compilers. To target one of these variants individually, use "cflags_c", "cflags_cc", "cflags_objc", and "cflags_objcc", respectively. These variant-specific versions of cflags* will be appended on the compiler command line after "cflags". See also "asmflags" for flags for assembly-language files and "swiftflags" for swift files.
1. Those set on the current target (not in a config). 2. Those set on the "configs" on the target in order that the configs appear in the list. 3. Those set on the "all_dependent_configs" on the target in order that the configs appear in the list. 4. Those set on the "public_configs" on the target in order that those configs appear in the list. 5. all_dependent_configs pulled from dependencies, in the order of the "deps" list. This is done recursively. If a config appears more than once, only the first occurence will be used. 6. public_configs pulled from dependencies, in the order of the "deps" list. If a dependency is public, they will be applied recursively.
A list of strings. "cflags" are passed to all invocations of the C, C++, Objective C, and Objective C++ compilers. To target one of these variants individually, use "cflags_c", "cflags_cc", "cflags_objc", and "cflags_objcc", respectively. These variant-specific versions of cflags* will be appended on the compiler command line after "cflags". See also "asmflags" for flags for assembly-language files and "swiftflags" for swift files.
1. Those set on the current target (not in a config). 2. Those set on the "configs" on the target in order that the configs appear in the list. 3. Those set on the "all_dependent_configs" on the target in order that the configs appear in the list. 4. Those set on the "public_configs" on the target in order that those configs appear in the list. 5. all_dependent_configs pulled from dependencies, in the order of the "deps" list. This is done recursively. If a config appears more than once, only the first occurence will be used. 6. public_configs pulled from dependencies, in the order of the "deps" list. If a dependency is public, they will be applied recursively.
A list of strings. "cflags" are passed to all invocations of the C, C++, Objective C, and Objective C++ compilers. To target one of these variants individually, use "cflags_c", "cflags_cc", "cflags_objc", and "cflags_objcc", respectively. These variant-specific versions of cflags* will be appended on the compiler command line after "cflags". See also "asmflags" for flags for assembly-language files and "swiftflags" for swift files.
1. Those set on the current target (not in a config). 2. Those set on the "configs" on the target in order that the configs appear in the list. 3. Those set on the "all_dependent_configs" on the target in order that the configs appear in the list. 4. Those set on the "public_configs" on the target in order that those configs appear in the list. 5. all_dependent_configs pulled from dependencies, in the order of the "deps" list. This is done recursively. If a config appears more than once, only the first occurence will be used. 6. public_configs pulled from dependencies, in the order of the "deps" list. If a dependency is public, they will be applied recursively.
A list of strings. "cflags" are passed to all invocations of the C, C++, Objective C, and Objective C++ compilers. To target one of these variants individually, use "cflags_c", "cflags_cc", "cflags_objc", and "cflags_objcc", respectively. These variant-specific versions of cflags* will be appended on the compiler command line after "cflags". See also "asmflags" for flags for assembly-language files and "swiftflags" for swift files.
1. Those set on the current target (not in a config). 2. Those set on the "configs" on the target in order that the configs appear in the list. 3. Those set on the "all_dependent_configs" on the target in order that the configs appear in the list. 4. Those set on the "public_configs" on the target in order that those configs appear in the list. 5. all_dependent_configs pulled from dependencies, in the order of the "deps" list. This is done recursively. If a config appears more than once, only the first occurence will be used. 6. public_configs pulled from dependencies, in the order of the "deps" list. If a dependency is public, they will be applied recursively.
When true (the default), the "gn check" command (as well as "gn gen" with the --check flag) will check this target's sources and headers for proper dependencies. When false, the files in this target will be skipped by default. This does not affect other targets that depend on the current target, it just skips checking the includes of the current target's files. If there are a few conditionally included headers that trip up checking, you can exclude headers individually by annotating them with "nogncheck" (see "gn help nogncheck"). The topic "gn help check" has general information on how checking works and advice on how to pass a check in problematic cases.
source_set("busted_includes") { # This target's includes are messed up, exclude it from checking. check_includes = false ... }
For create_bundle targets, code_signing_args is the list of arguments to pass to the code signing script. Typically you would use source expansion (see "gn help source_expansion") to insert the source file names. See also "gn help create_bundle".
Outputs from the code signing step of a create_bundle target. Must refer to files in the build directory. See also "gn help create_bundle".
An absolute or buildfile-relative file name of a Python script to run for a create_bundle target to perform code signing step. See also "gn help create_bundle".
A list of files used as input for code signing script step of a create_bundle target. Non-absolute paths will be resolved relative to the current build file. See also "gn help create_bundle".
A static library normally doesn't include code from dependencies, but instead forwards the static libraries and source sets in its deps up the dependency chain until a linkable target (an executable or shared library) is reached. The final linkable target only links each static library once, even if it appears more than once in its dependency graph. In some cases the static library might be the final desired output. For example, you may be producing a static library for distribution to third parties. In this case, the static library should include code for all dependencies in one complete package. However, complete static libraries themselves are never linked into other complete static libraries. All complete static libraries are for distribution and linking them in would cause code duplication in this case. If the static library is not for distribution, it should not be complete. GN treats non-complete static libraries as source sets when they are linked into complete static libraries. This is done because some tools like AR do not handle dependent static libraries properly. This makes it easier to write "alink" rules. In rare cases it makes sense to list a header in more than one target if it could be considered conceptually a member of both. libraries.
static_library("foo") { complete_static_lib = true deps = [ "bar" ] }
A list of config labels.
When used on a target, the include_dirs, defines, etc. in each config are appended in the order they appear to the compile command for each file in the target. They will appear after the include_dirs, defines, etc. that the target sets directly. Since configs apply after the values set on a target, directly setting a compiler flag will prepend it to the command line. If you want to append a flag instead, you can put that flag in a one-off config and append that config to the target's configs list. The build configuration script will generally set up the default configs applying to a given target type (see "set_defaults"). When a target is being defined, it can add to or remove from this list.
It is possible to create composite configs by specifying configs on a config. One might do this to forward values, or to factor out blocks of settings from very large configs into more manageable named chunks. In this case, the composite config is expanded to be the concatenation of its own values, and in order, the values from its sub-configs *before* anything else happens. This has some ramifications: - A target has no visibility into a config's sub-configs. Target code only sees the name of the composite config. It can't remove sub-configs or opt in to only parts of it. The composite config may not even be defined before the target is. - You can get duplication of values if a config is listed twice, say, on a target and in a sub-config that also applies. In other cases, the configs applying to a target are de-duped. It's expected that if a config is listed as a sub-config that it is only used in that context. (Note that it's possible to fix this and de-dupe, but it's not normally relevant and complicates the implementation.)
1. Those set on the current target (not in a config). 2. Those set on the "configs" on the target in order that the configs appear in the list. 3. Those set on the "all_dependent_configs" on the target in order that the configs appear in the list. 4. Those set on the "public_configs" on the target in order that those configs appear in the list. 5. all_dependent_configs pulled from dependencies, in the order of the "deps" list. This is done recursively. If a config appears more than once, only the first occurence will be used. 6. public_configs pulled from dependencies, in the order of the "deps" list. If a dependency is public, they will be applied recursively.
# Configs on a target. source_set("foo") { # Don't use the default RTTI config that BUILDCONFIG applied to us. configs -= [ "//build:no_rtti" ] # Add some of our own settings. configs += [ ":mysettings" ] } # Create a default_optimization config that forwards to one of a set of more # specialized configs depending on build flags. This pattern is useful # because it allows a target to opt in to either a default set, or a more # specific set, while avoid duplicating the settings in two places. config("super_optimization") { cflags = [ ... ] } config("default_optimization") { if (optimize_everything) { configs = [ ":super_optimization" ] } else { configs = [ ":no_optimization" ] } }
The contents of the file for a generated_file target. See "gn help generated_file".
Valid for `rust_library` targets and `executable`, `static_library`, `shared_library`, and `source_set` targets that contain Rust sources. If crate_name is not set, then this rule will use the target name.
Valid for `rust_library` targets and `executable`, `static_library`, `shared_library`, and `source_set` targets that contain Rust sources. This file is usually the `main.rs` or `lib.rs` for binaries and libraries, respectively. If crate_root is not set, then this rule will look for a lib.rs file (or main.rs for executable) or a single file in sources, if sources contains only one file.
Valid for `rust_library` targets and `executable`, `static_library`, `shared_library`, and `source_set` targets that contain Rust sources. Options for this field are "cdylib", "staticlib", "proc-macro", and "dylib". This field sets the `crate-type` attribute for the `rustc` tool on static libraries, as well as the appropiate output extension in the `rust_output_extension` attribute. Since outputs must be explicit, the `lib` crate type (where the Rust compiler produces what it thinks is the appropriate library type) is not supported. It should be noted that the "dylib" crate type in Rust is unstable in the set of symbols it exposes, and most usages today are potentially wrong and will be broken in the future. Static libraries, rust libraries, and executables have this field set automatically.
Lists files or directories required to run the given target. These are typically data files or directories of data files. The paths are interpreted as being relative to the current build file. Since these are runtime dependencies, they do not affect which targets are built or when. To declare input files to a script, use "inputs". Appearing in the "data" section does not imply any special handling such as copying them to the output directory. This is just used for declaring runtime dependencies. Runtime dependencies can be queried using the "runtime_deps" category of "gn desc" or written during build generation via "--runtime-deps-list-file". GN doesn't require data files to exist at build-time. So actions that produce files that are in turn runtime dependencies can list those generated files both in the "outputs" list as well as the "data" list. By convention, directories are listed with a trailing slash: data = [ "test/data/" ] However, no verification is done on these so GN doesn't enforce this. The paths are just rebased and passed along when requested. Note: On iOS and macOS, create_bundle targets will not be recursed into when gathering data. See "gn help create_bundle" for details. See "gn help runtime_deps" for how these are used.
A list of target labels. Specifies dependencies of a target that are not actually linked into the current target. Such dependencies will be built and will be available at runtime. This is normally used for things like plugins or helper programs that a target needs at runtime. Note: On iOS and macOS, create_bundle targets will not be recursed into when gathering data_deps. See "gn help create_bundle" for details. See also "gn help deps" and "gn help data".
executable("foo") { deps = [ "//base" ] data_deps = [ "//plugins:my_runtime_plugin" ] }
These keys are used to identify metadata to collect. If a walked target defines this key in its metadata, its value will be appended to the resulting collection. See "gn help generated_file".
A list of strings These strings will be passed to the C/C++ compiler as #defines. The strings may or may not include an "=" to assign a value.
1. Those set on the current target (not in a config). 2. Those set on the "configs" on the target in order that the configs appear in the list. 3. Those set on the "all_dependent_configs" on the target in order that the configs appear in the list. 4. Those set on the "public_configs" on the target in order that those configs appear in the list. 5. all_dependent_configs pulled from dependencies, in the order of the "deps" list. This is done recursively. If a config appears more than once, only the first occurence will be used. 6. public_configs pulled from dependencies, in the order of the "deps" list. If a dependency is public, they will be applied recursively.
defines = [ "AWESOME_FEATURE", "LOG_LEVEL=3" ]
If nonempty, this string specifies that the current action or action_foreach target will generate the given ".d" file containing the dependencies of the input. Empty or unset means that the script doesn't generate the files. A depfile should be used only when a target depends on files that are not already specified by a target's inputs and sources. Likewise, depfiles should specify only those dependencies not already included in sources or inputs. The .d file should go in the target output directory. If you have more than one source file that the script is being run over, you can use the output file expansions described in "gn help action_foreach" to name the .d file according to the input. The format is that of a Makefile and all paths must be relative to the root build directory. Only one output may be listed and it must match the first output of the action. Although depfiles are created by an action, they should not be listed in the action's "outputs" unless another target will use the file as an input.
action_foreach("myscript_target") { script = "myscript.py" sources = [ ... ] # Locate the depfile in the output directory named like the # inputs but with a ".d" appended. depfile = "$relative_target_output_dir/{{source_name}}.d" # Say our script uses "-o <d file>" to indicate the depfile. args = [ "{{source}}", "-o", depfile ] }
A list of target labels. Specifies private dependencies of a target. Private dependencies are propagated up the dependency tree and linked to dependent targets, but do not grant the ability to include headers from the dependency. Public configs are not forwarded.
Source sets, shared libraries, and non-complete static libraries will be propagated up the dependency tree across groups, non-complete static libraries and source sets. Executables, shared libraries, and complete static libraries will link all propagated targets and stop propagation. Actions and copy steps also stop propagation, allowing them to take a library as an input but not force dependents to link to it. Propagation of all_dependent_configs and public_configs happens independently of target type. all_dependent_configs are always propagated across all types of targets, and public_configs are always propagated across public deps of all types of targets. Data dependencies are propagated differently. See "gn help data_deps" and "gn help runtime_deps". See also "public_deps".
A list, each value being a scope indicating a pair of crate name and the path to the Rust library. These libraries will be passed as `--extern crate_name=path` to compiler invocation containing the current target.
executable("foo") { sources = [ "main.rs" ] externs = [{ crate_name = "bar", path = "path/to/bar.rlib" }] } This target would compile the `foo` crate with the following `extern` flag: `--extern bar=path/to/bar.rlib`.
A list of source directories. The directories in this list will be added to the framework search path for the files in the affected target.
1. Those set on the current target (not in a config). 2. Those set on the "configs" on the target in order that the configs appear in the list. 3. Those set on the "all_dependent_configs" on the target in order that the configs appear in the list. 4. Those set on the "public_configs" on the target in order that those configs appear in the list. 5. all_dependent_configs pulled from dependencies, in the order of the "deps" list. This is done recursively. If a config appears more than once, only the first occurence will be used. 6. public_configs pulled from dependencies, in the order of the "deps" list. If a dependency is public, they will be applied recursively.
framework_dirs = [ "src/include", "//third_party/foo" ]
A list of framework names. The frameworks named in that list will be linked with any dynamic link type target.
1. Those set on the current target (not in a config). 2. Those set on the "configs" on the target in order that the configs appear in the list. 3. Those set on the "all_dependent_configs" on the target in order that the configs appear in the list. 4. Those set on the "public_configs" on the target in order that those configs appear in the list. 5. all_dependent_configs pulled from dependencies, in the order of the "deps" list. This is done recursively. If a config appears more than once, only the first occurence will be used. 6. public_configs pulled from dependencies, in the order of the "deps" list. If a dependency is public, they will be applied recursively.
frameworks = [ "Foundation.framework", "Foo.framework" ]
A list of label patterns (see "gn help label_pattern") that allow dependent targets to include private headers. Applies to all binary targets. Normally if a target lists headers in the "public" list (see "gn help public"), other headers are implicitly marked as private. Private headers can not be included by other targets, even with a public dependency path. The "gn check" function performs this validation. A friend declaration allows one or more targets to include private headers. This is useful for things like unit tests that are closely associated with a target and require internal knowledge without opening up all headers to be included by all dependents. A friend target does not allow that target to include headers when no dependency exists. A public dependency path must still exist between two targets to include any headers from a destination target. The friend annotation merely allows the use of headers that would otherwise be prohibited because they are private. The friend annotation is matched only against the target containing the file with the include directive. Friend annotations are not propagated across public or private dependencies. Friend annotations do not affect visibility.
static_library("lib") { # This target can include our private headers. friend = [ ":unit_tests" ] public = [ "public_api.h", # Normal public API for dependent targets. ] # Private API and sources. sources = [ "a_source_file.cc", # Normal targets that depend on this one won't be able to include this # because this target defines a list of "public" headers. Without the # "public" list, all headers are implicitly public. "private_api.h", ] } executable("unit_tests") { sources = [ # This can include "private_api.h" from the :lib target because it # depends on that target and because of the friend annotation. "my_test.cc", ] deps = [ ":lib", # Required for the include to be allowed. ] }
A list of source directories. The directories in this list will be added to the include path for the files in the affected target.
1. Those set on the current target (not in a config). 2. Those set on the "configs" on the target in order that the configs appear in the list. 3. Those set on the "all_dependent_configs" on the target in order that the configs appear in the list. 4. Those set on the "public_configs" on the target in order that those configs appear in the list. 5. all_dependent_configs pulled from dependencies, in the order of the "deps" list. This is done recursively. If a config appears more than once, only the first occurence will be used. 6. public_configs pulled from dependencies, in the order of the "deps" list. If a dependency is public, they will be applied recursively.
include_dirs = [ "src/include", "//third_party/foo" ]
Inputs are compile-time dependencies of the current target. This means that all inputs must be available before compiling any of the sources or executing any actions. Inputs are typically only used for action and action_foreach targets.
For action and action_foreach targets, inputs should be the inputs to script that don't vary. These should be all .py files that the script uses via imports (the main script itself will be an implicit dependency of the action so need not be listed). For action targets, inputs and sources are treated the same, but from a style perspective, it's recommended to follow the same rule as action_foreach and put helper files in the inputs, and the data used by the script (if any) in sources. Note that another way to declare input dependencies from an action is to have the action write a depfile (see "gn help depfile"). This allows the script to dynamically write input dependencies, that might not be known until actually executing the script. This is more efficient than doing processing while running GN to determine the inputs, and is easier to keep in-sync than hardcoding the list.
It may be tempting to write a script that enumerates all files in a directory as inputs. Don't do this! Even if you specify all the files in the inputs or sources in the GN target (or worse, enumerate the files in an exec_script call when running GN, which will be slow), the dependencies will be broken. The problem happens if a file is ever removed because the inputs are not listed on the command line to the script. Because the script hasn't changed and all inputs are up to date, the script will not re-run and you will get a stale build. Instead, either list all inputs on the command line to the script, or if there are many, create a separate list file that the script reads. As long as this file is listed in the inputs, the build will detect when it has changed in any way and the action will re-run.
Any input dependencies will be resolved before compiling any sources or linking the target. Normally, all actions that a target depends on will be run before any files in a target are compiled. So if you depend on generated headers, you do not typically need to list them in the inputs section. Inputs for binary targets will be treated as implicit dependencies, meaning that changes in any of the inputs will force all sources in the target to be recompiled. If an input only applies to a subset of source files, you may want to split those into a separate target to avoid unnecessary recompiles.
action("myscript") { script = "domything.py" inputs = [ "input.data" ] }
A list of strings. These flags are passed on the command-line to the linker and generally specify various linking options. Most targets will not need these and will use "libs" and "lib_dirs" instead. ldflags are NOT pushed to dependents, so applying ldflags to source sets or static libraries will be a no-op. If you want to apply ldflags to dependent targets, put them in a config and set it in the all_dependent_configs or public_configs.
1. Those set on the current target (not in a config). 2. Those set on the "configs" on the target in order that the configs appear in the list. 3. Those set on the "all_dependent_configs" on the target in order that the configs appear in the list. 4. Those set on the "public_configs" on the target in order that those configs appear in the list. 5. all_dependent_configs pulled from dependencies, in the order of the "deps" list. This is done recursively. If a config appears more than once, only the first occurence will be used. 6. public_configs pulled from dependencies, in the order of the "deps" list. If a dependency is public, they will be applied recursively.
A list of directories. Specifies additional directories passed to the linker for searching for the required libraries. If an item is not an absolute path, it will be treated as being relative to the current build file. libs and lib_dirs work differently than other flags in two respects. First, they are inherited across static library boundaries until a shared library or executable target is reached. Second, they are uniquified so each one is only passed once (the first instance of it will be the one used).
1. Those set on the current target (not in a config). 2. Those set on the "configs" on the target in order that the configs appear in the list. 3. Those set on the "all_dependent_configs" on the target in order that the configs appear in the list. 4. Those set on the "public_configs" on the target in order that those configs appear in the list. 5. all_dependent_configs pulled from dependencies, in the order of the "deps" list. This is done recursively. If a config appears more than once, only the first occurence will be used. 6. public_configs pulled from dependencies, in the order of the "deps" list. If a dependency is public, they will be applied recursively. For "libs" and "lib_dirs" only, the values propagated from dependencies (as described above) are applied last assuming they are not already in the list.
lib_dirs = [ "/usr/lib/foo", "lib/doom_melon" ]
A list of library names or library paths. These libraries will be linked into the final binary (executable or shared library) containing the current target. libs and lib_dirs work differently than other flags in two respects. First, they are inherited across static library boundaries until a shared library or executable target is reached. Second, they are uniquified so each one is only passed once (the first instance of it will be the one used).
There are several different things that can be expressed in libs: File paths Values containing '/' will be treated as references to files in the checkout. They will be rebased to be relative to the build directory and specified in the "libs" for linker tools. This facility should be used for libraries that are checked in to the version control. For libraries that are generated by the build, use normal GN deps to link them. System libraries Values not containing '/' will be treated as system library names. These will be passed unmodified to the linker and prefixed with the "lib_switch" attribute of the linker tool. Generally you would set the "lib_dirs" so the given library is found. Your BUILD.gn file should not specify the switch (like "-l"): this will be encoded in the "lib_switch" of the tool.
1. Those set on the current target (not in a config). 2. Those set on the "configs" on the target in order that the configs appear in the list. 3. Those set on the "all_dependent_configs" on the target in order that the configs appear in the list. 4. Those set on the "public_configs" on the target in order that those configs appear in the list. 5. all_dependent_configs pulled from dependencies, in the order of the "deps" list. This is done recursively. If a config appears more than once, only the first occurence will be used. 6. public_configs pulled from dependencies, in the order of the "deps" list. If a dependency is public, they will be applied recursively. For "libs" and "lib_dirs" only, the values propagated from dependencies (as described above) are applied last assuming they are not already in the list.
On Windows: libs = [ "ctl3d.lib" ] On Linux: libs = [ "ld" ]
Metadata is a collection of keys and values relating to a particular target. Values must be lists, allowing for sane and predictable collection behavior. Generally, these keys will include three types of lists: lists of ordinary strings, lists of filenames intended to be rebased according to their particular source directory, and lists of target labels intended to be used as barriers to the walk. Verfication of these categories occurs at walk time, not creation time (since it is not clear until the walk which values are intended for which purpose).
group("doom_melon") { metadata = { # These keys are not built in to GN but are interpreted when consuming # metadata. my_barrier = [] my_files = [ "a.txt", "b.txt" ] } }
Valid for binary targets that contain Swift sources. If module_name is not set, then this rule will use the target name.
Controls how the "contents" of a generated_file target is formatted. See "gn help io_conversion".
For library and executable targets, overrides the directory for the final output. This must be in the root_build_dir or a child thereof. This should generally be in the root_out_dir or a subdirectory thereof (the root_out_dir will be the same as the root_build_dir for the default toolchain, and will be a subdirectory for other toolchains). Not putting the output in a subdirectory of root_out_dir can result in collisions between different toolchains, so you will need to take steps to ensure that your target is only present in one toolchain. Normally the toolchain specifies the output directory for libraries and executables (see "gn help tool"). You will have to consult that for the default location. The default location will be used if output_dir is undefined or empty.
shared_library("doom_melon") { output_dir = "$root_out_dir/plugin_libs" ... }
Normally the file extension for a target is based on the target type and the operating system, but in rare cases you will need to override the name (for example to use "libfreetype.so.6" instead of libfreetype.so on Linux). This value should not include a leading dot. If undefined, the default specified on the tool will be used. If set to the empty string, no output extension will be used. The output_extension will be used to set the "{{output_extension}}" expansion which the linker tool will generally use to specify the output file name. See "gn help tool".
shared_library("freetype") { if (is_linux) { # Call the output "libfreetype.so.6" output_extension = "so.6" } ... } # On Windows, generate a "mysettings.cpl" control panel applet. Control panel # applets are actually special shared libraries. if (is_win) { shared_library("mysettings") { output_extension = "cpl" ... } }
Normally the output name of a target will be based on the target name, so the target "//foo/bar:bar_unittests" will generate an output file such as "bar_unittests.exe" (using Windows as an example). Sometimes you will want an alternate name to avoid collisions or if the internal name isn't appropriate for public distribution. The output name should have no extension or prefixes, these will be added using the default system rules. For example, on Linux an output name of "foo" will produce a shared library "libfoo.so". There is no way to override the output prefix of a linker tool on a per- target basis. If you need more flexibility, create a copy target to produce the file you want. This variable is valid for all binary output target types.
static_library("doom_melon") { output_name = "fluffy_bunny" }
A boolean that overrides the output prefix for a target. Defaults to false. Some systems use prefixes for the names of the final target output file. The normal example is "libfoo.so" on Linux for a target named "foo". The output prefix for a given target type is specified on the linker tool (see "gn help tool"). Sometimes this prefix is undesired. See also "gn help output_extension".
shared_library("doom_melon") { # Normally this will produce "libdoom_melon.so" on Linux. Setting this flag # will produce "doom_melon.so". output_prefix_override = true ... }
Outputs is valid for "copy", "action", and "action_foreach" target types and indicates the resulting files. Outputs must always refer to files in the build directory. copy Copy targets should have exactly one entry in the outputs list. If there is exactly one source, this can be a literal file name or a source expansion. If there is more than one source, this must contain a source expansion to map a single input name to a single output name. See "gn help copy". action_foreach Action_foreach targets must always use source expansions to map input files to output files. There can be more than one output, which means that each invocation of the script will produce a set of files (presumably based on the name of the input file). See "gn help action_foreach". action Action targets (excluding action_foreach) must list literal output file(s) with no source expansions. See "gn help action".
Valid for create_bundle target, corresponds to the path for the partial Info.plist created by the asset catalog compiler that needs to be merged with the application Info.plist (usually done by the code signing script). The file will be generated regardless of whether the asset compiler has been invoked or not. See "gn help create_bundle".
A fully-qualified label representing the pool that will be used for the action. Pools are defined using the pool() {...} declaration.
action("action") { pool = "//build:custom_pool" ... }
Precompiled headers will be used when a target specifies this value, or a config applying to this target specifies this value. In addition, the tool corresponding to the source files must also specify precompiled headers (see "gn help tool"). The tool will also specify what type of precompiled headers to use, by setting precompiled_header_type to either "gcc" or "msvc". The precompiled header/source variables can be specified on a target or a config, but must be the same for all configs applying to a given target since a target can only have one precompiled header. If you use both C and C++ sources, the precompiled header and source file will be compiled once per language. You will want to make sure to wrap C++ includes in __cplusplus #ifdefs so the file will compile in C mode.
When using GCC-style precompiled headers, "precompiled_source" contains the path of a .h file that is precompiled and then included by all source files in targets that set "precompiled_source". The value of "precompiled_header" is not used with GCC-style precompiled headers.
When using MSVC-style precompiled headers, the "precompiled_header" value is a string corresponding to the header. This is NOT a path to a file that GN recognises, but rather the exact string that appears in quotes after an #include line in source code. The compiler will match this string against includes or forced includes (/FI). MSVC also requires a source file to compile the header with. This must be specified by the "precompiled_source" value. In contrast to the header value, this IS a GN-style file name, and tells GN which source file to compile to make the .pch file used for subsequent compiles. For example, if the toolchain specifies MSVC headers: toolchain("vc_x64") { ... tool("cxx") { precompiled_header_type = "msvc" ... You might make a config like this: config("use_precompiled_headers") { precompiled_header = "build/precompile.h" precompiled_source = "//build/precompile.cc" # Either your source files should #include "build/precompile.h" # first, or you can do this to force-include the header. cflags = [ "/FI$precompiled_header" ] } And then define a target that uses the config: executable("doom_melon") { configs += [ ":use_precompiled_headers" ] ...
See "gn help precompiled_header".
The source file that goes along with the precompiled_header when using "msvc"-style precompiled headers. It will be implicitly added to the sources of the target. See "gn help precompiled_header".
Correspond to the type of the product of a create_bundle target. Only meaningful to Xcode (used as part of the Xcode project generation). When generating Xcode project files, only create_bundle target with a non-empty product_type will have a corresponding target in Xcode project.
A list of files that other targets can include. These permissions are checked via the "check" command (see "gn help check"). If no public files are declared, other targets (assuming they have visibility to depend on this target) can include any file in the sources list. If this variable is defined on a target, dependent targets may only include files on this whitelist unless that target is marked as a friend (see "gn help friend"). Header file permissions are also subject to visibility. A target must be visible to another target to include any files from it at all and the public headers indicate which subset of those files are permitted. See "gn help visibility" for more. Public files are inherited through the dependency tree. So if there is a dependency A -> B -> C, then A can include C's public headers. However, the same is NOT true of visibility, so unless A is in C's visibility list, the include will be rejected. GN only knows about files declared in the "sources" and "public" sections of targets. If a file is included that is not known to the build, it will be allowed. It is common for test targets to need to include private headers for their associated code. In this case, list the test target in the "friend" list of the target that owns the private header to allow the inclusion. See "gn help friend" for more. When a binary target has no explicit or implicit public headers (a "public" list is defined but is empty), GN assumes that the target can not propagate any compile-time dependencies up the dependency tree. In this case, the build can be parallelized more efficiently. Say there are dependencies: A (shared library) -> B (shared library) -> C (action). Normally C must complete before any source files in A can compile (because there might be generated includes). But when B explicitly declares no public headers, C can execute in parallel with A's compile steps. C must still be complete before any dependents link.
These exact files are public: public = [ "foo.h", "bar.h" ] No files are public (no targets may include headers from this one): # This allows starting compilation in dependent targets earlier. public = []
A list of config labels. Targets directly depending on this one will have the configs listed in this variable added to them. These configs will also apply to the current target. Generally, public configs are used to apply defines and include directories necessary to compile this target's header files. See also "gn help all_dependent_configs".
Public configs are applied to all targets that depend directly on this one. These dependant targets can further push this target's public configs higher in the dependency tree by depending on it via public_deps (see "gn help public_deps"). static_library("toplevel") { # This target will get "my_config" applied to it. However, since this # target uses "deps" and not "public_deps", targets that depend on this # one won't get it. deps = [ ":intermediate" ] } static_library("intermediate") { # Depending on "lower" in any way will apply "my_config" to this target. # Additionall, since this target depends on "lower" via public_deps, # targets that depend on this one will also get "my_config". public_deps = [ ":lower" ] } static_library("lower") { # This will get applied to all targets that depend on this one. public_configs = [ ":my_config" ] } Public config propagation happens in a second phase once a target and all of its dependencies have been resolved. Therefore, a target will not see these force-added configs in their "configs" variable while the script is running, and they can not be removed. As a result, this capability should generally only be used to add defines and include directories rather than setting complicated flags that some targets may not want. Public configs may or may not be propagated across toolchain boundaries depending on the value of the propagates_configs flag (see "gn help toolchain") on the toolchain of the target declaring the public_config.
If you want the config to apply to targets that depend on this one, but NOT this one, define an extra layer of indirection using a group: # External targets depend on this group. group("my_target") { # Config to apply to all targets that depend on this one. public_configs = [ ":external_settings" ] deps = [ ":internal_target" ] } # Internal target to actually compile the sources. static_library("internal_target") { # Force all external targets to depend on the group instead of directly # on this so the "external_settings" config will get applied. visibility = [ ":my_target" ] ... }
1. Those set on the current target (not in a config). 2. Those set on the "configs" on the target in order that the configs appear in the list. 3. Those set on the "all_dependent_configs" on the target in order that the configs appear in the list. 4. Those set on the "public_configs" on the target in order that those configs appear in the list. 5. all_dependent_configs pulled from dependencies, in the order of the "deps" list. This is done recursively. If a config appears more than once, only the first occurence will be used. 6. public_configs pulled from dependencies, in the order of the "deps" list. If a dependency is public, they will be applied recursively.
Public dependencies are like private dependencies (see "gn help deps") but additionally express that the current target exposes the listed deps as part of its public API. This has several ramifications: - public_configs that are part of the dependency are forwarded to direct dependents. - Public headers in the dependency are usable by dependents (includes do not require a direct dependency or visibility). - If the current target is a shared library, other shared libraries that it publicly depends on (directly or indirectly) are propagated up the dependency tree to dependents for linking. See also "gn help public_configs".
Say you have three targets: A -> B -> C. C's visibility may allow B to depend on it but not A. Normally, this would prevent A from including any headers from C, and C's public_configs would apply only to B. If B lists C in its public_deps instead of regular deps, A will now inherit C's public_configs and the ability to include C's public headers. Generally if you are writing a target B and you include C's headers as part of B's public headers, or targets depending on B should consider B and C to be part of a unit, you should use public_deps instead of deps.
# This target can include files from "c" but not from # "super_secret_implementation_details". executable("a") { deps = [ ":b" ] } shared_library("b") { deps = [ ":super_secret_implementation_details" ] public_deps = [ ":c" ] }
A boolean that triggers a rebase of collected metadata strings based on their declared file. Defaults to false. Metadata generally declares files as strings relative to the local build file. However, this data is often used in other contexts, and so setting this flag will force the metadata collection to be rebased according to the local build file's location and thus allow the filename to be used anywhere. Setting this flag will raise an error if any target's specified metadata is not a string value. See also "gn help generated_file".
Sometimes the arguments passed to a script can be too long for the system's command-line capabilities. This is especially the case on Windows where the maximum command-line length is less than 8K. A response file allows you to pass an unlimited amount of data to a script in a temporary file for an action or action_foreach target. If the response_file_contents variable is defined and non-empty, the list will be treated as script args (including possibly substitution patterns) that will be written to a temporary file at build time. The name of the temporary file will be substituted for "{{response_file_name}}" in the script args. The response file contents will always be quoted and escaped according to Unix shell rules. To parse the response file, the Python script should use "shlex.split(file_contents)".
action("process_lots_of_files") { script = "process.py", inputs = [ ... huge list of files ... ] # Write all the inputs to a response file for the script. Also, # make the paths relative to the script working directory. response_file_contents = rebase_path(inputs, root_build_dir) # The script expects the name of the response file in --file-list. args = [ "--enable-foo", "--file-list={{response_file_name}}", ] }
An absolute or buildfile-relative file name of a Python script to run for a action and action_foreach targets (see "gn help action" and "gn help action_foreach").
A list of files. Non-absolute paths will be resolved relative to the current build file.
For binary targets (source sets, executables, and libraries), the known file types will be compiled with the associated tools. Unknown file types and headers will be skipped. However, you should still list all C/C+ header files so GN knows about the existence of those files for the purposes of include checking. As a special case, a file ending in ".def" will be treated as a Windows module definition file. It will be appended to the link line with a preceding "/DEF:" string. There must be at most one .def file in a target and they do not cross dependency boundaries (so specifying a .def file in a static library or source set will have no effect on the executable or shared library they're linked into). For Rust targets that do not specify a crate_root, then the crate_root will look for a lib.rs file (or main.rs for executable) or a single file in sources, if sources contains only one file.
action_foreach The sources are the set of files that the script will be executed over. The script will run once per file. action The sources will be treated the same as inputs. See "gn help inputs" for more information and usage advice. copy The source are the source files to copy.
A list of strings. "swiftflags" are passed to any invocation of a tool that takes an .swift file as input.
1. Those set on the current target (not in a config). 2. Those set on the "configs" on the target in order that the configs appear in the list. 3. Those set on the "all_dependent_configs" on the target in order that the configs appear in the list. 4. Those set on the "public_configs" on the target in order that those configs appear in the list. 5. all_dependent_configs pulled from dependencies, in the order of the "deps" list. This is done recursively. If a config appears more than once, only the first occurence will be used. 6. public_configs pulled from dependencies, in the order of the "deps" list. If a dependency is public, they will be applied recursively.
Boolean. Defaults to false. When a target is marked "testonly = true", it must only be depended on by other test-only targets. Otherwise, GN will issue an error that the depenedency is not allowed. This feature is intended to prevent accidentally shipping test code in a final product.
source_set("test_support") { testonly = true ... }
A list of labels and label patterns that define which targets can depend on the current one. These permissions are checked via the "check" command (see "gn help check"). If visibility is not defined, it defaults to public ("*"). If visibility is defined, only the targets with labels that match it can depend on the current target. The empty list means no targets can depend on the current target. Tip: Often you will want the same visibility for all targets in a BUILD file. In this case you can just put the definition at the top, outside of any target, and the targets will inherit that scope and see the definition.
See "gn help label_pattern" for more details on what types of patterns are supported. If a toolchain is specified, only targets in that toolchain will be matched. If a toolchain is not specified on a pattern, targets in all toolchains will be matched.
Only targets in the current buildfile ("private"): visibility = [ ":*" ] No targets (used for targets that should be leaf nodes): visibility = [] Any target ("public", the default): visibility = [ "*" ] All targets in the current directory and any subdirectory: visibility = [ "./*" ] Any target in "//bar/BUILD.gn": visibility = [ "//bar:*" ] Any target in "//bar/" or any subdirectory thereof: visibility = [ "//bar/*" ] Just these specific targets: visibility = [ ":mything", "//foo:something_else" ] Any target in the current directory and any subdirectory thereof, plus any targets in "//bar/" and any subdirectory thereof. visibility = [ "./*", "//bar/*" ]
Defaults to [""]. These keys are used to control the next step in a collection walk, acting as barriers. If a specified key is defined in a target's metadata, the walk will use the targets listed in that value to determine which targets are walked. If no walk_keys are specified for a generated_file target (i.e. "[""]"), the walk will touch all deps and data_deps of the specified target recursively. See "gn help generated_file".
A list of framework names. The frameworks named in that list will be weak linked with any dynamic link type target. Weak linking instructs the dynamic loader to attempt to load the framework, but if it is not able to do so, it leaves any imported symbols unresolved. This is typically used when a framework is present in a new version of an SDK but not on older versions of the OS that the software runs on.
1. Those set on the current target (not in a config). 2. Those set on the "configs" on the target in order that the configs appear in the list. 3. Those set on the "all_dependent_configs" on the target in order that the configs appear in the list. 4. Those set on the "public_configs" on the target in order that those configs appear in the list. 5. all_dependent_configs pulled from dependencies, in the order of the "deps" list. This is done recursively. If a config appears more than once, only the first occurence will be used. 6. public_configs pulled from dependencies, in the order of the "deps" list. If a dependency is public, they will be applied recursively.
weak_frameworks = [ "OnlyOnNewerOSes.framework" ]
Does not synchronously write the file, but rather schedules it to be written at the end of generation. If the file exists and the contents are identical to that being written, the file will not be updated. This will prevent unnecessary rebuilds of targets that depend on this file. Path must be within the output directory. See "gn help runtime_deps" for how the runtime dependencies are computed. The format of this file will list one file per line with no escaping. The files will be relative to the root_build_dir. The first line of the file will be the main output file of the target itself. The file contents will be the same as requesting the runtime deps be written on the command line (see "gn help --runtime-deps-list-file").
A list of strings. Valid for create_bundle target. Those flags are directly passed to xcassets compiler, corresponding to {{xcasset_compiler_flags}} substitution in compile_xcassets tool.
The value defined in this scope will be copied to the EXTRA_ATTRIBUTES property of the generated Xcode project. They are only meaningful when generating with --ide=xcode. See "gn help create_bundle" for more information.
Each unit and ui test target must have a test application target, and this value is used to specify the relationship. Only meaningful to Xcode (used as part of the Xcode project generation). See "gn help create_bundle" for more information.
create_bundle("chrome_xctest") { test_application_name = "chrome" ... }
Build arguments are variables passed in from outside of the build that build files can query to determine how the build works.
First, system default arguments are set based on the current system. The built-in arguments are: - host_cpu - host_os - current_cpu - current_os - target_cpu - target_os Next, project-specific overrides are applied. These are specified inside the default_args variable of //.gn. See "gn help dotfile" for more. If specified, arguments from the --args command line flag are used. If that flag is not specified, args from previous builds in the build directory will be used (this is in the file args.gn in the build directory). Last, for targets being compiled with a non-default toolchain, the toolchain overrides are applied. These are specified in the toolchain_args section of a toolchain definition. The use-case for this is that a toolchain may be building code for a different platform, and that it may want to always specify Posix, for example. See "gn help toolchain" for more. If you specify an override for a build argument that never appears in a "declare_args" call, a nonfatal error will be displayed.
gn args out/FooBar Create the directory out/FooBar and open an editor. You would type something like this into that file: enable_doom_melon=false os="android" gn gen out/FooBar --args="enable_doom_melon=true os=\"android\"" This will overwrite the build directory with the given arguments. (Note that the quotes inside the args command will usually need to be escaped for your shell to pass through strings values.)
If you want to use an argument, you use declare_args() and specify default values. These default values will apply if none of the steps listed in the "How build arguments are set" section above apply to the given argument, but the defaults will not override any of these. Often, the root build config file will declare global arguments that will be passed to all buildfiles. Individual build files can also specify arguments that apply only to those files. It is also useful to specify build args in an "import"-ed file if you want such arguments to apply to multiple buildfiles.
When gn starts, it will search the current directory and parent directories for a file called ".gn". This indicates the source root. You can override this detection by using the --root command-line argument The .gn file in the source root will be executed. The syntax is the same as a buildfile, but with very limited build setup-specific meaning. If you specify --root, by default GN will look for the file .gn in that directory. If you want to specify a different file, you can additionally pass --dotfile: gn gen out/Debug --root=/home/build --dotfile=/home/my_gn_file.gn
arg_file_template [optional] Path to a file containing the text that should be used as the default args.gn content when you run `gn args`. buildconfig [required] Path to the build config file. This file will be used to set up the build file execution environment for each toolchain. check_targets [optional] A list of labels and label patterns that should be checked when running "gn check" or "gn gen --check". If neither check_targets or no_check_targets (see below) is specified, all targets will be checked. It is an error to specify both check_targets and no_check_targets. If it is the empty list, no targets will be checked. To bypass this list, request an explicit check of targets, like "//*". The format of this list is identical to that of "visibility" so see "gn help visibility" for examples. no_check_targets [optional] A list of labels and label patterns that should *not* be checked when running "gn check" or "gn gen --check". All other targets will be checked. If neither check_targets (see above) or no_check_targets is specified, all targets will be checked. It is an error to specify both check_targets and no_check_targets. The format of this list is identical to that of "visibility" so see "gn help visibility" for examples. check_system_includes [optional] Boolean to control whether system style includes are checked by default when running "gn check" or "gn gen --check". System style includes are includes that use angle brackets <> instead of double quotes "". If this setting is omitted or set to false, these includes will be ignored by default. They can be checked explicitly by running "gn check --check-system" or "gn gen --check=system" exec_script_whitelist [optional] A list of .gn/.gni files (not labels) that have permission to call the exec_script function. If this list is defined, calls to exec_script will be checked against this list and GN will fail if the current file isn't in the list. This is to allow the use of exec_script to be restricted since is easy to use inappropriately. Wildcards are not supported. Files in the secondary_source tree (if defined) should be referenced by ignoring the secondary tree and naming them as if they are in the main tree. If unspecified, the ability to call exec_script is unrestricted. Example: exec_script_whitelist = [ "//base/BUILD.gn", "//build/my_config.gni", ] root [optional] Label of the root build target. The GN build will start by loading the build file containing this target name. This defaults to "//:" which will cause the file //BUILD.gn to be loaded. Note that build_file_extension applies to the default case as well. The command-line switch --root-target will override this value (see "gn help --root-target"). script_executable [optional] Path to specific Python executable or other interpreter to use in action targets and exec_script calls. By default GN searches the PATH for Python to execute these scripts. If set to the empty string, the path specified in action targets and exec_script calls will be executed directly. secondary_source [optional] Label of an alternate directory tree to find input files. When searching for a BUILD.gn file (or the build config file discussed above), the file will first be looked for in the source root. If it's not found, the secondary source root will be checked (which would contain a parallel directory hierarchy). This behavior is intended to be used when BUILD.gn files can't be checked in to certain source directories for whatever reason. The secondary source root must be inside the main source tree. default_args [optional] Scope containing the default overrides for declared arguments. These overrides take precedence over the default values specified in the declare_args() block, but can be overriden using --args or the args.gn file. This is intended to be used when subprojects declare arguments with default values that need to be changed for whatever reason. build_file_extension [optional] If set to a non-empty string, this is added to the name of all build files to load. GN will look for build files named "BUILD.$build_file_extension.gn". This is intended to be used during migrations or other situations where there are two independent GN builds in the same directories. ninja_required_version [optional] When set specifies the minimum required version of Ninja. The default required version is 1.7.2. Specifying a higher version might enable the use of some of newer features that can make the build more efficient.
buildconfig = "//build/config/BUILDCONFIG.gn" check_targets = [ "//doom_melon/*", # Check everything in this subtree. "//tools:mind_controlling_ant", # Check this specific target. ] root = "//:root" secondary_source = "//build/config/temporary_buildfiles/" default_args = { # Default to release builds for this project. is_debug = false is_component_build = false }
1. Look for ".gn" file (see "gn help dotfile") in the current directory and walk up the directory tree until one is found. Set this directory to be the "source root" and interpret this file to find the name of the build config file. 2. Execute the build config file identified by .gn to set up the global variables and default toolchain name. Any arguments, variables, defaults, etc. set up in this file will be visible to all files in the build. 3. Load the //BUILD.gn (in the source root directory). 4. Recursively evaluate rules and load BUILD.gn in other directories as necessary to resolve dependencies. If a BUILD file isn't found in the specified location, GN will look in the corresponding location inside the secondary_source defined in the dotfile (see "gn help dotfile"). 5. When a target's dependencies are resolved, write out the `.ninja` file to disk. 6. When all targets are resolved, write out the root build.ninja file. Note that the BUILD.gn file name may be modulated by .gn arguments such as build_file_extension.
Build files are loaded in parallel. This means it is impossible to interrogate a target from GN code for any information not derivable from its label (see "gn help label"). The exception is the get_target_outputs() function which requires the target being interrogated to have been defined previously in the same file. Targets are declared by their type and given a name: static_library("my_static_library") { ... target parameter definitions ... } There is also a generic "target" function for programmatically defined types (see "gn help target"). You can define new types using templates (see "gn help template"). A template defines some custom code that expands to one or more other targets. Before executing the code inside the target's { }, the target defaults are applied (see "gn help set_defaults"). It will inject implicit variable definitions that can be overridden by the target code as necessary. Typically this mechanism is used to inject a default set of configs that define the global compiler and linker flags.
All targets encountered in the default toolchain (see "gn help toolchain") will have build rules generated for them, even if no other targets reference them. Their dependencies must resolve and they will be added to the implicit "all" rule (see "gn help ninja_rules"). Targets in non-default toolchains will only be generated when they are required (directly or transitively) to build a target in the default toolchain. See also "gn help ninja_rules".
The only difference between "public_deps" and "deps" except for pushing configs around the build tree and allowing includes for the purposes of "gn check". A target's "data_deps" are guaranteed to be built whenever the target is built, but the ordering is not defined. The meaning of this is dependencies required at runtime. Currently data deps will be complete before the target is linked, but this is not semantically guaranteed and this is undesirable from a build performance perspective. Since we hope to change this in the future, do not rely on this behavior.
GN build files are read as sequences of tokens. While splitting the file into tokens, the next token is the longest sequence of characters that form a valid token.
White space is comprised of spaces (U+0020), horizontal tabs (U+0009), carriage returns (U+000D), and newlines (U+000A). Comments start at the character "#" and stop at the next newline. White space and comments are ignored except that they may separate tokens that would otherwise combine into a single token.
Identifiers name variables and functions. identifier = letter { letter | digit } . letter = "A" ... "Z" | "a" ... "z" | "_" . digit = "0" ... "9" .
The following keywords are reserved and may not be used as identifiers: else false if true
An integer literal represents a decimal integer value. integer = [ "-" ] digit { digit } . Leading zeros and negative zero are disallowed.
A string literal represents a string value consisting of the quoted characters with possible escape sequences and variable expansions. string = `"` { char | escape | expansion } `"` . escape = `\` ( "$" | `"` | char ) . BracketExpansion = "{" ( identifier | ArrayAccess | ScopeAccess ) "}" . Hex = "0x" [0-9A-Fa-f][0-9A-Fa-f] expansion = "$" ( identifier | BracketExpansion | Hex ) . char = /* any character except "$", `"`, or newline */ . After a backslash, certain sequences represent special characters: \" U+0022 quotation mark \$ U+0024 dollar sign \\ U+005C backslash All other backslashes represent themselves. To insert an arbitrary byte value, use $0xFF. For example, to insert a newline character: "Line one$0x0ALine two". An expansion will evaluate the variable following the '$' and insert a stringified version of it into the result. For example, to concat two path components with a slash separating them: "$var_one/$var_two" Use the "${var_one}" format to be explicitly deliniate the variable for otherwise-ambiguous cases.
The following character sequences represent punctuation: + += == != ( ) - -= < <= [ ] ! = > >= { } && || . ,
The input tokens form a syntax tree following a context-free grammar: File = StatementList . Statement = Assignment | Call | Condition . LValue = identifier | ArrayAccess | ScopeAccess . Assignment = LValue AssignOp Expr . Call = identifier "(" [ ExprList ] ")" [ Block ] . Condition = "if" "(" Expr ")" Block [ "else" ( Condition | Block ) ] . Block = "{" StatementList "}" . StatementList = { Statement } . ArrayAccess = identifier "[" Expr "]" . ScopeAccess = identifier "." identifier . Expr = UnaryExpr | Expr BinaryOp Expr . UnaryExpr = PrimaryExpr | UnaryOp UnaryExpr . PrimaryExpr = identifier | integer | string | Call | ArrayAccess | ScopeAccess | Block | "(" Expr ")" | "[" [ ExprList [ "," ] ] "]" . ExprList = Expr { "," Expr } . AssignOp = "=" | "+=" | "-=" . UnaryOp = "!" . BinaryOp = "+" | "-" // highest priority | "<" | "<=" | ">" | ">=" | "==" | "!=" | "&&" | "||" . // lowest priority All binary operators are left-associative.
The GN language is dynamically typed. The following types are used: - Boolean: Uses the keywords "true" and "false". There is no implicit conversion between booleans and integers. - Integers: All numbers in GN are signed 64-bit integers. - Strings: Strings are 8-bit with no enforced encoding. When a string is used to interact with other systems with particular encodings (like the Windows and Mac filesystems) it is assumed to be UTF-8. See "String literals" above for more. - Lists: Lists are arbitrary-length ordered lists of values. See "Lists" below for more. - Scopes: Scopes are like dictionaries that use variable names for keys. See "Scopes" below for more.
Lists are created with [] and using commas to separate items: mylist = [ 0, 1, 2, "some string" ] A comma after the last item is optional. Lists are dereferenced using 0-based indexing: mylist[0] += 1 var = mylist[2] Lists can be concatenated using the '+' and '+=' operators. Bare values can not be concatenated with lists, to add a single item, it must be put into a list of length one. Items can be removed from lists using the '-' and '-=' operators. This will remove all occurrences of every item in the right-hand list from the left-hand list. It is an error to remove an item not in the list. This is to prevent common typos and to detect dead code that is removing things that no longer apply. It is an error to use '=' to replace a nonempty list with another nonempty list. This is to prevent accidentally overwriting data when in most cases '+=' was intended. To overwrite a list on purpose, first assign it to the empty list: mylist = [] mylist = otherlist
All execution happens in the context of a scope which holds the current state (like variables). With the exception of loops and conditions, '{' introduces a new scope that has a parent reference to the old scope. Variable reads recursively search all nested scopes until the variable is found or there are no more scopes. Variable writes always go into the current scope. This means that after the closing '}' (again excepting loops and conditions), all local variables will be restored to the previous values. This also means that "foo = foo" can do useful work by copying a variable into the current scope that was defined in a containing scope. Scopes can also be assigned to variables. Such scopes can be created by functions like exec_script, when invoking a template (the template code refers to the variables set by the invoking code by the implicitly-created "invoker" scope), or explicitly like: empty_scope = {} myvalues = { foo = 21 bar = "something" } Inside such a scope definition can be any GN code including conditionals and function calls. After the close of the scope, it will contain all variables explicitly set by the code contained inside it. After this, the values can be read, modified, or added to: myvalues.foo += 2 empty_scope.new_thing = [ 1, 2, 3 ] Scope equality is defined as single-level scopes identical within the current scope. That is, all values in the first scope must be present and identical within the second, and vice versa. Note that this means inherited scopes are always unequal by definition.
Input and output conversions are arguments to file and process functions that specify how to convert data to or from external formats. The possible values for parameters specifying conversions are: "" (the default) input: Discard the result and return None. output: If value is a list, then "list lines"; otherwise "value". "list lines" input: Return the file contents as a list, with a string for each line. The newlines will not be present in the result. The last line may or may not end in a newline. After splitting, each individual line will be trimmed of whitespace on both ends. output: Renders the value contents as a list, with a string for each line. The newlines will not be present in the result. The last line will end in with a newline. "scope" input: Execute the block as GN code and return a scope with the resulting values in it. If the input was: a = [ "hello.cc", "world.cc" ] b = 26 and you read the result into a variable named "val", then you could access contents the "." operator on "val": sources = val.a some_count = val.b output: Renders the value contents as a GN code block, reversing the input result above. "string" input: Return the file contents into a single string. output: Render the value contents into a single string. The output is: a string renders with quotes, e.g. "str" an integer renders as a stringified integer, e.g. "6" a boolean renders as the associated string, e.g. "true" a list renders as a representation of its contents, e.g. "[\"str\", 6]" a scope renders as a GN code block of its values. If the Value was: Value val; val.a = [ "hello.cc", "world.cc" ]; val.b = 26 the resulting output would be: "{ a = [ \"hello.cc\", \"world.cc\" ] b = 26 }" "value" input: Parse the input as if it was a literal rvalue in a buildfile. Examples of typical program output using this mode: [ "foo", "bar" ] (result will be a list) or "foo bar" (result will be a string) or 5 (result will be an integer) Note that if the input is empty, the result will be a null value which will produce an error if assigned to a variable. output: Render the value contents as a literal rvalue. Strings render with escaped quotes. "json" input: Parse the input as a JSON and convert it to equivalent GN rvalue. output: Convert the Value to equivalent JSON value. The data type mapping is: a string in JSON maps to string in GN an integer in JSON maps to integer in GN a float in JSON is unsupported and will result in an error an object in JSON maps to scope in GN an array in JSON maps to list in GN a boolean in JSON maps to boolean in GN a null in JSON is unsupported and will result in an error Nota that the input dictionary keys have to be valid GN identifiers otherwise they will produce an error. "trim ..." (input only) Prefixing any of the other transformations with the word "trim" will result in whitespace being trimmed from the beginning and end of the result before processing. Examples: "trim string" or "trim list lines" Note that "trim value" is useless because the value parser skips whitespace anyway.
File patterns are VERY limited regular expressions. They must match the entire input string to be counted as a match. In regular expression parlance, there is an implicit "^...$" surrounding your input. If you want to match a substring, you need to use wildcards at the beginning and end. There are only two special tokens understood by the pattern matcher. Everything else is a literal. - "*" Matches zero or more of any character. It does not depend on the preceding character (in regular expression parlance it is equivalent to ".*"). - "\b" Matches a path boundary. This will match the beginning or end of a string, or a slash.
"*asdf*" Matches a string containing "asdf" anywhere. "asdf" Matches only the exact string "asdf". "*.cc" Matches strings ending in the literal ".cc". "\bwin/*" Matches "win/foo" and "foo/win/bar.cc" but not "iwin/foo".
A label pattern is a way of expressing one or more labels in a portion of the source tree. They are not general regular expressions. They can take the following forms only: - Explicit (no wildcard): "//foo/bar:baz" ":baz" - Wildcard target names: "//foo/bar:*" (all targets in the //foo/bar/BUILD.gn file) ":*" (all targets in the current build file) - Wildcard directory names ("*" is only supported at the end) "*" (all targets) "//foo/bar/*" (all targets in any subdir of //foo/bar) "./*" (all targets in the current build file or sub dirs) Any of the above forms can additionally take an explicit toolchain in parenthesis at the end of the label pattern. In this case, the toolchain must be fully qualified (no wildcards are supported in the toolchain name). "//foo:bar(//build/toolchain:mac)" An explicit target in an explicit toolchain. ":*(//build/toolchain/linux:32bit)" All targets in the current build file using the 32-bit Linux toolchain. "//foo/*(//build/toolchain:win)" All targets in //foo and any subdirectory using the Windows toolchain.
Everything that can participate in the dependency graph (targets, configs, and toolchains) are identified by labels. A common label looks like: //base/test:test_support This consists of a source-root-absolute path, a colon, and a name. This means to look for the thing named "test_support" in "base/test/BUILD.gn". You can also specify system absolute paths if necessary. Typically such paths would be specified via a build arg so the developer can specify where the component is on their system. /usr/local/foo:bar (Posix) /C:/Program Files/MyLibs:bar (Windows)
A canonical label includes the label of the toolchain being used. Normally, the toolchain label is implicitly inherited from the current execution context, but you can override this to specify cross-toolchain dependencies: //base/test:test_support(//build/toolchain/win:msvc) Here GN will look for the toolchain definition called "msvc" in the file "//build/toolchain/win" to know how to compile this target.
If you want to refer to something in the same buildfile, you can omit the path name and just start with a colon. This format is recommended for all same-file references. :base Labels can be specified as being relative to the current directory. Stylistically, we prefer to use absolute paths for all non-file-local references unless a build file needs to be run in different contexts (like a project needs to be both standalone and pulled into other projects in difference places in the directory hierarchy). source/plugin:myplugin ../net:url_request
If a name is unspecified, it will inherit the directory name. Stylistically, we prefer to omit the colon and name when possible: //net -> //net:net //tools/gn -> //tools/gn:gn
Metadata is information attached to targets throughout the dependency tree. GN allows for the collection of this data into files written during the generation step, enabling users to expose and aggregate this data based on the dependency tree.
Similar to the write_file() function, the generated_file target type creates a file in the specified location with the specified content. The primary difference between write_file() and this target type is that the write_file function does the file write at parse time, while the generated_file target type writes at target resolution time. See "gn help generated_file" for more detail. When written at target resolution time, generated_file enables GN to collect and write aggregated metadata from dependents. A generated_file target can declare either 'contents' to write statically known contents to a file or 'data_keys' to aggregate metadata and write the result to a file. It can also specify 'walk_keys' (to restrict the metadata collection), 'output_conversion', and 'rebase'.
Targets can declare a 'metadata' variable containing a scope, and this metadata may be collected and written out to a file specified by generated_file aggregation targets. The 'metadata' scope must contain only list values since the aggregation step collects a list of these values. During the target resolution, generated_file targets will walk their dependencies recursively, collecting metadata based on the specified 'data_keys'. 'data_keys' is specified as a list of strings, used by the walk to identify which variables in dependencies' 'metadata' scopes to collect. The walk begins with the listed dependencies of the 'generated_file' target. The 'metadata' scope for each dependency is inspected for matching elements of the 'generated_file' target's 'data_keys' list. If a match is found, the data from the dependent's matching key list is appended to the aggregate walk list. Note that this means that if more than one walk key is specified, the data in all of them will be aggregated into one list. From there, the walk will then recurse into the dependencies of each target it encounters, collecting the specified metadata for each. For example: group("a") { metadata = { doom_melon = [ "enable" ] my_files = [ "foo.cpp" ] my_extra_files = [ "bar.cpp" ] } deps = [ ":b" ] } group("b") { metadata = { my_files = [ "baz.cpp" ] } } generated_file("metadata") { outputs = [ "$root_build_dir/my_files.json" ] data_keys = [ "my_files", "my_extra_files" ] deps = [ ":a" ] } The above will produce the following file data: foo.cpp bar.cpp baz.cpp The dependency walk can be limited by using the 'walk_keys'. This is a list of labels that should be included in the walk. All labels specified here should also be in one of the deps lists. These keys act as barriers, where the walk will only recurse into the targets listed. An empty list in all specified barriers will end that portion of the walk. group("a") { metadata = { my_files = [ "foo.cpp" ] my_files_barrier = [ ":b" ] } deps = [ ":b", ":c" ] } group("b") { metadata = { my_files = [ "bar.cpp" ] } } group("c") { metadata = { my_files = [ "doom_melon.cpp" ] } } generated_file("metadata") { outputs = [ "$root_build_dir/my_files.json" ] data_keys = [ "my_files" ] walk_keys = [ "my_files_barrier" ] deps = [ ":a" ] } The above will produce the following file data (note that `doom_melon.cpp` is not included): foo.cpp bar.cpp A common example of this sort of barrier is in builds that have host tools built as part of the tree, but do not want the metadata from those host tools to be collected with the target-side code.
Metadata can be used to collect information about the different targets in the build, and so a common use is to provide post-build tooling with a set of data necessary to do aggregation tasks. For example, if each test target specifies the output location of its binary to run in a metadata field, that can be collected into a single file listing the locations of all tests in the dependency tree. A local build tool (or continuous integration infrastructure) can then use that file to know which tests exist, and where, and run them accordingly. Another use is in image creation, where a post-build image tool needs to know various pieces of information about the components it should include in order to put together the correct image.
All generated targets (see "gn help execution") will be added to an implicit build rule called "all" so "ninja all" will always compile everything. The default rule will be used by Ninja if no specific target is specified (just typing "ninja"). If there is a target named "default" in the root build file, it will be the default build rule, otherwise the implicit "all" rule will be used.
GN generates Ninja "phony" rules for targets in the default toolchain. The phony rules can collide with each other and with the names of generated files so are generated with the following priority: 1. Actual files generated by the build always take precedence. 2. Targets in the toplevel //BUILD.gn file. 3. Targets in toplevel directories matching the names of the directories. So "ninja foo" can be used to compile "//foo:foo". This only applies to the first level of directories since usually these are the most important (so this won't apply to "//foo/bar:bar"). 4. The short names of executables if there is only one executable with that short name. Use "ninja doom_melon" to compile the "//tools/fruit:doom_melon" executable. 5. The short names of all targets if there is only one target with that short name. 6. Full label name with no leading slashes. So you can use "ninja tools/fruit:doom_melon" to build "//tools/fruit:doom_melon". 7. Labels with an implicit name part (when the short names match the directory). So you can use "ninja foo/bar" to compile "//foo/bar:bar". These "phony" rules are provided only for running Ninja since this matches people's historical expectations for building. For consistency with the rest of the program, GN introspection commands accept explicit labels. To explicitly compile a target in a non-default toolchain, you must give Ninja the exact name of the output file relative to the build directory.
GN's header checker helps validate that the includes match the build dependency graph. Sometimes an include might be conditional or otherwise problematic, but you want to specifically allow it. In this case, it can be whitelisted. Include lines containing the substring "nogncheck" will be excluded from header checking. The most common case is a conditional include: #if defined(ENABLE_DOOM_MELON) #include "tools/doom_melon/doom_melon.h" // nogncheck #endif If the build file has a conditional dependency on the corresponding target that matches the conditional include, everything will always link correctly: source_set("mytarget") { ... if (enable_doom_melon) { defines = [ "ENABLE_DOOM_MELON" ] deps += [ "//tools/doom_melon" ] } But GN's header checker does not understand preprocessor directives, won't know it matches the build dependencies, and will flag this include as incorrect when the condition is false.
The topic "gn help check" has general information on how checking works and advice on fixing problems. Targets can also opt-out of checking, see "gn help check_includes".
Runtime dependencies of a target are exposed via the "runtime_deps" category of "gn desc" (see "gn help desc") or they can be written at build generation time via write_runtime_deps(), or --runtime-deps-list-file (see "gn help --runtime-deps-list-file"). To a first approximation, the runtime dependencies of a target are the set of "data" files, data directories, and the shared libraries from all transitive dependencies. Executables, shared libraries, and loadable modules are considered runtime dependencies of themselves.
Executable targets and those executable targets' transitive dependencies are not considered unless that executable is listed in "data_deps". Otherwise, GN assumes that the executable (and everything it requires) is a build-time dependency only.
Action and copy targets that are listed as "data_deps" will have all of their outputs and data files considered as runtime dependencies. Action and copy targets that are "deps" or "public_deps" will have only their data files considered as runtime dependencies. These targets can list an output file in both the "outputs" and "data" lists to force an output file as a runtime dependency in all cases. The different rules for deps and data_deps are to express build-time (deps) vs. run-time (data_deps) outputs. If GN counted all build-time copy steps as data dependencies, there would be a lot of extra stuff, and if GN counted all run-time dependencies as regular deps, the build's parallelism would be unnecessarily constrained. This rule can sometimes lead to unintuitive results. For example, given the three targets: A --[data_deps]--> B --[deps]--> ACTION GN would say that A does not have runtime deps on the result of the ACTION, which is often correct. But the purpose of the B target might be to collect many actions into one logic unit, and the "data"-ness of A's dependency is lost. Solutions: - List the outputs of the action in its data section (if the results of that action are always runtime files). - Have B list the action in data_deps (if the outputs of the actions are always runtime files). - Have B list the action in both deps and data deps (if the outputs might be used in both contexts and you don't care about unnecessary entries in the list of files required at runtime). - Split B into run-time and build-time versions with the appropriate "deps" for each.
The results of static_library or source_set targets are not considered runtime dependencies since these are assumed to be intermediate targets only. If you need to list a static library as a runtime dependency, you can manually compute the .a/.lib file name for the current platform and list it in the "data" list of a target (possibly on the static library target itself).
Linker tools can specify which of their outputs should be considered when computing the runtime deps by setting runtime_outputs. If this is unset on the tool, the default will be the first output only.
Source expansion is used for the action_foreach and copy target types to map source file names to output file names or arguments. To perform source expansion in the outputs, GN maps every entry in the sources to every entry in the outputs list, producing the cross product of all combinations, expanding placeholders (see below). Source expansion in the args works similarly, but performing the placeholder substitution produces a different set of arguments for each invocation of the script. If no placeholders are found, the outputs or args list will be treated as a static list of literal file names that do not depend on the sources. See "gn help copy" and "gn help action_foreach" for more on how this is applied.
This section discusses only placeholders for actions. There are other placeholders used in the definition of tools. See "gn help tool" for those. {{source}} The name of the source file including directory (*). This will generally be used for specifying inputs to a script in the "args" variable. "//foo/bar/baz.txt" => "../../foo/bar/baz.txt" {{source_file_part}} The file part of the source including the extension. "//foo/bar/baz.txt" => "baz.txt" {{source_name_part}} The filename part of the source file with no directory or extension. This will generally be used for specifying a transformation from a source file to a destination file with the same name but different extension. "//foo/bar/baz.txt" => "baz" {{source_dir}} The directory (*) containing the source file with no trailing slash. "//foo/bar/baz.txt" => "../../foo/bar" {{source_root_relative_dir}} The path to the source file's directory relative to the source root, with no leading "//" or trailing slashes. If the path is system-absolute, (beginning in a single slash) this will just return the path with no trailing slash. This value will always be the same, regardless of whether it appears in the "outputs" or "args" section. "//foo/bar/baz.txt" => "foo/bar" {{source_gen_dir}} The generated file directory (*) corresponding to the source file's path. This will be different than the target's generated file directory if the source file is in a different directory than the BUILD.gn file. "//foo/bar/baz.txt" => "gen/foo/bar" {{source_out_dir}} The object file directory (*) corresponding to the source file's path, relative to the build directory. this us be different than the target's out directory if the source file is in a different directory than the build.gn file. "//foo/bar/baz.txt" => "obj/foo/bar" {{source_target_relative}} The path to the source file relative to the target's directory. This will generally be used for replicating the source directory layout in the output directory. This can only be used in actions and bundle_data targets. It is an error to use in process_file_template where there is no "target". "//foo/bar/baz.txt" => "baz.txt"
Paths containing directories (except the source_root_relative_dir) will be different depending on what context the expansion is evaluated in. Generally it should "just work" but it means you can't concatenate strings containing these values with reasonable results. Details: source expansions can be used in the "outputs" variable, the "args" variable, and in calls to "process_file_template". The "args" are passed to a script which is run from the build directory, so these directories will relative to the build directory for the script to find. In the other cases, the directories will be source- absolute (begin with a "//") because the results of those expansions will be handled by GN internally.
Non-varying outputs: action("hardcoded_outputs") { sources = [ "input1.idl", "input2.idl" ] outputs = [ "$target_out_dir/output1.dat", "$target_out_dir/output2.dat" ] } The outputs in this case will be the two literal files given. Varying outputs: action_foreach("varying_outputs") { sources = [ "input1.idl", "input2.idl" ] outputs = [ "{{source_gen_dir}}/{{source_name_part}}.h", "{{source_gen_dir}}/{{source_name_part}}.cc" ] } Performing source expansion will result in the following output names: //out/Debug/obj/mydirectory/input1.h //out/Debug/obj/mydirectory/input1.cc //out/Debug/obj/mydirectory/input2.h //out/Debug/obj/mydirectory/input2.cc
Do "gn help --the_switch_you_want_help_on" for more. Individual commands may take command-specific switches not listed here. See the help on your specific command for more.
* --args: Specifies build arguments overrides. * --color: Force colored output. * --dotfile: Override the name of the ".gn" file. * --fail-on-unused-args: Treat unused build args as fatal errors. * --markdown: Write help output in the Markdown format. * --ninja-executable: Set the Ninja executable. * --nocolor: Force non-colored output. * -q: Quiet mode. Don't print output on success. * --root: Explicitly specify source root. * --root-target: Override the root target. * --runtime-deps-list-file: Save runtime dependencies for targets in file. * --script-executable: Set the executable used to execute scripts. * --threads: Specify number of worker threads. * --time: Outputs a summary of how long everything took. * --tracelog: Writes a Chrome-compatible trace log to the given file. * -v: Verbose logging. * --version: Prints the GN version number and exits.