[TOC]
You just run gn from the command line. There is a script in
depot_tools, which is presumably in your PATH, with this name. The
script will find the binary in the source tree containing the current
directory and run it.
In GYP, the system would generate Debug and Release build
directories for you and configure them accordingly. GN doesn't do this.
Instead, you set up whatever build directory you want with whatever
configuration you want. The Ninja files will be automatically
regenerated if they're out of date when you build in that directory.
To make a build directory:
gn gen out/my_build
Set build arguments on your build directory by running:
gn args out/my_build
This will bring up an editor. Type build args into that file like this:
is_component_build = true
is_debug = false
You can see the list of available arguments and their default values by typing
gn args --list out/my_build
on the command line. Note that you have to specify the build directory for this command because the available arguments can change according to what's set.
Chrome developers can also read the Chrome-specific build configuration instructions for more information.
Run gn args out/Default (substituting your build directory as needed) and
add one or more of the following lines for common cross-compiling options.
target_os = "chromeos"
target_os = "android"
target_cpu = "arm"
target_cpu = "x86"
target_cpu = "x64"
See GNCrossCompiles for more info.
Run gn args out/Default (substituting your build directory as needed).
Add:
use_goma = true
goma_dir = "~/foo/bar/goma"
If your goma is in the default location (~/goma) then you can omit the
goma_dir line.
This is a build arg like the goma flags. run gn args out/Default and add:
is_component_build = true
Create a tools/gn/tutorial/BUILD.gn file and enter the following:
executable("hello_world") {
sources = [
"hello_world.cc",
]
}
There should already be a hello_world.cc file in that directory,
containing what you expect. That's it! Now we just need to tell the
build about this file. Open the BUILD.gn file in the root directory
and add the label of this target to the dependencies of one of the root
groups (a "group" target is a meta-target that is just a collection of
other targets):
group("root") {
deps = [
...
"//url",
"//tools/gn/tutorial:hello_world",
]
}
You can see the label of your target is "//" (indicating the source root), followed by the directory name, a colon, and the target name.
From the command line in the source root directory:
gn gen out/Default
ninja -C out/Default hello_world
out/Default/hello_world
GN encourages target names for static libraries that aren't globally unique. To build one of these, you can pass the label with no leading "//" to ninja:
ninja -C out/Default tools/gn/tutorial:hello_world
Let's make a static library that has a function to say hello to random
people. There is a source file hello.cc in that directory which has a
function to do this. Open the tools/gn/tutorial/BUILD.gn file and add
the static library to the bottom of the existing file:
static_library("hello") {
sources = [
"hello.cc",
]
}
Now let's add an executable that depends on this library:
executable("say_hello") {
sources = [
"say_hello.cc",
]
deps = [
":hello",
]
}
This executable includes one source file and depends on the previous
static library. The static library is referenced by its label in the
deps. You could have used the full label //tools/gn/tutorial:hello
but if you're referencing a target in the same build file, you can use
the shortcut :hello.
From the command line in the source root directory:
ninja -C out/Default say_hello
out/Default/say_hello
Note that you didn't need to re-run GN. GN will automatically rebuild
the ninja files when any build file has changed. You know this happens
when ninja prints [1/1] Regenerating ninja files at the beginning of
execution.
Our hello library has a new feature, the ability to say hello to two
people at once. This feature is controlled by defining TWO_PEOPLE. We
can add defines like so:
static_library("hello") {
sources = [
"hello.cc",
]
defines = [
"TWO_PEOPLE",
]
}
However, users of the library also need to know about this define, and
putting it in the static library target defines it only for the files
there. If somebody else includes hello.h, they won't see the new
definition. To see the new definition, everybody will have to define
TWO_PEOPLE.
GN has a concept called a "config" which encapsulates settings. Let's create one that defines our preprocessor define:
config("hello_config") {
defines = [
"TWO_PEOPLE",
]
}
To apply these settings to your target, you only need to add the config's label to the list of configs in the target:
static_library("hello") {
...
configs += [
":hello_config",
]
}
Note that you need "+=" here instead of "=" since the build
configuration has a default set of configs applied to each target that
set up the default build stuff. You want to add to this list rather than
overwrite it. To see the default configs, you can use the print
function in the build file or the desc command-line subcommand (see
below for examples of both).
This nicely encapsulates our settings, but still requires everybody that
uses our library to set the config on themselves. It would be nice if
everybody that depends on our hello library can get this
automatically. Change your library definition to:
static_library("hello") {
sources = [
"hello.cc",
]
all_dependent_configs = [
":hello_config"
]
}
This applies the hello_config to the hello target itself, plus all
targets that transitively depend on the current one. Now everybody that
depends on us will get our settings. You can also set public_configs
which applies only to targets that directly depend on your target (not
transitively).
Now if you compile and run, you'll see the new version with two people:
> ninja -C out/Default say_hello
ninja: Entering directory 'out/Default'
[1/1] Regenerating ninja files
[4/4] LINK say_hello
> out/Default/say_hello
Hello, Bill and Joy.
You declare which arguments you accept and specify default values via
declare_args.
declare_args() {
enable_teleporter = true
enable_doom_melon = false
}
See gn help buildargs for an overview of how this works.
See gn help declare_args for specifics on declaring them.
It is an error to declare a given argument more than once in a given scope, so care should be used in scoping and naming arguments.
You can run GN in verbose mode to see lots of messages about what it's
doing. Use -v for this.
There is a print command which just writes to stdout:
static_library("hello") {
...
print(configs)
}
This will print all of the configs applying to your target (including the default ones).
You can run gn desc <build_dir> <targetname> to get information about
a given target:
gn desc out/Default //tools/gn/tutorial:say_hello
will print out lots of exciting information. You can also print just one
section. Lets say you wanted to know where your TWO_PEOPLE define
came from on the say_hello target:
> gn desc out/Default //tools/gn/tutorial:say_hello defines --blame
...lots of other stuff omitted...
From //tools/gn/tutorial:hello_config
(Added by //tools/gn/tutorial/BUILD.gn:12)
TWO_PEOPLE
You can see that TWO_PEOPLE was defined by a config, and you can also
see the which line caused that config to be applied to your target (in
this case, the all_dependent_configs line).
Another particularly interesting variation:
gn desc out/Default //base:base_i18n deps --tree
See gn help desc for more.
You can see what took a long time by running it with the --time command line flag. This will output a summary of timings for various things.
You can also make a trace of how the build files were executed:
gn --tracelog=mylog.trace
and you can load the resulting file in Chrome's about:tracing page to
look at everything.