fx is the front-door to a collection of scripts that make many tasks related to Fuchsia development easier. It contains a large number of subcommands, which can be discovered by running
fx help. If you use
zsh as a shell, you can get some auto-completion for
fx by sourcing
scripts/fx-env.sh into your shell.
It is strongly recommended that you
source scripts/fx-env.sh into your shell. This is tested and regularly used with Bash and ZSH. It may work for other compatible shells.
# In your fuchsia checkout: $ cd fuchsia # Add a configuration to your shell to include fx-env.sh $ echo "source \"$PWD/scripts/fx-env.sh\"" >> "$HOME/.$(basename "$SHELL")rc" # If you would like additional convenience tools from the Fuchsia scripts, also # optionally run the following: $ echo "fx-update-path" >> "$HOME/.$(basename "$SHELL")rc" # Restart your shell $ exec "$SHELL"
The above method provides the most well defined feature set, and should be generally non-invasive. If it causes bugs in your shell environment, please file project bugs.
If for some reason you need to work with multiple Fuchsia checkouts (recommended workflows below should obviate such a need), then you may wish to do something other than the above. In this case, there are a few well supported methods:
$PATHwhile working in a particular Fuchsia directory.
It is NOT recommended (though presently works) to copy
fx to other places, such as
~/bin/fx (as this could one day break), or to add
$FUCHSIA_DIR/scripts to your
$PATH (as reviewers of code in
//scripts) do not block the addition of files in that directory which could lead to unpredictable behaviors (for example, that directory contains binaries with generic names like “bootstrap” which may unintentionally override the behavior of other systems).
The first thing you will want to do after checking out a Fuchsia tree is to build Fuchsia, and then get it onto a device.
fx has some commands to help with this:
fx setconfigure a build
fx buildexecute a build
fx flash ; fx mkzedbootflash a target; or prepare a zedboot USB key
fx serveserve a build
fx updateupdate a target
fx run-testexecute tests
fx shellconnect to a target shell
First let's configure the build. To do this we need to make a few choices:
tools, and if you're working on features, you probably want
Armed with our above choices (if you didn't read above, do so now), you are ready to configure your build:
$ fx set workstation.x64 --with //bundles:tests
This command stores the configuration in an
args.gn file in the build directory (which is
out/default by default). You can edit this file using the
fx args command to create more elaborate configurations.
fx list-productsfor a list of other product configurations)
x64(on arm64 boards, the board choice is very important! Run
fx list-boardsfor a list of board configurations)
So what are
universe? (new names)
Configurations ultimately specify dependencies (mostly packages) that contribute to output artifacts of the build (mostly images and package repositories). The build is parameterized to determine which dependencies (mostly packages) are added to which output artifacts (images or package repositories). The three axes are called “base”, “cache”, and “universe”:
The “board” and “product” configurations pick a predefined set of members for each of these package sets. Most commonly the board configurations specify a set of boot-critical drivers to add to the base dependency set, and could for example include some optional but common peripheral drivers in the cache set. The board configuration may also include some board-specific development tools (more commonly host tools, rather than target packages) for interacting with the board in “universe”. The product configurations make choices to add more or less software to the base, cace or universe package sets based on the definition and feature set of the product they represent. A speaker product, for example, adds many audio-media-related packages to the base. A workstation product adds a wide range of GUI, media and many other packages to the base.
There are many more than below, but the following three particularly important configurations to be familiar with:
bringupis a minimal feature set product that is focused on being very simple and very lean. It exists to provide fast builds and small images (primarily used in a netboot rather than paved fashion), and is great for working on very low-level facilities, such as the Zircon kernel or board-specific drivers and configurations. It lacks most network capabilities, and therefore is not able to add new software at runtime or upgrade itself.
coreis a minimal feature set that can install additional software (such as items added to the “universe” dependency set). It is the starting point for all higher-level product configurations. It has common network capabilities and can update a system over-the-air.
workstationis a basis for a general purpose development environment, good for working on UI, media and many other high-level features. This is also the best environment for enthusiasts to play with and explore.
As with products, there are many more, but the following bundles are most important to be familiar with:
toolscontains a broad array of the most common developer tools. This includes tools for spawning components from command-line shells, tools for reconfiguring and testing networks, making http requests, debugging programs, changing audio volume, and so on. The core product includes
bundles:toolsin the universe package set by default.
testscauses all test programs to be built. Most test programs can be invoked using
run-test-componenton the device, or via
kitchen_sinkis a target that causes all other build targets to be included. It is useful when testing the impact of core changes, or when making large scale changes in the code base. It also may be a fun configuration for enthusiasts to play with, as it includes all software available in the source tree. Note that kitchen sink will produce more than 20GB of build artifacts and requires at least 2GB of storage on the target device (size estimates from Q1/2019).
For most use cases, only
fx build is needed. The build is optimized for fast incremental rebuilds, as such, repeating this command does the minimal work required after code has been changed, and no work if the source files are unchanged.
fx build, a few other build related commands provide more granular control:
fx build-zirconbuilds only the Zircon portion of the build.
fx buildbuilds only the Fuchsia portion (all above Zircon, but not Zircon itself) of the build.
fx cleanclear out all build artifacts.
fx clean-buildperform a clean, then a build.
fx genrepeat the
gn genprocess that
fx setperformed. Users making fine grained build argument changes (e.g. by editing
args.gndirectly) can run
fx gento reconfigure their build.
fx build can be given the name of a specific target or file to build. For example, a target with the label
//foo/bar:blah can be built with
fx build foo/bar:blah.
Note that this only works for targets declared in the default GN toolchain. For targets in other toolchains, the path of an output file may be used instead. For example, an executable target with the label
//foo/bar:blah(//build/toolchain:host_x64) can be built with
fx build <output_dir>/host_x64/blah.
The exact preparation required to put Fuchsia onto a target device varies by specific device, but there are two general groups in common use today, made convenient behind
fx flashis used with most
arm64devices to perform a raw write of Zedboot to the device, preparing it for Paving.
fx mkzedbootis used with most
x64devices to prepare a bootable USB key that boots into Zedboot, preparing the device for Paving.
Zedboot is a special configuration of Zircon that contains a simple network stack, a simple device advertisement and discovery protocols, and a suite of protocols to write Fuchsia to a target disk and/or to network boot a target system. Zedboot is a term used for both the overall process, as well as a special build configuration. Many people come to know it as “the blue screen with ASCII art”.
To enter Zedboot on an arm64 target, power on the device while triggering a boot into fastboot flashing mode (often this involves holding a particular button while rebooting or powering on that varies by particular hardware target). Once in flashing mode, execute
fx flash on the host system.
To enter Zedboot on an x64 target, first produce a Zedboot USB key using
fx mkzedboot <path-to-usb-device> (to list suitable USB devices on your system, execute
fx list-usb-disks). Remove the USB key after completion, insert it to the target device, and reboot the target device, selecting “Boot from USB” from the boot options, or in the device BIOS. There are additional instructions for preparing a Pixelbook.
Paving is in many ways similar to “flashing” from other worlds, however, it has some differences. Specifically, paving refers to a group of processes and protocols in Fuchsia to transfer a set of artifacts to a target system that will be written into various partitions on a target system. By contrast, the process of “flashing” is more of a raw process of writing a raw data stream to a raw disk device, and not strictly partition-oriented.
Users can start a paving process by first flashing Zedboot using
fx flash, or by booting a Zedboot USB key made by
fx mkzedboot, then executing
fx pave on the host system. In general most users actually will want to use
fx serve instead of
fx serve is covered in the next section.
In Fuchsia, “netboot” refers to sending a set of artifacts to a Zedboot instance that instead of making changes to the disk, will just be booted from RAM. Users can perform a “netboot” by first booting a device into Zedboot by using either
fx flash (arm64) or
fx mkzedboot (x64), and then executing
fx netboot on the host system.
netboot artifacts are not produced by all builds by default, because for larger builds such as the “workstation” product configuration such builds are extremely large, and producing them many times a day is both slow as well as measurably wearing on host disk hardware. The
bringup configuration always prepares
netboot artifacts. For all other build configurations, a user can optionally build the netboot artifacts using
fx build netboot.
A lot of build configurations for Fuchsia include software that is not immediately included in the base images that a build produces, that are written to devices during paving. Such software is instead made available to target devices on-demand, which is often colloquially referred to as “ephemeral software”.
fx serve performs two functions internally:
fx pavestart a paving server, used for “fresh installs” of a Fuchsia device from a Zedboot state.
fx serve-updatesstart a package repository server, used for dynamic installation of software at runtime, as well as whole-system updates.
fx serve-updates command also searches for a device to configure, and upon discovery (which may be restricted/modulated with
fx set-device or
fx -d) the target device is configured to use the repository server as a source of dynamic packages and system updates.
As described in prior sections, there are different groups of software on a Fuchsia device:
For new user development workflows, the most general command to assist with updating a target device is
fx update. The
fx update command first updates all “cache” software, and then performs an
fx ota or, a core system update. This command reboots the target device when it is complete. The end result of this process should be indistinguishable in terms of software versions from performing a fresh pave of a device.
fx update process causes a device reboot, it is sometimes not the most efficient process for diagnosis, debugging or other non-testing based workflows or needs. In these cases a user has some options for how to ensure that software on a device is being regularly updated.
fx serve process configures a Fuchsia software repository with automatic update features. The repository informs the target device of newly updated software every time the underlying repository is updated (which happens at the end of every successful
fx build). For many software components, the easiest way to update them during development is to ensure that they are not included in the base set, but instead included in either “cache” or “universe”. In that case, simply restarting the software on the target (e.g. by closing it completely, or by invoking
killall) will result in the software being immediately updated when it is started again.
fx push-package and
fx build-push perform manual, forceful updates of packages on a target device. These commands do not however know how to re-start software on the device, as such they provide little benefit over simply restarting software correctly which, along with
fx serve, will cause software to be updated as a natural course of restarting. These commands are sometimes used to diagnose issues, or in cases where automatic updates are disabled in special build configurations.
Note: some software may not appear to be updating because it is being run inside of a “runner” process or some other surrounding environment that is “holding on” to resources for the previous package version, only spawning programs from the old package. As packages in Fuchsia are immutable and content-addressed, when host environments retain resources in this manner, there is nothing that the update system can do to forcefully trigger updates in the rest of the system. Users who find themselves with this issue mostly need to find efficient workflow methods to fully restart the relevant software stack.
XXX: TODO: NEEDED: simple example command to restart all GUI / shell components, e.g. if basemgr spawned things under a job or a component topology, a method to restart that topology (e.g.
killall basemgr.cmx, however that's no good today).
The Fuchsia codebase contains many tests. Most of these tests are themselves Components, and can be launched on the target device in the same way as other components. On the target device, some programs also assist with test specific concerns for component launching, such as
/bin/run-test-component. The process can also conveniently be controlled from the development host by way of
fx run-test <package-name> requires the user to specify a particular package name to execute. A package may contain one or more tests. Arguments after
fx run-test package-name are passed to the program
runtests. One particularly common use case is to execute:
fx run-test <package-of-many-tests> -t <name-of-one-test> to execute only a single test. To list the packages that are members of the current build configuration, run
Some users find that an effective high focus workflow is to have the system build, push and execute tests whenever they save their source code. This can be achived with
fx very easily, for example:
$ fx -i run-test rolldice-tests
The above command will execute the rolldice tests every time a change is made to the source code in the tree. The
-i flag to
fx to repeat the rest of it's command every time the source code in the tree is changed. As the
fx run-test command first performs a build, then executes a test on a target, this combination provides a convenient auto-test loop, great for high focus workflows like test driven development.
Most product configurations include an SSH server with a Fuchsia specific configuration. The command
fx shell is a convenient wrapper to connect to the target device over SSH and provides access to a very simply POSIX-style shell. Users should note that while the shell is a fork of a POSIX shell, it does not provide all features of a common Unix shell. In particular users will find that CTRL+C has odd quirks, and may often find quirks for sub-shell expressions and certain more advanced IO redirections or environment variable propagations. These misfeatures are side effects of Fuchsia not being a POSIX system.
Nonetheless the shell made available via
fx shell is extremely useful for imperatively executing programs on the Fuchsia target, as well as exploring some of the diagnostic / debug interfaces made available in a filesystem tree, such as
/dev. It is also useful for invoking programs such as
/bin/run that provides facilities for launching Fuchsia components. If the
tools bundle is available in the build configuration, many tools common to unix shell environments have been ported and are available, such as
fortune and so on.
fx syslog captures all logs from low-level and high-level programs, including the kernel, drivers and other userspace programs.
fx syslog depends upon a working high level network stack and SSH. As such,
fx syslog does not work with Zedboot or “bringup” product configurations. If a device is in a state where
fx syslog ceases to function, it is often useful to switch to
fx log to capture more information about probable causes.
fx log captures only a low-level log stream called “klog”. The klog stream includes logs from the Zircon kernel itself, as well as a subset of userspace software (most notably drivers and low-level core software).
fx log depends on a lightweight network stack called
netsvc that has a tendency to remain available even after problems in higher-level software. The netsvc suite is also always available in “bringup” product configurations, as such,
fx log is most useful when working on low-level software, such as the Zircon kernel, or drivers.
fx cp provides a basic wrapper around
scp, similar to how
fx shell is a wrapper around
# copy ./book.txt from the host, to /tmp/book.txt on the target $ fx cp book.txt /tmp/book.txt # copy /tmp/poem.txt on the target to poem.txt on the host $ fx cp --to-host /tmp/poem.txt poem.txt
fx run starts a Fuchsia build under QEMU, a general purpose virtual machine. Users on Linux host systems with KVM can execute
fx run -k for significantly improved performance.
The QEMU environment does not support any GUI programs. It also has limiting network capabilities. Hardware devices are recommended for day-to-day Fuchsia software development whenever possible.
Some users will have more than one Fuchsia device on a network, and will want to limit the effects of various commands to particular of those devices. The
fx set-device command exists to help with this use case.
fx set-device command binds a particular device node name to a particular build directory. This is particularly useful when a user wishes to keep several different devices in several build configurations, and could be setup as follows:
$ fx --dir out/workstation set workstation.x64 $ fx build $ fx set-device <workstation-node-name> $ fx --dir out/core set core.arm64 $ fx build $ fx set-device <core-node-name> # Start a server for the workstation: $ fx --dir=out/workstation serve # Set the default build-dir and target device to the arm64 core, and # connect to a shell on that device: $ fx use out/core $ fx shell
Additionally, for users who wish to execute a command against a single Fuchsia device from the current default build directory, as a one-off command, the
fx global flag
-d allows overriding the target node name for a single command invocation.
On some devices (most arm64 devices at present) there are also some useful flags:
fx reboot -rreboot into “recovery” (Zedboot)
fx reboot -breboot into “bootloader” (Flash)
fx whereiscl <query>
This command tells whether the given CL is merged, and if so whether it passed Global Integration. The query can be either a Gerrit review URL, a CL number, a Change-Id, or a git revisioin.
$ fx whereiscl fxr/286748 CL status: MERGED GI status: PASSED $ fx whereiscl https://fuchsia-review.googlesource.com/c/fuchsia/+/287311/1/garnet/go/src/amber/source/source.go CL status: NEW $ fx whereiscl I94c56fa4e59842d398bfa90a48c45b388f095184 CL status: MERGED GI status: PASSED $ fx whereiscl 6575aee CL status: MERGED GI status: PENDING
-xflag turns on tracing for the
fxscripts, printing out all expressions evaluated during the
fx execexecutes an abitrary program that follows inside of the current
fxenvironment. As an example
fx exec envprints all environment variables in that environment (
fx exec env | grep FUCHSIAis likely of interest).
fx help <command> provides the best introductory documentation for that command. Some commands also support/provide
fx <command> -h or
fx <command> --help, however this help is not available for all commands. This is unusual, but is a function of implementation details. Internally many
fx commands just run other programs, most often those produced by the build, and flags are in many cases passed on unaltered to those programs. In those cases, passing the usual
--help flags may not provide documentation for
fx <command>, but instead for the program invoked downstream of
Users should always start with
fx help <command>.
fx help with no other arguments provides a list of all available commands in
fx, as well as documentation for
fx global flags.