Although Go strives to be a cross platform language, cross compilation from one platform to another is not as simple as it could be, as you need the Go sources bootstrapped to each platform and architecture.

The first step towards cross compiling was Dave Cheney's golang-crosscompile package, which automatically bootstrapped the necessary sources based on your existing Go installation. Although this was enough for a lot of cases, certain drawbacks became apparent where the official libraries used CGO internally: any dependency to third party platform code is unavailable, hence those parts don't cross compile nicely (native DNS resolution, system certificate access, etc).

A step forward in enabling cross compilation was Alan Shreve's gonative package, which instead of bootstrapping the different platforms based on the existing Go installation, downloaded the official pre-compiled binaries from the golang website and injected those into the local toolchain. Since the pre-built binaries already contained the necessary platform specific code, the few missing dependencies were resolved, and true cross compilation could commence... of pure Go code.

However, there was still one feature missing: cross compiling Go code that used CGO itself, which isn't trivial since you need access to OS specific headers and libraries. This becomes very annoying when you need access only to some trivial OS specific functionality (e.g. query the CPU load), but need to configure and maintain separate build environments to do it.

My solution to the challenge of cross compiling Go code with embedded C/C++ snippets (i.e. CGO_ENABLED=1) is based on the concept of lightweight Linux containers. All the necessary Go tool-chains, C cross compilers and platform headers/libraries have been assembled into a single Docker container, which can then be called as if a single command to compile a Go package to various platforms and architectures.

Thanks to the following projects for supporting XGO

• Gitea - A painless self-hosted Git service.
• Offen - The fair and lightweight alternative to common web analytics tools.
• Vikunja - The to-do app to organize your life.
• Woodpecker CI - Woodpecker is a simple CI engine with great extensibility.

You too can sponsor this project to ensure its continued maintenance.

Although you could build the container manually, it is available as an automatic trusted build from Docker's container registry (not insignificant in size):

docker pull techknowlogick/xgo:latest

To prevent having to remember a potentially complex Docker command every time, a lightweight Go wrapper was written on top of it.

go get src.techknowlogick.com/xgo

For go >= 1.17, go get is deprecated, so you'll have to use this command:

go install src.techknowlogick.com/xgo@latest

Simply specify the import path you want to build, and xgo will do the rest:

$ xgo github.com/project-iris/iris
...

$ ls -al
-rwxr-xr-x  1 root  root    6776500 Nov 24 16:44 iris-darwin-10.6-386
-rwxr-xr-x  1 root  root    8755532 Nov 24 16:44 iris-darwin-10.6-amd64
-rwxr-xr-x  1 root  root   10135248 Nov 24 16:44 iris-linux-386
-rwxr-xr-x  1 root  root   12598472 Nov 24 16:44 iris-linux-amd64
-rwxr-xr-x  1 root  root   10040464 Nov 24 16:44 iris-linux-arm
-rwxr-xr-x  1 root  root    7516368 Nov 24 16:44 iris-windows-4.0-386.exe
-rwxr-xr-x  1 root  root    9549416 Nov 24 16:44 iris-windows-4.0-amd64.exe

If the path is not a canonical import path, but rather a local path (starts with a dot . or a dash /), xgo will use the local GOPATH contents for the cross compilation.

A handful of flags can be passed to go build. The currently supported ones are

• -v: prints the names of packages as they are compiled
• -x: prints the build commands as compilation progresses
• -race: enables data race detection (supported only on amd64, rest built without)
• -tags='tag list': list of build tags to consider satisfied during the build
• -ldflags='flag list': arguments to pass on each go tool link invocation
• -gcflags='flag list': arguments to pass on each go tool compile invocation
• -buildmode=mode: binary type to produce by the compiler

As newer versions of the language runtime, libraries and tools get released, these will get incorporated into xgo too as extensions layers to the base cross compilation image (only Go 1.13 and above will be supported).

You can select which Go release to work with through the -go command line flag to xgo and if the specific release was already integrated, it will automatically be retrieved and installed.

$ xgo -go go-1.13.2 github.com/project-iris/iris

Additionally, a few wildcard release strings are also supported:

• latest will use the latest Go release (this is the default)
• go-1.13.x will use the latest point release of a specific Go version

xgo by default uses the name of the package being cross compiled as the output file prefix. This can be overridden with the -out flag.

$ xgo -out iris-v0.3.2 github.com/project-iris/iris
...

$ ls -al
-rwxr-xr-x  1 root  root   6776500 Nov 24 16:44 iris-v0.3.2-darwin-10.6-386
-rwxr-xr-x  1 root  root   8755532 Nov 24 16:44 iris-v0.3.2-darwin-10.6-amd64
-rwxr-xr-x  1 root  root  10135248 Nov 24 16:44 iris-v0.3.2-linux-386
-rwxr-xr-x  1 root  root  12598472 Nov 24 16:44 iris-v0.3.2-linux-amd64
-rwxr-xr-x  1 root  root  10040464 Nov 24 16:44 iris-v0.3.2-linux-arm
-rwxr-xr-x  1 root  root   7516368 Nov 24 16:44 iris-v0.3.2-windows-4.0-386.exe
-rwxr-xr-x  1 root  root   9549416 Nov 24 16:44 iris-v0.3.2-windows-4.0-amd64.exe

Similarly to go get, xgo also uses the master branch of a repository during source code retrieval. To switch to a different branch before compilation pass the desired branch name through the --branch argument.

$ xgo --branch release-branch.go1.4 golang.org/x/tools/cmd/goimports
...

$ ls -al
-rwxr-xr-x  1 root  root   4139868 Nov 24 16:40 goimports-darwin-10.6-386
-rwxr-xr-x  1 root  root   5186720 Nov 24 16:40 goimports-darwin-10.6-amd64
-rwxr-xr-x  1 root  root   4189456 Nov 24 16:40 goimports-linux-386
-rwxr-xr-x  1 root  root   5264136 Nov 24 16:40 goimports-linux-amd64
-rwxr-xr-x  1 root  root   4209416 Nov 24 16:40 goimports-linux-arm
-rwxr-xr-x  1 root  root   4348416 Nov 24 16:40 goimports-windows-4.0-386.exe
-rwxr-xr-x  1 root  root   5415424 Nov 24 16:40 goimports-windows-4.0-amd64.exe

Yet again similarly to go get, xgo uses the repository remote corresponding to the import path being built. To switch to a different remote while preserving the original import path, use the --remote argument.

$ xgo --remote github.com/golang/tools golang.org/x/tools/cmd/goimports
...

If you used the above branch or remote selection machanisms, it may happen that the path you are trying to build is only present in the specific branch and not the default repository, causing Go to fail at locating it. To circumvent this, you may specify only the repository root for xgo, and use an additional --pkg parameter to select the exact package within, honoring any prior branch and remote selections.

$ xgo --pkg cmd/goimports golang.org/x/tools
...

$ ls -al
-rwxr-xr-x  1 root  root   4164448 Nov 24 16:38 goimports-darwin-10.6-386
-rwxr-xr-x  1 root  root   5223584 Nov 24 16:38 goimports-darwin-10.6-amd64
-rwxr-xr-x  1 root  root   4217184 Nov 24 16:38 goimports-linux-386
-rwxr-xr-x  1 root  root   5295768 Nov 24 16:38 goimports-linux-amd64
-rwxr-xr-x  1 root  root   4233120 Nov 24 16:38 goimports-linux-arm
-rwxr-xr-x  1 root  root   4373504 Nov 24 16:38 goimports-windows-4.0-386.exe
-rwxr-xr-x  1 root  root   5450240 Nov 24 16:38 goimports-windows-4.0-amd64.exe

This argument may at some point be integrated into the import path itself, but for now it exists as an independent build parameter. Also, there is not possibility for now to build mulitple commands in one go.

By default xgo will try and build the specified package to all platforms and architectures supported by the underlying Go runtime. If you wish to restrict the build to only a few target systems, use the comma separated --targets CLI argument:

• --targets=linux/arm: builds only the ARMv5 Linux binaries (arm-6/arm-7 allowed)
• --targets=windows/*,darwin/*: builds all Windows and OSX binaries
• --targets=*/arm: builds ARM binaries for all platforms
• --targets=*/*: builds all suppoted targets (default)

The supported targets are:

• Platforms: darwin, linux, windows
• Achitectures: 386, amd64, arm-5, arm-6, arm-7, arm64, mips, mipsle, mips64, mips64le, riscv64

By default xgo tries to cross compile to the lowest possible versions of every supported platform, in order to produce binaries that are portable among various versions of the same operating system. This however can lead to issues if a used dependency is only supported by more recent systems. As such, xgo supports the selection of specific platform versions by appending them to the OS target string.

• --targets=darwin-10.9/*: cross compile to Mac OS X Mavericks
• --targets=windows-6.0/*: cross compile to Windows Vista

The supported platforms are:

• All Windows APIs up to Windows 8.1 limited by mingw-w64 (API level ids)
• OSX APIs in the range of 10.6 - 10.14

The main differentiator of xgo versus other cross compilers is support for basic embedded C/C++ code and target-platform specific OS SDK availability. The current xgo release introduces an experimental CGO dependency cross compilation, enabling building Go programs that require external C/C++ libraries.

It is assumed that the dependent C/C++ library is configure/make based, was properly prepared for cross compilation and is available as a tarball download (.tar, .tar.gz or .tar.bz2). Further plans include extending this to cmake based projects, if need arises (please open an issue if it's important to you).

Such dependencies can be added via the --deps argument. They will be retrieved prior to starting the cross compilation and the packages cached to save bandwidth on subsequent calls.

A complex sample for such a scenario is building the Ethereum CLI node, which has the GNU Multiple Precision Arithmetic Library as it's dependency.

$ xgo --deps=https://gmplib.org/download/gmp/gmp-6.1.0.tar.bz2  \
    --targets=windows/* github.com/ethereum/go-ethereum/cmd/geth
...

$ ls -al
-rwxr-xr-x 1 root root 16315679 Nov 24 16:39 geth-windows-4.0-386.exe
-rwxr-xr-x 1 root root 19452036 Nov 24 16:38 geth-windows-4.0-amd64.exe

Some trivial arguments may be passed to the dependencies' configure script via --depsargs.

$ xgo --deps=https://gmplib.org/download/gmp/gmp-6.1.0.tar.bz2  \
    --targets=ios/* --depsargs=--disable-assembly               \
    github.com/ethereum/go-ethereum/cmd/geth
...

$ ls -al
-rwxr-xr-x 1 root root 14804160 Nov 24 16:32 geth-ios-5.0-arm

Note, that since xgo needs to cross compile the dependencies for each platform and architecture separately, build time can increase significantly.

To give the user more power, xgo provides two hook scripts which will be sourced during the build if they are available in a provided hooks directory (provided by the --hooksdir argument, which mounts that directory into the container). The first one is setup.sh which will be sourced after everything is set up. This script can e.g. be used to install additional packages in the container. The second is build.sh which will be sourced for each target (before the actual build process).

All environment variables set up by xgo are available in the scripts.

Within build.sh there are several target specific environment variables:

• XGOOS and XGOARCH are expanded to the actual value defined by the build targets.
• CC: C cross compiler to use for the build
• HOST: Target platform to build
• PREFIX: File-system path where to install the built binaries.

For further reference, see build.sh