shipping executables
Shell C++
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.

How to publish native executables for (almost) all platforms

blogged article

NOTE: If you find some bugs, please inform me :)

Abstract (in 10 seconds)

By investigating Ninja & Chromium depot_tools approach, I'll introduce the way to mobile your C++ executable for (almost) all platforms.


You sometimes would like to publish native executables. One day, you make a nice command written in C++, and using it on Linux x64. You may write a nice zsh script using this nice executable and it works as you want.

But you may control dotfiles on git repository and sometimes you need to sync your nice environment between your working machines. In the meantime, you may build your nice command on each environment. However, whenever you fix / enhance your native module, you need to rebuild it on all your machines. Some machines are the same architecture as your main machine. But some machines may be OSX. Another machines are Windows, etc...

Why don't you write it with Script languages?

You need to setup script languages environment on the each machines. And I sometimes would like to construct commands based on C++ libraries, such as LLVM.

NOTE: LLVM can be built as static library. So it is possible to bundle LLVM libraries to one executable file.

Why don't you write it with golang?

As the same as the above.


When adding a directory to PATH, cmd.exe will search a command through PATH directories. cmd.exe will find {name}.exe executable file. ".exe" extension! It is not (usually) considered in Linux & OSX. So placing {name}.exe and {name} command in the same directory, you can provide Windows and Unix binary at the same time.

So this stroy is very easy. The way is buidling command as name.exe and placing it on the appropriate directory. That's all.

OSX & Linux

You need to distinguish them. So the easiest way to start it is making {name} command sh script to execute an appropriate binary.

Switching binary with shell script

In the script, you can use uname command to distinguish the platform. This command works on FreeBSD, OSX and Linux. This is the definition of (almost) all platforms.

According to the uname man page, uname -s will answer the kernel name. For example,


And you need to consider is that machine architecture (such as x64). You can get this by uname too. uname -m will return hardware architecture. For example,

    amd64  # On FreeBSD
    x86_64 # On Linux

You can see uname implementation of FreeBSD here. And actual code to gain a architecture name is here. And architecture candidates on FreeBSD seems the same to this.

amd64 arm i386 ia64 mips pc98 powerpc sparc64 x86

And on the Linux case. According to the Linux's Makefile (here), we can get subarchitecture by this script.

uname -m | sed -e s/i.86/x86/ -e s/x86_64/x86/ \
               -e s/sun4u/sparc64/ \
               -e s/arm.*/arm/ -e s/sa110/arm/ \
               -e s/s390x/s390/ -e s/parisc64/parisc/ \
               -e s/ppc.*/powerpc/ -e s/mips.*/mips/ \
               -e s/sh[234].*/sh/ -e s/aarch64.*/arm64/

So by adding amd64 case to it, we can get the command line to get sub architecture.

uname -m | sed -e s/i.86/x86/ -e s/x86_64/x86/ -e s/amd64/x86/ \
               -e s/sun4u/sparc64/ \
               -e s/arm.*/arm/ -e s/sa110/arm/ \
               -e s/s390x/s390/ -e s/parisc64/parisc/ \
               -e s/ppc.*/powerpc/ -e s/mips.*/mips/ \
               -e s/sh[234].*/sh/ -e s/aarch64.*/arm64/

And finally, you need to get if CPU is 64bit / 32bit. To get it on Linux, FreeBSD and Darwin, you can use getconf LONG_BIT according to the depot_tools source.

Then here's the complete switch script.

OS="$(uname -s)"
ARCH="$(uname -m | sed -e s/i.86/x86/ -e s/x86_64/x86/ -e s/amd64/x86/ \
                       -e s/sun4u/sparc64/ \
                       -e s/arm.*/arm/ -e s/sa110/arm/ \
                       -e s/s390x/s390/ -e s/parisc64/parisc/ \
                       -e s/ppc.*/powerpc/ -e s/mips.*/mips/ \
                       -e s/sh[234].*/sh/ -e s/aarch64.*/arm64/)"
BIT="$(getconf LONG_BIT)"

if [ -e "$COMMAND" ]; then
    exec "$COMMAND"
    echo "Command ${COMMAND} not found."
    echo "Unsupported architecture ${OS} ${ARCH} ${BIT}"

So placing {name}-Linux-x86-64 in the same directory to this script, you can make command available on Linux x64 environment.

Building executables with -static option (Linux and FreeBSD)

Native executables (built by g++, clang or so) usually depends on libc, libgcc, libcompiler-rt, libsup++, libc++ and so on. But as you may know, you can build native executables not depending on them. You can use -static option.

g++ -static

Then g++ produces a executable not depending on the above libraries. You can ensure it by executing ldd {name}. So moving it to the other platform, it works correctly if OS, ARCH, BIT are the same.

NOTE: Without -static option, ldd a.out result may be the following.

$ ldd a.out =>  (0x00007fff799fe000) => /usr/lib/x86_64-linux-gnu/ (0x00007f67bd054000) => /lib/x86_64-linux-gnu/ (0x00007f67bcc8c000) => /lib/x86_64-linux-gnu/ (0x00007f67bc987000)
        /lib64/ (0x00007f67bd383000) => /lib/x86_64-linux-gnu/ (0x00007f67bc771000)

Since is contained in the linux kernel (Linux Virtual Dynamic Shared Objects), it isn't required as a shared library. And since,,, and /lib64/ are placed with the same name in a lot of Linux distributions. So if you don't worry about the edge cases, -static for system libraries is not necessary. Compiled binaries basically work on the other Linux systems. For example, ldd ninja-linux64 shows the same result. But of cource, the other libraries should be linked statically. This shows how to control linking with libraries. By following the way described in that article, we can control which libraries are linked statically. It's possible to link system libraries as dynamic and the others as static.

Building executables on OSX

However, -static option doesn't work on OSX. According to the man gcc.

This option will not work on Mac OS X unless all libraries (including libgcc.a) have also been compiled with -static.
Since neither a static version of libSystem.dylib nor crt0.o are provided, this option is not useful to most people.

So if you build with -static (e.g. clang++ -static), you may get the following error.

ld: library not found for -lcrt0.o
clang: error: linker command failed with exit code 1 (use -v to see invocation)

This is also described on this article.

But don't worry about that. Since OSX environment layout is limited, basically dynamically linked system library is placed on appropriate sites. You can see dynamic linked libraries on the binary by otool -L executable. For example,

    /usr/lib/libc++.1.dylib (compatibility version 1.0.0, current version 120.0.0)
    /usr/lib/libSystem.B.dylib (compatibility version 1.0.0, current version 1197.1.1)

So in this time, we just build executables as name-Darwin-x86-64 and place it.

Cross compiling the 32 / 64 bit

Just googled and found the article and this. Especially, the latter article works correctly on my environment (Ubuntu 13.10, FreeBSD 10)


So I've created the sample project portable to test it. It includes the above script and executables. (Windows 32bit, OSX 64bit, Linux 32bit, Linux 64bit. Their architecture is x86)

So just copying (or git clone) it and append this directory to the PATH, you can sync, execute, publish this command on the (almost) all platforms.