Embedding binary data in executables

By Christian Stigen Larsen
Posted 07 Jun 2016 — updated 07 Jun 2016

Applications usually load resources like images from disk. But, in some situations, it may be better to embed binary data right into the executable file. Here are a few ways to do that.

In language like C and C++, the straight-forward way to include binary data in the executable is to convert it to a character array:

const char data[] = {0x00, 0x01, ...};

While there's nothing wrong with that approach, it requires a tool to convert binary data to code (for example, xxd -i). Besides, I find it a bit inelegant, and I'll present some alternatives for you.

Using the GNU linker

This is by far the easiest solution — but, unfortunately, doesn't work on Mac OS X. It does for Linux, though!

Let's say you have an image cat.png and want to embed it into your application. You can create an object file with

$ ld -r -b binary cat.png -o cat.o

The object file will have three symbols in it,

$ nm cat.o

(That's not actual output, but gives you an idea on how to see the symbols. You can also use objdump -x cat.o).

To use them from C, declare some extern variables

extern const char cat_start;
extern const char cat_end;
extern const int cat_size;

and add cat.o to the compiler:

$ gcc cat.o program.c -oprogram

If you have a function display_png_image, you can simply call


Using assembly

If you don't have a working GNU ld — or, if you want utter and complete control — you can use an assembler like nasm.

Just use the incbin directive to include the data. Add some helper symbols for the end of the data, and use a macro to calculate the byte length:

bits 64

section .rodata

global _cat_start
global _cat_end
global _cat_size

_cat_start:   incbin "cat.png"
_cat_size:    dd $-_cat_start

On OS X, compile with

$ nasm -fmacho64 cat.asm -o cat.o

Finally, link your program exactly as before:

$ gcc cat.o program.c -o program

So what's so good about this approach? It lets you put the binary data in the read-only data section, meaning it will be truly read-only. If you used ld instead, you'd have to use objcopy to move the symbols to the .rodata section.

Using objcopy

You can also use objcopy from the GNU binutils package. However, I wasn't able to get it working completely on OS X. Meaning, there isn't an easy, cross-platform way to use it, so I won't write much about it.

But, you can start by doing something like

$ objcopy -I binary -O mach-o-x86_64 \
  --rename-section .data=.const [...] cat.png cat.o

On my system, the linker didn't like the resulting object file. I'll post an update if I get it working. On Linux, it should be straight-forward.

What about other languages?

Most statically compiled languages will let you link in object files. I haven't tried, but I guess you could easily do it in languages like Swift, Rust and so on.

What would you use it for?

Of course, you can embed stuff like images and music into your application. For desktop applications, this may be an advantage in certain situations. But there are other cool uses as well.

Mike Pall, the original author of LuaJIT, gives an example where he compiles Lua programs to object files and wrap them up in an archive file. He then proceeds to link a host program in C with them, so the scripts can be executed without loading anything from disk.

I'm sure there are many other use cases as well.