At first I haven't manage yet to do what I want. In this thread, there is a solution: http://mail.haskell.org/pipermail/haskell-cafe/2015-April/119021.html for the dynamic case. See below for the complete rootfs I use (similar thus to https://github.com/snoyberg/haskell-scratch).
We want a minimal Docker image with the following program.
> cat Example.hs
main = putStrLn "Hello, world."
Using Halcyon (in a Docker image, but that is not important), we can produce a dynamically-linked binary:
> docker run -v $(pwd):/source noteed/ghc-7-8-4 sh -c \
"/app/ghc/bin/ghc -O2 -L/app/ghc/usr/lib --make /source/Example.hs"
> $ ldd Example
linux-vdso.so.1 => (0x00007fff5558d000)
libgmp.so.10 => /usr/lib/x86_64-linux-gnu/libgmp.so.10 (0x00007f39563fb000)
libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007f39560f5000)
librt.so.1 => /lib/x86_64-linux-gnu/librt.so.1 (0x00007f3955eec000)
libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007f3955ce8000)
libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007f3955ad2000)
libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f395570c000)
libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007f39554ee000)
/lib64/ld-linux-x86-64.so.2 (0x00007f3956690000)
We can also produce a statically-linked binary:
> docker run -v $(pwd):/source noteed/ghc-7-8-4 sh -c \
"apt-get install -q -y libgmp-dev && \
/app/ghc/bin/ghc -O2 --make \
-static -optc-static -optl-static -optl-pthread \
/source/Example.hs"
> ldd Example
not a dynamic executable
In the static case, the rootfs looks simply like this:
> tree rootfs
rootfs
└── Example
0 directories, 1 file
I also tried the approach outlined here for the dynamic case.
> tree rootfs
rootfs
├── Example
├── lib
│ └── x86_64-linux-gnu
│ ├── libc.so.6
│ ├── libdl.so.2
│ ├── libgcc_s.so.1
│ ├── libm.so.6
│ ├── libpthread.so.0
│ └── librt.so.1
├── lib64
│ └── ld-linux-x86-64.so.2
└── usr
└── lib
└── x86_64-linux-gnu
└── libgmp.so.10
6 directories, 10 files
In both cases, building the image is as follow:
> tar -C rootfs -c . | docker import - example
Unfortunately, running
> docker run example /Example
results in an Example process using 100% CPU and doing nothing.
I have tried the static case with the true-asm binary, and it works.
Here is the complete rootfs (for the dynamically compiled Example):
rootfs
├── Example
├── lib
│ └── x86_64-linux-gnu
│ ├── libc.so.6
│ ├── libdl.so.2
│ ├── libgcc_s.so.1
│ ├── libm.so.6
│ ├── libpthread.so.0
│ ├── librt.so.1
│ └── libz.so.1
├── lib64
│ └── ld-linux-x86-64.so.2
└── usr
└── lib
└── x86_64-linux-gnu
├── gconv
│ ├── gconv-modules
│ ├── gconv-modules.cache
│ ├── UTF-16.so
│ ├── UTF-32.so
│ └── UTF-7.so
└── libgmp.so.10
7 directories, 15 files
And thus:
> tar -C rootfs -c . | docker import - example
cdec64252aa2f9740e45e93890c78c48c1bb4b8765bd77b791755de58574b6c3
> docker run example /Example
Hello, world.
This is quite awesome !
> docker save example > example.tar
> xz example.tar
> du -h example.tar.xz
1.5M example.tar.xz
(strip was applied to the binary.)
This works with chroot too:
> sudo chroot rootfs /Example
You can follow instruction from
https://www.kernel.org/doc/Documentation/filesystems/ramfs-rootfs-initramfs.txt using our Example as ini t executable (add threadDelay to ensure that our init doesn't exit):
> mv rootfs/Example rootfs/init
> cd rootfs ; find . | cpio -o -H newc | gzip > ../rootfs.cpio.gz ; cd ..
> qemu-system-x86_64 \
-nographic \
-kernel /boot/vmlinuz-3.19.0-7-generic \
-initrd rootfs.cpio.gz \
-append "console=ttyS0" \
/dev/zero
-nographic and console=ttyS0 are needed only to get qemu output in the
current terminal.
Use C-a h for help and C-a x to exit qemu.
@noteed What about using something like https://github.com/nilcons/ghc-musl to compile the statically-linked binary?