- Notifications
You must be signed in to change notification settings - Fork3
build-once run-anywhere OCaml programs
License
dinosaure/esperanto
Folders and files
Name | Name | Last commit message | Last commit date | |
---|---|---|---|---|
Repository files navigation
Esperanto makes OCaml a build-once run-anywhere language, like itself, exceptit doesn't need an interpreter or virtual machine. Indeed, OCaml is able toproduce anative executable which requires few libraries:
- a standard C library
libasmrun.a
(thecaml-runtime)unix.cmxa
andthreads.cmxa
Esperanto replaces the host's C library byCosmopolitan. Then,it outputs a POSIX-approved polyglot format that runs on many platforms. Formore details, please read theαcτµαlly pδrταblε εxεcµταblε.
The goal of esperanto is to provide atoolchain which is able to compile anOCaml project toaarch64
andx86_64
and link these artifacts together toproduce an actually portable executable. Esperanto is primarily a distributiontool rather than a development tool. We advise the user to have a finalizedproject and then use Esperanto to produce a distribution.
The pipeline used to produce such a portable executable must respect certainconstraints:
- not depend on dynamic objects (
*.so
) - the C files required for compilation can also be compiled with theCosmopolitan C library
Let's take the example of a simple program to understand the basic "pipeline"required to produce a.com
(for COsMpolitan) executable. We recommend usingEsperanto in a [Docker][docker] or "clean" system (like a special[opam
switch][switch]). Once again, the aim is to produce an artifact from afinalized project, not to develop an application. So you need opam. Let's takeDocker as an example:
$ docker run -it --platform amd64 ocaml/opam:ubuntu-23-04-ocaml-4.14$ opam pin add -y https://github.com/dinosaure/esperanto.git
Esperanto installs 4 packages:
conf-cosmopolitan
, this package verifies that your system can, indeed, runape programs.esperanto-cosmopolitan
, this package is the Cosmopolitan toolchain asavailable from [cosmocc.zip][cosmo.zip] (v3.1.3) packaged foropam
. Itverifies the integrity of distribution files according to these[fingerprints][./toolchain/cosmocc.digest]
Finally, 2 versions of the OCaml compiler are available:
- To produce an
x86_64
program (x86_64-esperanto
) - To produce an
aarch64
program (aarch64-esperanto
)
Let's start to make a simpleHello World! example:
$ cat>main.ml<<EOFlet () = print_endline "Hello World!"EOF$ ocamlfind -toolchain x86_64_esperanto opt -o main.x86_64.elf main.ml$ ocamlfind -toolchain aarch64_esperanto opt -o main.aarch64.elf main.ml$ apelink -o main.com \ -l$(opam var bin)/ape-x86_64.elf \ -l$(opam var bin)/ape-aarch64.elf \ -M$(opam var bin)/ape-m1.c \ main.x86_64.elf main.aarch64.elf$ sh -c"./main.com"Hello World!
Producing an executable from your OCaml project that is portable on all systemsrequires 2 precise points:
- project dependencies must be available to compile in a consistent context.Such consistency can be achieved by
dune
and a"monorepo" format for yourproject. - C dependencies must be able to compile with Cosmopolitan.[cosmocc.zip][cosmocc.zip] offers a patched version of GCC and Cosmopolitanexposes many functions. However, you may need to patch your C code (inparticular, you can use
__COSMOPOLITAN__
).
If these 2 points are met (getting the dependency sources and havingCosmopolitan-compatible C code), we're ready to produce the final artifact usingthe Esperanto toolchain. So, 3 steps are necessary:
- compile project for x86_64 architecture
- compile project for aarch64 architecture
- link the 2
ELF
files into a portable binary
To achieve our 2 objectives, we're going to usedune
to provide us with thefamous contexts linked to our 2 toolchains (x86_64
andaarch64
) andopam monorepo
to download locally the dependencies needed for our project (theconstraint implies that all dependencies use dune). Let's play with [hxd][hxd],a small project.
You can take your favorite project and compile executables with theEsperanto toolchain. You need to specify two new contexts:
$ git clone https://github.com/dinosaure/hxd.git$cd hxd$ opam repo add opam-overlays https://github.com/dune-universe/opam-overlays.git$ sudo apt install pkg-config$ cat>dune-workspace<<EOF(lang dune 2.0)(context (default))(context (default (name x86_64-esperanto) (toolchain x86_64_esperanto) (host default)))(context (default (name aarch64-esperanto) (toolchain aarch64_esperanto) (host default)))EOF
Finally, as a static program, you must fetch dependencies withopam monorepo and build your project with them:
$ opam monorepo lock --ocaml-version=4.14.1$ opam monorepo pull$ dune build bin/xxd.exe$ apelink -o xxd.com \ -l$(opam var bin)/ape-x86_64.elf \ -l$(opam var bin)/ape-aarch64.elf \ -M$(opam var bin)/ape-m1.c \ _build/bin/x86_64-esperanto/xxd.exe.dbg \ _build/bin/aarch64-esperanto/xxd.exe.dbg
Now, you have a portable/polyglot program! The user canassimilate theprogram with the--assimilate
option. Assimilation consists in "specializing"the executable for the host system.
$ sh -c"./bin/xxd.com --assimilate"$ uname -px86_64$ file bin/xxd.combin/xxd.com: ELF 64-bit LSB executable, x86-64
The--assimilate
option modify the executable itself to becomereallynative to your platform - in other words, the executable no longer becomespolyglot!
Such design comes fromMirageOS andSolo5. For C stubs, thecosmocc
toolchain provides the__COSMOPOLITAN__
definition which permits toorchestrate your compilation of C files according to the OCaml toolchain.
Currently, we support OCaml 4.14.
Currently,zsh < 5.9.0
does not support well Cosmopolitan/APE binaries.However, the recent version fix the initial issue. In the case of you havezsh < 5.9.0
, you can use the--assimilate
option to modify the executableto areal native application:
$ zsh --versionzsh 5.8.1$ zsh$ ./a.outzsh:exec format error: ./a.out$ sh -c"./a.out --assimilate"$ ./a.outHello World!
If the issue persist with Pythonsubprocess
or old versions of fish, we alsoadvise an "assimilation" in these cases. However, as explained above, the binaryno longer becomespolyglot!
An other issue on Linux is aboutbinfmt_misc
which wants to interpret thebinary as a Windows executable (due to the header of the APE binary). A goodexplanation is available here:jart/cosmopolitan#2
About
build-once run-anywhere OCaml programs