OCaml Examples

Install Opam

To install opam, follow these instructions.

Building with Dune

To install, run

opam install dune

followed by

opam init

Check version

Close the terminal. Then re-open it to determine the version of dune you are running with

dune --version

Quiet output

To prevent build progress reports from being printed before your programs run, create a global dune config file.

On unix systems, this will be located at

~/.config/dune/config

Add the following code to the file, replacing 3.15 with the version number that you are running.

(lang dune 3.15)
(display quiet)

Useful shell function

Dune New Project (dnp)

dnp() {
  dune init project $1 && cd $1 && dune build
}

Useful shell aliases

Dune Compile and Run (dcr)

alias dcr='dune build && dune exec ./bin/main.exe'

Dune Compile (dc)

alias dc='dune build'

Dune Run (dr)

alias dr='dune exec ./bin/main.exe'

Alternative build system

Setting up ocamlbuild

1. install ocamlbuild with

opam install ocamlbuild

2. install ocamlfind with

opam install ocamlfind

3. use opam to install any dependencies required by the project.

4. in the project directory, create a _tags file with the following code

true: -traverse

5. Use the following functions and alias as shortcuts to compile and run your code.

The functions can be run with either zero parameters, or a list of dependencies.

OCaml build (ob)

ob() {
  if [ $# -eq 0 ]; then
    ocamlbuild -use-ocamlfind main.native
  else
    local pkgs="$*"
    ocamlbuild -use-ocamlfind -pkgs "$pkgs" main.native
  fi
}

OCaml build and run (obr)

obr() {
  if [ $# -eq 0 ]; then
    ocamlbuild -use-ocamlfind main.native && ./main.native
  else
    local pkgs="$*"
    ocamlbuild -use-ocamlfind -pkgs "$pkgs" main.native && ./main.native
  fi
}

Run OCaml binary (rob)

alias rob='./main.native'

Bogus linting errors

The OCaml Platform extension for VSCode is configured to Dune rather than OCamlbuild, so using OCamlbuild projects with it will generate bogus linting errors.

I don't know if this problem also occurs in terminal-based editors where an OCaml LSP is set up, but I dare say that it might.

One workaround for this problem is to use VSCodium instead of vscode, and to install the OCaml Platform Syntax extension rather than the OCaml Platform extension, but of course, this will only give you OCaml syntax highlighting, not linting or autocompletion.

Can I delete the build folder?

You can delete the build folder, but keeping it will make re-compilation faster.

OCaml Syntax Primer

The OCaml entry on Learn X in Y Minutes is very good.

Language Overview

The Read World OCaml site has a great Guided Tour.

YouTube Tutorials

For video tutorials, I recommend Michael Ryan Clarkson's extensive OCaml Programming playlist.

AI Code Generation

Large Language Models (LLMs) are only effective for me because I don't trust their output.

I always test the code that they generate!

However, given that OCaml is an old language that is fairly popular in academia, there is a suprising amount of training data for them to work with.

Function Generation With LLMs

I have had good results with prompts using the following template

Write a [name] function in OCaml that takes
[name the parameters and their types] and returns
a [type] such that [describe what the function does].
Then show me the code.

Another strategy is to provide the LLM with a list of function signatures for a particular utility or feature and ask the LLM to implement the function bodies, one by one, asking for your feedback at each step.

Example of Program Generation with Chat-GPT4

I wrote a lexer for parsing s-expressions and had the following lengthy conversation with Chat-GPT about how to turn the lexer into a parser, then an evaluator and finally a REPL.

Along the way, the LLM made errors, but when I pointed them out, it was able to self-correct and arrive at a solution.

Hacking the context window limit

Although feeding errors back to Chat-GPT4 can yield good results, sometimes a conversation will drag on so long that the LLM "forgets" what you are talking about and the chatbots answers rapidly become useless.

This is because of a fundamental limit built into the LLM. Namely the maximum size of its context window. The size of the context grows as the conversation progresses.

To avoid running up against this limit when you are working on a code generation task, simply delete the chat you are using, start a new one, attach the source code that needs fixing to the chat (the "Advanced Data Analytics" plugin should be enabled), tell the chatbot that there are syntax errors or unexpected behaviour and ask for its help.

This process can be repeated as often as necessary!