[Translation available in Japanese]

So you want to learn OCaml. Where do you start? What do you do?

I’ve been an OCaml beginner probably a dozen times — picking it up, dropping it, and picking it up again so many times I’ve lost count.

This time it’s stuck, and I think it’s because the community has fundamentally changed.

Here’s what worked for me.

What books are available?

Read Real World OCaml (colloquially: RWO), and accept no substitutes. It might be the best computer language book I’ve ever read in my life.

In addition, I would advise against reading other books, as they tend to be incorrect and/or in French.

(EDIT: a commenter points out that OCaml from the Very Beginning doesn’t suck, and is more beginner-oriented. I’ve never read it, but Ron Minksy says nice things about it. So there’s a data point.)

Here are some nice things about RWO.

• Cogent and well-written.
• Fully available as HTML online. This means (1) you don’t have to buy the book, and (2) you can copy + paste code segments directly.
• Provides clear operational justification for nearly every important language feature.
  • By “operational justification”, I mean that the authors provide enough knowledge of how the feature works under the covers that you could be safe justifying its use for a production system.
  • A good example is their coverage of pattern matching. It turns out that matching desugars down to a finite state machine. So, if two patterns overlap, checking them will look like going through a bunch of very cleverly nested if statements, rather than checking one pattern, and always starting completely over again for the next pattern. Providing this knowledge is valuable to someone like me, who needs this sort of intuition when writing production code.
  • This is in contrast to a lot of books sort of walk you through the core abstractions (e.g., “here’s a for loop”), and the core library (e.g., “call this to get a List<T>), and give you no other context.
• A whole section on the runtime. A whole section!
  • By "runtime” I mean “anything that is running when your program is executing.” That includes things like GC, concurrency management, etc.
  • This is a HUGE DEAL because, if you don’t have good access to knowledge about how the runtime of your production language works, you are taking a huge risk. Yet, almost no language books actually cover much about the runtime at all!
  • This section provides a lot of the knowledge you would need to make this risk manageable. It is so good, I completely reversed my attitude about OCaml. Before, I would never have considered pushing OCaml to production before, whereas now, I feel I understand enough that the risk is plausibly small.
  • Consider this vs another book that’s dear to my heart, Learn You A Haskell. It’s entertaining, and got me through important things (like monads), but it provides almost no coverage of how the langauge runs. This is a critical failure! It basically relegates Haskell to a fun toy I use in my spare time.
• A whole section on tooling to solve important tasks. Includes things like parsing and lexing, serialization, concurrent libraries, etc.
• (There’s also a section about the language, the features, and the Core API, but that goes without saying.)
• Ignores the horrible standard library and tooling. The book is focused around the vastly superior toolchain put out (mostly) by Jane Street Capital and OCaml Labs. Before RWO, the standard library and the standard tooling (e.g., package managers) were a massive stumbling point for learning OCaml, because they were so uniformly terrible. Thankfully this new toolchain is amazing, well-documented, and well-supported, and RWO rightfully centers around this instead.

What tools should I use?

OCaml has a very strong type system. A combination of this fact, plus the fact that the types are usually inferred (i.e., they are usually not written down explicitly), makes OCaml a language where your intuition about what should be correct will be regularly shot down, and then shoved in your face until you get it right.

Please.

Please, please make this easy on yourself.

Invest in tooling that will shorten the gap between writing something and the compiler telling you it’s totally wrong.

Follow the RWO installation instructions to the end. In particular:

• Install utop (the OCaml REPL). You’ll use this to experiment a lot and confirm your intuition by running code.
• Install OPAM (the OCaml package manager). This should go without saying. It used to be hard and annoying to switch the version of OCaml or install simple libraries. OPAM is truly a joy to use compared to those days, and even compared to other languages.
• Editor tooling. Just do it. You want Tuareg and Merlin, which work on both emacs and vim. Don’t use emacs and vim? Use them for OCaml then. The tools are that much better than using another editor and no tools.

Since it takes time and energy to invest in tooling, I’ll try to entice you by showing you some stuff that’s cool that you can do with them.

Files are compiled on save, which means that things that don’t compile are highlighted in yellow:

As you type, Merlin will produce a list of autocomplete suggestions:

Merlin also has a hotkey (on emacs it’s C-c <TAB>) that will bring up a list of suggested autocompletes:

Another hotkey (on emacs it’s C-c C-t) takes the expression that the cursor is currently at and tells you what type it is!!! (It’s included at the bottom of the screen.) This is incredibly convenient because the compiler then doesn’t have an opportunity to complain about types.

If you press this same key combo on an expression that is a type, it simply brings up the type definition!!

There are a lot of things you can do here. This is just a taste. It’s an optional thing, but really, it’s well worth the time savings.

What are some good examples of source I can read?

Good question! It’s important to see really skilled programmers use OCaml in a really idiomatic way!

The best open OCaml system is probably the Jane Street Core and the Jane Street Core Kernel. For example, here is their map implementation, which is very good.

The source for utop and anything coming out of OCaml Labs is good, but unfortunately, good open source OCaml projects are still somewhat hard to come by in general. Eventually this will probably change.

What are some good tutorials?

The standard OCaml tutorials are now awesome. They used to be non-existent, then they were terrible. Now they are awesome.

The 99 OCaml problems exercise is ok. It might not be the best way to learn, depending on your background, but I found it to be a useful way to learn common manipulations for standard OCaml data structures. In particular, I found that writing something, and then looking at the solution, showed me where I was not using idiomatic OCaml. That was helpful.

There are both contained on the the official OCaml “learn” page . It’s a pretty good resource to start, and it will be consistently updated in the future.

Where is good documentation?

A good start is the documentation for Core.Std. This explains the standard data structures of the Jane Street Core, which is a good place to start.

To get started with that, it may be necessary to look through the relevant online RWO chapters. (Certainly this was true for me.)

Who can I talk to?

I have friends who I talk to. I hear IRC is good, and I hear there are mailing lists. Personally I don’t use either.

What do you recommend for beginners?

Everyone’s path varies, but personally I had the most luck with the following.

• Start by implementing applications commensurate to your level of experience. I picked a Boggle solver, a command line parser, a regex parser, and a couple machine learning problems. You might pick something easier or harder, depending.
• Use your editor to check types of expressions often. This reduces compiler frustration. Use your editor compile-when-you-save feature to check if things compile. It’s way faster to do this interactively with an editor than to compile outside your editor.
• I find it very useful to start by sketching out my types and their interfaces before I write any code. YMMV.
• Try finding a dual in OCaml to all the things you like to do in another language. For example, Python’s list comprehensions are a special case of the monadic bind. Sweet! Once you have all these down, you’ll be more comfortable in your language, because translating your thoughts to code gives you something solid to fall back on.
• Read a lot of source if that helps. I found the Jane Street core very useful for understanding the idioms of OCaml.
• 99 problems, etc., let you implement something, and then see idiomatic ways to do it. This can be very very useful.
• Get code reviews. I have friends who help me with this. It’s incredibly helpful for understanding where you do something not OCaml-kosher.

Conclusions

I guess I don’t really have anything to say in parting, other than that if you have comments about what does/does not work, I’d be happy to hear them.

comments powered by Disqus