Talking Small

For the May language of the month, I got to play with a language that I've been looking forward to sinking my teeth into for a long time: Smalltalk. I've had a fascination with Smalltalk since I learned of it in college, but I didn't "get" it at the time due to lack of learning materials.

Like Lisp, Smalltalk has multiple implementations to choose from; while Lisp implementations tend to be more like dialects (split across the Common Lisp and Scheme languages), Smalltalk implementations are more like separate languages that share grammar but differ in vocabulary. Many of the offerings are commercial; a few, such as Squeak, GNU Smalltalk, and Pharo, are free as in speech and as in beer. Pharo 5.0 recently came out, so let's give that a spin!

Learning To Talk the (Small)talk

Once you've downloaded and started Pharo, learning the language and the environment is pretty simple. You need only open a playground (a sort of scratch buffer for any kind of text, including code), type in ProfStef go, right click and click 'Do it'. This starts up an interactive tutorial that introduces you to the basics of the language and libraries.

For a more in-depth Smalltalk education, I consulted Pharo By Example. It's a free book that covers the language, the Pharo libraries, and the tooling more in depth. Unfortunately, the version on that site is a little out of date; I recommend checking out the PDFs distributed in this GitHub repo for a more up-to-date experience.

The preface to Pharo by Example contains this quote, which I think is very important to internalize as one is familiarizing oneself with Smalltalk. As someone who really likes to dig deep and learn how things work on why, I find this especially pertinent:

Try not to care. Beginning Smalltalk programmers often
have trouble because they think they need to understand
all the details of how a thing works before they can use it.
This means it takes quite a while before they can master
Transcript show: 'Hello World' . One of the great leaps in OO is
to be able to answer the question “How does this work?”
- Alan Knight

Smalltalk is Small (ish)

Smalltalk as a language is very minimalistic; some boast that its syntax is simple enough to fit on an index card. This means that to deliver on power, it relies heavily on its standard library. The Smalltalk environment, which I'll cover briefly, is also extremely important - it's the part that Smalltalk is so famous for. The de-emphasis on syntax is why you won't see a lot of code in this post.

One thing that interests me about the small size of the Smalltalk language is this: why don't more Smalltalkers write their own implementation as an exercise, the way that Lispers do? I'm guessing it has to do with the size of the standard library you'd have to reproduce to have something workable, or the tooling is too difficult to reproduce for a single person. It might also come down to culture - considering that quote from above, it seems that Smalltalkers have pride in leaving some work for others and focusing on what matters to them.

And Now For Something Completely Different

One of the reasons that people recommend learning Smalltalk is that everything is so different. You don't really have files to work with; your code is contained in native, living structures that you can poke and prod with tooling or snippets of code. All of your code lives in what's called an image; you can think of these sort of like Docker images, if you're familiar with those, but Docker images focus more on transience and immutable infrastructure. You can acheive that with Smalltalk images, but it also allows you to develop incrementally in a way that feels really natural. Even the tools you're given are citizens of this community of objects - you can inspect and modify them as you wish.

Tooling

Speaking of tools, Smalltalk has some of the best. Here are just a few examples of some of the incredible tools that Pharo ships with.

Method Finder

When you start working with Smalltalk, seeing all of the classes and their methods splayed before you can be a bit intimidating. "How do I convert a string to lowercase?", you might wonder. You can use the browser to scan through the various string methods, or you can try the spotter to search for methods containing words like 'lower' or 'down'.

You know what you want to do; you want to transform the string 'Foo' into 'foo'. Wouldn't it be cool if you could just ask Smalltalk which method(s) you can use to do that? Well, you can! If you open the Method Finder and click the box to search by example, you can enter the following, and the method finder will reply with 'Foo' asLowercase:

  'Foo' . 'foo'

I wanted to find a method to clamp a number to a particular range; here's an example of how I used the method finder to do that for me:

This makes me think of Haskell's Hoogle, only in a dynamically typed language, and driven by examples rather than types.

Coding in the debugger

I'm an admitted printf-debugger. I don't use debuggers themselves too often; I'll use gdb to examine the occasional core dump or segfault, but I tend to rely on printf. That being said, using Pharo's debugger was a fun experience. A lot of Smalltalkers talk about "coding in the debugger", which means they write code that they know will fail, focusing on the higher-level details of the problem they're solving. A typical example is invoking a method you haven't written yet; when you're finished thinking at the high level and run the code, the debugger complains "hey, I don't understand this method!", and politely offers to create the method for you. You don't need to do anything special for this - the debugger is always waiting to help you out. I feel like this is a strong embodiment of the Alan Knight quote from above.

After you create and implement the method, you can tell the debugger to proceed, which will invoke the method you just wrote and your application will carry on its merry way. Here's a trivalized example of that in action:

Ch-ch-changes

Let's say you're working on your Smalltalk program, and you do something silly like write an infinitely recursive function. That's one drawback to Pharo - it's single threaded (as far as the OS is concerned), and it implements its own scheduler, so if you bog down that scheduler, your entire environment will freeze.

This actually happened to me while I was working on a program in Pharo, and my only recourse was to kill the VM. Imagine if while working in a more traditional language and testing your program, that your program locked up, and the only way to proceed would be to kill your REPL along with your editor.

Fortunately, Pharo ships with a tool to pull you out of the fire here. Whenever you make a change to Pharo code, whether it's deleting a method, adding a class, or changing a method, that change gets logged in what's called a changes file. There's a tool to work with this file, which allows you to recover from a catastrophe. It also allows you to review the entire history of your image and all of your code within - to bring back my editor analogy from above, it's as if your editor had an undo history that persisted across invocations, forever.

Drawbacks

While working with Smalltalk is fun and a learning experience, there were a few things I didn't like so much.

Long if/else chains are unwieldy. You end up with something like condition ifTrue: [ ... ] ifFalse: [ condition2 ifTrue: [ ... ] ifFalse: [ ... ]], and it doesn't feel great. There's probably an idiomatic way of doing this that I'm missing, though.

Classes are organized into packages to enhance searchability, but that's more like a convention - the namespace is flat. I wonder how often name collisions are a problem in the Smalltalk world!

In Smalltalk, multi-argument methods use a keyword-like "split name" scheme, so instead of this:

  email.sendWithSubject(subject, 'me@example.com')

...you end up with this:

  email sendWithSubject: subject to: 'me@example.com'.

To me, this makes a lot of sense and enhances the readability of the language, but it's not perfect. There are no keywords in method calls - the method's name is 'sendWithSubject:to:' - so if you want something like keywords, you need create a method for each combination of "keywords" you support. Because of this, some Smalltalkers tend to use setter methods and chain them together, like so:

email := Email new.
email subject: subject.
email to: 'me@example.com'.
email send.

"I'm a comment!  There's a shorter way to write this since the invocant
remains the same:"

email := Email new.
email subject: subject;
      to: 'me@example.com';
      send.

I really enjoyed digging into Smalltalk, and I can't wait to share the program I wrote with you. Stay tuned for a writeup on that in the near future!

Published on 2016-06-17