Talks
Keynote: Developing a Language
#programming-language-design
#smalltalk
#ruby
#abstract-syntax-tree-ast
#parsing
#domain-specific-language-dsl
Summarized using AI
Keynote: Developing a Language
by Evan PhoenixThe video titled "Developing a Language" features Evan Phoenix as a speaker at the LA RubyConf 2011. The talk explores the conceptual and practical aspects of creating programming languages, delving into the philosophical underpinnings of language and abstract ideas. The session is interactive, inviting audience engagement and emphasizing learning through implementation. Here are the key points discussed throughout the talk:
Introduction & Ground Rules: Evan begins by thanking the audience and organizers and sets the stage for an interactive session, encouraging questions throughout the presentation rather than saving them for the end.
The Purpose of Creating Programming Languages: He poses philosophical questions about language and thought, discussing how ideas and language are intertwined. A good programming language can facilitate better expression of ideas.
Learning Through Creation: Evan emphasizes that the best way to learn about programming languages is by actually creating one. He introduces "Prattle", a Smalltalk-inspired language that he and the audience will begin to develop collaboratively.
Key Concepts from Smalltalk: Evan mentions four fundamental concepts from Smalltalk (self, true, false, nil) and explains their similarities with Ruby. He introduces parsing as a key step in language development.
Parsing Logic: The audience is walked through the process of writing parsers for various data types and constructs within the new language, including numbers, logical values, and strings.
Building the REPL: He discusses the importance of implementing a Read-Eval-Print Loop (REPL) for immediate feedback when testing the language code.
Method Calls and Sending: He expands the language's functionality by adding method sending features and explains how to manage arguments in method calls, reflecting Smalltalk syntax.
Incremental Development: The session emphasizes the necessity of validating each additional feature through testing, underscoring the importance of structure and order in building a language.
Final Thoughts and Encouragement: Evan wraps up by motivating the audience to explore programming language development, reassured that challenges are manageable and resources for learning are accessible.
00:00:08.880
Hello everyone!
00:00:29.840
Thank you all for coming out! I hope everyone has a great day today. Alright, let's hear from you quickly.
00:00:36.399
Let's give a round of applause for the organizers, Kobe, JR, and everyone involved!
00:00:47.280
This is going to be an interesting talk. We'll see how it goes!
00:00:54.239
A few quick ground rules: if you get lost and want to ask a question, please raise your hand and I'll answer it right there.
00:01:00.559
This is not the kind of talk where we can wait until the end to discuss questions because we’ll have no context by then.
00:01:05.680
Is everyone cool with that? Great!
00:01:12.240
This talk is titled 'Developing a Language'. Here is my Twitter handle and my GitHub account.
00:01:19.280
I created this slide just a couple of minutes ago.
00:01:25.520
Now, this is talk number 11. You know what that means!
00:01:30.640
Oh wait, these are out of order. Oh well, that's okay.
00:01:35.680
I want to give a shout out for LARB. So you all know about LARB, right?
00:01:40.880
Who doesn't know about Tuesday night hack? Alright!
00:01:47.040
As I was saying, this is the 11th talk and the last talk of the day, so everyone's thinking about dinner.
00:01:54.159
But we won’t get to dinner just yet. First, I’m going to give you some topics to discuss with your friends over dinner.
00:02:00.479
For that discussion, we are actually going to write a new programming language together.
00:02:06.880
Let’s see how this goes!
00:02:12.640
The big question is, why do we create programming languages? This leads to a philosophical question.
00:02:19.120
Which came first in society or human evolution: language or abstract ideas?
00:02:25.360
Is the ability to express an idea the genesis of having that abstract idea?
00:02:30.560
Or do we need to have the abstract idea first in order to realize we have to express it?
00:02:35.920
It’s tough to say. In many ways, having an abstract idea is tied to having the language to express it.
00:02:41.840
If you have an abstract idea, it's hard to do anything with it without the ability to communicate it.
00:02:48.000
A wise person once said that a programmer, like a poet, works only slightly removed from thought.
00:02:54.959
So programming is essentially about abstract ideas and the language used to express those ideas.
00:03:01.200
If ideas are manifestations of language, does a better language lead to better ideas?
00:03:06.640
This is the notion behind the pragmatic programmer idea: learning a new language makes you a better programmer.
00:03:11.680
Seeing how others express similar ideas can help you uncover ideas you didn’t know about.
00:03:17.760
However, newer ideas in programming languages are not always better than older ones.
00:03:23.519
You could spend your life learning about the ideas that others have already articulated.
00:03:28.720
Looking back at older languages is also valuable.
00:03:34.879
The best way to learn a language, however, is to implement one, and that's what we're going to do today.
00:03:41.840
Let’s take a look back in time at one of Ruby’s ancestors, which is Smalltalk.
00:03:48.640
This talk will not be a full-fledged Smalltalk presentation. Instead, the purpose is to whet your appetite and show you how you can express ideas. If you want to work on a language, you'll find that diving in is quite easy.
00:05:00.720
The original idea for this talk was to develop a piece of software called the Rubinius Language Kit.
00:05:08.800
The intent was for you to easily implement things using this kit.
00:05:15.440
But as happens with many good programmers, I decided that before starting on this, I needed to write a bunch of companion software. Consequently, I ended up building a simple parsing library.
00:05:30.640
Then I realized that it was already yesterday, and I hadn’t even gotten to the actual toolkit.
00:05:36.080
What I will present today is a language, and after this talk, we will figure out how to extract pieces from this and build a toolkit.
00:05:42.320
Thus, 'Prattle' is the name of our Smalltalk-inspired language we will create today. This serves as the genesis of our language toolkit.
00:06:00.880
If you have a laptop, you might want to get it out now.
00:06:06.960
There is a lot of code in these slides, and I had to condense it down for clarity. The code for all of this is already online. You can clone it if you wish.
00:06:13.199
I will give you a moment to do that while Mitch asks questions.
00:06:24.479
So hopefully, you can clone this now, and the Wi-Fi is satisfactory.
00:06:30.479
As I mentioned earlier, this whole project has turned into a sort of road to nowhere.
00:06:36.800
With that said, I present to you 'Prattle'!
00:06:42.160
Smalltalk is very much like Ruby, but in truth, Ruby is more akin to Smalltalk.
00:06:47.520
We'll start with the four most basic concepts in Smalltalk: self, true, false, and nil. They are identical to those in Ruby.
00:06:53.280
Now, as a quick note, when looking at the code, you can find it in the future if you're watching this online.
00:07:00.320
So, remove your windows to this part of the screen.
00:07:06.479
You'll be looking at one of the first few commits, which focuses on adding the ability to parse 'self'.
00:07:12.000
This part is essential because we're talking about writing both a parser and the ability to run the parsed code.
00:07:19.200
The initial commit looks somewhat daunting, but the key portions are what we want to focus on.
00:07:24.320
So, how are we going to parse the 'self' statement? We will create a rule called 'self' that matches a string literal for 'self'.
00:07:32.240
When matched, we will return a new self object.
00:07:39.440
Keep in mind, we are inside this self class, so the naming convention applies. This is quite straightforward.
00:07:45.440
Now you can say that you've successfully parsed a Smalltalk program. Congratulations!
00:07:52.320
Next, we will look at 'true', 'false', and 'nil'. It’s pretty straightforward.
00:07:57.760
Let’s move on to something a little more complex: numbers.
00:08:03.440
We want to create a rule for numbers, and we can do this from a regular expression.
00:08:10.080
This process makes it easy to recognize and parse numbers as we progress.
00:08:16.079
Now, we have five rules established, but we need to bind them together.
00:08:24.399
We need a root node that specifies what constitutes a valid Smalltalk program.
00:08:30.240
Whether it's true, false, nil, a number, or any of those valid identifiers.
00:08:36.000
We can specify valid inputs that can be parsed.
00:08:41.279
Writing a language is all about instant gratification. You need a REPL (Read-Eval-Print Loop) to input text and see what happens.
00:08:48.080
When you achieve this, you get a fantastic feeling of accomplishment!
00:08:54.720
In this case, we fired up the REPL that comes with the project.
00:09:02.000
We put in a number like '42', and it responds back with a parsed number node with the value of 42.
00:09:09.760
The REPL also confirms our string inputs like 'true' and 'nil' by returning the respective nodes.
00:09:15.760
Now we’re able to successfully parse and react to inputs!
00:09:21.840
The next biggest challenge is determining actions based on the values we parsed.
00:09:29.600
When we hit a true or a number or a nil, we need to define what the code will do.
00:09:36.640
For the upcoming examples, we will use the ability to inject methods directly into the bytecode.
00:09:43.680
You do not have to concatenate Ruby strings; instead, use a programmatic API to construct your code.
00:09:50.560
With 'self', the process is quick. We add a simple rule that says to push the value of self to the stack.
00:09:56.640
We implement this for all values we defined, ensuring that they are pushed correctly to the stack.
00:10:03.360
Now we will move on to a more complex rule: strings.
00:10:10.160
Parsing strings is slightly harder because we must account for escape characters like quotation marks.
00:10:17.200
Rather than diving into code, let’s look at the grammar output, which will be clearer.
00:10:23.360
In this case, characters will either have escape sequences or not have quotes.
00:10:30.240
A string consists of a quote, a body of text, and another quote.
00:10:37.280
Note that Smalltalk only supports single quotes for strings; double quotes are for comments.
00:10:44.319
In the future, you could implement support for double quotes.
00:10:50.960
Diagnosing issues is key, so let’s check our REPL as we run various strings.
00:10:57.680
If we enter '42' again, it should parse and return its value, confirming it’s working.
00:11:04.320
So far, we’ve been extracting values; however, it appears somewhat dull.
00:11:11.040
Let’s implement method sending so that the REPL is more interesting.
00:11:17.600
In this case, we will send the number three to the method 'class'.
00:11:24.320
In Ruby, this would look like calling '3.class'.
00:11:30.560
The rule to handle this involves parsing the number followed by a space and the method name.
00:11:37.200
The parsed output would also correspond to the grammar.
00:11:43.200
Now we can send the method 'class' to three. Fantastic!
00:11:50.320
But what if our object is now a 'true' value?
00:11:56.320
We need to ensure our grammar is flexible enough to handle both types.
00:12:02.240
So let’s create an 'atom' rule that encompasses both types for method sending.
00:12:08.800
The atom rule should encapsulate various entity types such as numbers, true, false, and nil.
00:12:14.480
This technique helps in constructing the unary send syntax.
00:12:20.480
Once that’s established, it’s straightforward to implement.
00:12:27.600
We will now introduce scenarios where the receiver can also be a unary send.
00:12:33.520
Thus, any valid expression can be a receiver.
00:12:40.320
This property authenticates the rule we established before.
00:12:46.320
So now we have enhanced our programming capabilities.
00:12:52.240
In this case, we are adapting to send the method to the unary send itself.
00:12:59.040
The processing grammar is set up to handle this modification gracefully.
00:13:06.240
Now we want to run the REPL to verify if it executes correctly.
00:13:12.240
Coupling the grammar and the bytecode should yield the correct output.
00:13:18.080
Let's execute to confirm our understanding of method calls.
00:13:24.080
Now we can call methods up to our expectations.
00:13:31.040
The next discussion will cover how to pass methods in Smalltalk style.
00:13:38.080
Passing methods is seamless; we will use the keyword format.
00:13:44.080
The format is simple: send the method name followed by a colon and then an atom as an argument.
00:13:50.960
This execution leads to methods being sent with arguments.
00:13:56.960
However, to send multiple arguments, you must lay them out properly.
00:14:03.840
This approach emphasizes the structuring of method calls.
00:14:10.480
Smalltalk enforces this structure by requiring arguments to be method names.
00:14:16.480
If two arguments are passed, they become part of the method name.
00:14:22.240
To manage this, we must determine how to break down the method calls efficiently.
00:14:28.240
With valid grammar rules in place, we can manage passed arguments.
00:14:34.080
We essentially separate pairs into the correct order.
00:14:41.680
As previously mentioned, we'll couple all this together using the REPL.
00:14:48.480
What we’ll see is that this becomes a very useful tool.
00:14:55.520
Let's test it out and ensure we've covered all methods.
00:15:02.480
Now, we have to consider how to manage variations properly.
00:15:08.480
While doing so, we must maintain accuracy.
00:15:14.800
So let's arrange each call to facilitate error-free executions.
00:15:21.920
This leads us to show how to handle call syntax for testing.
00:15:28.560
Everything appears well-ordered since we've put the fine details together.
00:15:35.160
Now, let’s switch gears and explore how unary sends can be received.
00:15:42.000
Let us check how this reflects when we run it in the REPL again.
00:15:48.440
As expected, it preserved the sequence for evaluation.
00:15:55.920
With that, we are prepared to handle object-oriented method sending in Smalltalk format.
00:16:03.120
Next, we’ll set up receiving class-based calls.
00:16:10.240
It's crucial to handle the syntax efficiently.
00:16:17.920
When applying these methods correctly, we can gain satisfactory outcomes.
00:16:24.000
Let’s execute this setup in the REPL for optimal results.
00:16:31.040
This illustrates how methods can be easily instantiated and utilized.
00:16:38.240
Now, we’re eager to see the transformations take place.
00:16:45.920
From multiple method calls to integration of additional language features.
00:16:52.320
In essence, we are bringing in flexibility for object interactions.
00:16:59.200
As you start implementing more robust features, validate outcomes in real-time.
00:17:06.080
We’re promoting interaction with extensive capabilities through the REPL.
00:17:13.040
Now, let’s ensure all the components tie together for solid execution.
00:17:20.080
Correlating these aspects points towards comprehensive evaluations.
00:17:27.280
If your expectations yield satisfying results, you’re on the right track.
00:17:34.160
We’re assuming a colorful variety of outputs should surface as we proceed.
00:17:41.440
These programmatic expressions will be a reference point.
00:17:48.080
As you expand into multiple send commands, the language will flourish.
00:17:55.200
The big takeaway: building a language depends on organized methods.
00:18:02.400
Each addition builds toward a satisfying programming experience.
00:18:09.120
With every improvement, test claims mature into detailed implementations.
00:18:16.000
Once checks and balances are established, a functioning system takes form.
00:18:23.000
Let’s explore further functionality based on incremental development.
00:18:30.560
We'll continue to develop interfaces, enhance coding modules.
00:18:37.600
With that, we’re moving towards complexities that show programming depth.
00:18:44.880
Bridges between small talk methods and Ruby conventions will enrich methods.
00:18:52.000
Now, initiating blocks falls into play, and size expands.
00:18:58.640
Blocks defined here will also add layers of versatility.
00:19:06.080
Coding in blocks pairs smoothly with the existing Ruby code interface.
00:19:13.680
Having multiple expressions supports further exploration without conflict.
00:19:20.720
Let’s investigate these blocks practically — data-defined responses should emerge.
00:19:26.880
This way, testing dialogues become part of efficient learning paths.
00:19:34.080
With blocks under consideration, techniques continue to streamline execution.
00:19:41.200
Let’s harness interactions to solidify our coding outcomes.
00:19:48.400
To finalize our aim, let’s summarize how the last steps connect.
00:19:55.680
Reflecting these associations as each point turns into core functions.
00:20:02.080
Now, we have arrived at the final threshold.
00:20:09.280
We will draw our larger conclusions on how development iteratively enhances languages.
00:20:16.080
In summation, consider building languages as an opportunity.
00:20:23.040
Experimentation thrives in this space, so take the plunge!
00:20:29.920
Learn through the challenges of language design. They’re not as daunting as they had seemed.
00:20:36.000
Feel confident — resources are available for you.
00:20:43.200
With these words, I encourage you to go out and explore!
00:20:50.240
Thank you all very much!
00:20:56.960
If you have questions, feel free to ask as we wrap up.
00:21:03.360
I appreciate everyone’s attention throughout this talk.
00:21:10.720
Thank you again, and see you next time!
Explore all talks recorded at LA RubyConf 2011
+5