philipp's blog

hecto, Chapter 2: Entering Raw Mode

🏗️ Construction Notice

The old (complete, but outdated) version of hecto is available here. You are looking at the 2024 version of hecto, which is still unfinished. You can follow the progress of the rewrite here. Once it’s done, this notice will be deleted.


Table of Contents

Chapter 2: Entering Raw Mode

philipp's blog Let’s try and read key presses from the user. Remove everything in the main.rs and replace it with the following code:

use std::io::{self, Read};

fn main() {
    for b in io::stdin().bytes() {
        let c = b.unwrap() as char;
        println!("{}", c);
    }
}

Play around with that program and try to find out how it works. To stop it, press Ctrl-C or Ctrl-D.

Once you think you’ve sufficiently understood the program, head over to this step on GitHub for an in-depth explanation of the code and syntax.

If you guessed that the code above would print out any character you type on a new line, you would’ve been wrong. Or half-right - this is what the program does, but it doesn’t immediately get the characters from the terminal for processing.

When you run ./hecto, your terminal gets hooked up to the standard input, and so your keyboard input gets read into the b variable. However, by default your terminal starts in canonical mode, also called cooked mode. In this mode, keyboard input is only sent to your program when the user presses Enter.

This means that our program will let you type whatever text you want, and once you hit enter, it outputs the same text, with every character in a new line, before it waits for new input. If you press CTRL-D, you tell io::stdin().bytes() that the file it’s supposed to read has ended. CTRL-C immediately terminates the program.

This behaviour is useful for many programs: it lets the user type in a line of text, use Backspace to fix errors until they get their input exactly the way they want it, and finally press Enter to send it to the program. This is how terminals work and has nothing to do with Rust. But this does not work well for programs with more complex user interfaces, like text editors. We want to process each key press as it comes in, so we can respond to it immediately.

Press q to quit?

To demonstrate how canonical mode works, we'll have the program exit when it reads a q key press from the user. To do so, add the following code below println :

if c == 'q' {
    break;
}

The break statement will end the for-each loop, and it does so if the character that is read is a q.

To quit this program, you will have to type a line of text that includes a q in it, and then press enter. The program will quickly read the line of text one character at a time until it reads the q, at which point the for..in loop will stop and the program will exit. Neither will the program end immediately once you press q (you have to hit Enter sometime after), nor will it process anything that comes after the q - Rust discards the additional characters while exiting.

Entering Raw Mode

What we want is raw mode, and we’re going to use a dependency to change our terminal to it1. External dependencies in Rust are called Crates (cargo, Crate … get it?), and the one we’re going to use is called crossterm.

We add crossterm by running the following command:

cargo add crossterm

This produces the following output:

    Updating crates.io index
      Adding crossterm v0.27.0 to dependencies.
             Features:
             + bracketed-paste
             + events
             + windows
             - event-stream
             - filedescriptor
             - serde
             - use-dev-tty
    Updating crates.io index

cargo updates its index from crates.io and then adds crossterm to the previously-empty dependencies section of the Cargo.toml. What follows is a list of features which are enabled (+) or disabled (-) for this dependency. Features go beyond this tutorial, but you can read all about it here.

Next time you run cargo build, you will see that cargo will build much more than just hecto, as it now also builds crossterm and the dependencies used by it.

Now, let’s use our shiny new dependency by replacing the contents of main.rs with the following code:

use std::io::{self, Read};
use crossterm::terminal::enable_raw_mode;

fn main() {
    enable_raw_mode().unwrap();
    for b in io::stdin().bytes() {
        let c = b.unwrap() as char;
        println!("{}", c);
        if c == 'q' {
    		disable_raw_mode().unwrap();
            break;
        }
    }
}

Try it out, and you will notice that every character you type in is immediately printed out, and as soon as you type q, the program ends.

Let’s investigate the code changes on GitHub.

Dependency Management

Now is also a good time to discuss dependency management a bit more in detail and connect what we see here with a few things from Chapter 1.

Using dependencies means that we can focus on our coding task without having to reinvent the wheel (and reinvent the same bugs) over and over again.

The Cargo.toml plays a strong role in this. Not only do we list dependencies there, but if we are building a library function ourselves, we can use the Cargo.toml to provide additional information about our package should we decide to publish it. It’s normal, and sometimes unfortunate, that dependencies have other dependencies, and these other dependencies depend on other things, too. This implies that we essentially need to trust the author of the library to a) be a nice person, and b) have trusted or checked that the author if their dependencies are also nice persons, and so on. This can be a serious issue2.

At this point, cargos ability to selectively recompile parts of the code which have changed comes in pretty handy, as recompiling the whole thing might take a bit. It’s not bad yet - that would be concerning for one dependency - but the difference between a fresh cargo build and a subsequent one is noticeable already.

If you poke around in the target directories, as we did in Chapter 1, you will now also notice that the aptly named folder deps is now filled. And you can also notice that the resulting executable for a release build is much smaller than the resulting executable for a debug build!

Let’s also try to understand a final piece of the puzzle, which we omitted in Chapter 1: The Cargo.lock. That file also contains information about the dependencies - even in more depth than the Cargo.toml. But what is it good for?

As described in the GitHub step above, cargo is nice enough to upgrade versions which do not include breaking changes. Let’s say crossterm introduces a bug fix but stays otherwise unchanged, then this behaviour means that we only need to ask cargo to update its index by running cargo update, then a subsequent build would use the updated version. You would only notice a longer build as cargo downloads it.

This poses a problem. Let’s say you have successfully released hecto and it is used around the world. Then someone submits a bug report, and you try to reproduce it. You open your old hecto code base, rebuild it, and you can’t find the error. This means one of the following:

That is pretty annoying. The Cargo.lock attempts to solve one of these problems by providing you with a very specific “fingerprint” of all the dependencies that went into your build. You can build based on the dependencies in the Cargo.lock by using:

$ cargo build --locked

cargo then checks if the Cargo.lock still fits the Cargo.toml (so defining a new dependency or version in the .tomlwhich is not part of the .lock yet creates an error), but it doesn’t use a version different from what has been used before, even if there have been updates in the meantime.

Entering Raw Mode, the Hard Way

To enter Raw Mode, we simply needed to call our dependency to do it for us. Neat!

What crossterm does for us is that it first gets the attributes of our terminals under the hood, disables a lot of them and sets these new attributes. crossterm uses methods from libc, which represent the raw bindings to platforms’ system libraries. Using them directly at this point would require us to use Rust functionality and concepts we don’t understand at this point, to modify obscure terminal settings which we also don’t understand yet, so a library function it is!

If that doesn’t satisfy you, maybe the following will:

  1. We will learn the required Rust concepts later in this tutorial. So once you’ve completed it, you should be able to set Raw Mode by yourself.
  2. The underlying terminal attributes needed for Raw Mode are explained in depth in the original kilo tutorial. The Rust way of doing this isn’t all that different.
  3. If you want to understand the theory without deep diving into the implementation details, you can use crossterm’s implementation as a starting point.

Display Keypresses

To get a better idea of how input in raw mode works, let’s print out each byte that we read. We’ll print each character’s binary representation, its numeric ASCII value, as well as the character it represents if it is a printable character. We do this by changing our main() as follows:

fn main() {
    enable_raw_mode().unwrap();
    for b in io::stdin().bytes() {
        let b = b.unwrap();
        let c = b as char;
        if c.is_control() {
            println!("Binary: {0:08b} ASCII: {0:#03} \r", b);
        } else {
            println!("Binary: {0:08b} ASCII: {0:#03} Character: {1:#?}\r", b, c);
        }
        if c == 'q' {
            disable_raw_mode().unwrap();
            break;
        }
    }
}

Head over to this step on GitHub to learn a bit more about code specifics.

Before we play around with this together, let me quickly tell you some fun facts about printing things in terminals. But first: typewriters.

Typewriters worked like this: A paper sheet was fixed on a drum, which in turn was fixed on a carriage. Every time you’d hammer on a key with a character, a mechanical stamp mechanism would stamp the corresponding character onto the paper and move the carriage a bit to the left. When you got near to the end of the paper, you’d hear an audible “Ding”, telling you that you should now care about a word break as the end of the paper was approaching. Then you would need to push a handle to the right. That handle did two things: It moved the carriage so that you could start the new line fully at the left, and it turned the drum so that you wouldn’t overwrite the previous line but started a new one instead.

All three concepts, the bell, the turning of the drum and the moving of the carriage are still present in computers, though historically, these concepts made their way via teletypes to computers3. To come back to hecto: You might have noticed further up that println! did start a new line, but it didn’t start it on the left, as if on a typewriter we only rotated the drum but did not move the carriage. We fixed this by adding the signal for carriage return (which is actually still called like this), \r, to the output above.

By the way, we’re a bit indecisive in our code about which parts we want Rust to handle (like the newline, which is added for us by println!), and what we want to do ourselves (like the Carriage Return). We’ll fix that later.

Now, let’s use our new program to learn a few things.

Under the hood, characters are just numbers, and numbers are represented in memory as 0s and 1s (We call these “Bits”) internally. Let’s say we have 01100001 somewhere in our memory. It’s easy to turn binary numbers into decimal numbers, and if we do this for 01100001, we get the number 97. There exists a standard which maps a number like this one to a character. This standard is called ASCII, and when you look up 97 in an ASCII table, the result is a lowercase a. Sure enough, if we type a into our new program, we see all three things printed out: a, 097 and 01100001. This is also why we could easily teach our program to treat a binary that we’ve read from the keyboard as a character. Basically we just told the computer “Please treat this number as a character”. The underlying binary number still stays the same, but Rust understands that we want to follow the convention from ASCII and prints out an a whenever we ask it to.

ASCII codes 0-31 are all unprintable characters (you’ll find the bell, the carriage return and the line feed from above in there as well), and 127 is also a control character. ASCII codes 32-126 are all printable. is_control() would therefore simply look at the range of the code4.

This is a very useful program, let’s play around with it some more. It shows us how various key presses translate into the characters we read. Most ordinary keys translate directly into the characters they represent. But try seeing what happens

You'll notice a few interesting things:

Looking closer at the byte representation, you’ll notice that ASCII is designed in a way that you can get from a character to its lowercase counterpart, or to CTRL and this character by just manipulating a couple of bytes. For example, CTRL seems to set the upper 3 bits to 0.

Error Handling

Let’s wrap up this chapter by improving the way we handle errors (as in: We didn’t handle them before, but now we want to). This is also where I’ll make good on my promise and explain the occasional unwrap() we’ve met along the way.

First, some short theory. In many other programming languages, if a function encounters an error, it will not return, but instead “throw an exception”, which you can catch further above and then try to recover. This means that we effectively have two ways a function can return control to whoever invoked it: Either by returning a value, or by throwing an exception, and both of these things require different pieces of code to handle, making it harder to reason about the code.

In Rust, you can return errors from functions just as any other return value, allowing for a much cleaner way to handle exceptional cases. To do this cleanly, Rust has the concept of a Result, which comes in two variants: Ok and Err. Here is a metaphor that we’re going to use for a while:

Let’s say I owe you money and you request that money from me. Then either I give you the money or I tell you that nope, I have nothing to give to you. This is my way of telling you that “requesting money owed” is a function that either returns you some money, or it throws an exception. Note that returning 0 EUR in case I have no money would be invalid in case I actually owe you money, so throwing an exception here comes closer to the mental model of getting the money owed.

Now, instead of money, I return you a box, and I do that both when I have money and when I have no money. In case I have your money, it’s a pink box labeled Ok. Wrapped inside you will find the money you need. Otherwise it’s a black box labeled Err. That box contains information about why I didn’t give you anything (I have no money, or maybe you owe me money?!)

Now - what .unwrap() does is, to stay in the metaphor, to assume the box is pink and take the result out of it. If the box is black, your program will stop then and there.

The behaviour we want is that we pass Errors upwards as far as we can - let them deal with that black box up there! - and at the highest level, when there is no one left to pass the Error to but ourselves, we would finally abort the program.

Let’s build this in this step on GitHub..

Wrap-up and Outlook

In this chapter, we learned about dependencies in Rust, reading user input, the basics of error handling and a surprising amount of information about how terminals and typewriters work. If you found this chapter a bit light on coding, then you have something to look forward to: In the next chapter, we’re going to build the foundations of our text editor.

This post was last updated 1 week ago.

  1. To stay as close to the basics as we can, this tutorial avoids dependencies whenever it can. However, using dependencies are a crucial part of using Rust, so omitting them entirely would be pointless.

  2. This link tells you all you need to know about just how bad it can be to recursively rely on trust.

  3. The Bell was used to get the attention of a nearby operator.

  4. We’ve sidestepped a problem here: Everything I said about ASCII so far is true, but Rust uses a superset of ASCII for chars. We’ll get to the joy of text encodings later in the tutorial.

#hecto