Forays into Servo #0

My first experience with browser engines was Ladybird, which I contributed quite a few PRs (around 80) to in the summer of 2024. Coming around to fall of this year, I felt the drive to get back into web browser hacking. The only issue was that Ladybird is no longer accepting non-team contributions.¹

Although I was disappointed at the news, in retrospect, I'm happy that I looked for other projects to work on. Namely, Servo! As a bonus, Servo has an amazing open source community that I will talk about throughout the post.

This post serves as my first progress report on my work, keeping the spirit of my original Ladybird posts!

First Issue

Unlike with how my work in Ladybird started, I did not begin work on Servo with a clear goal in mind. Fortunately, the Servo maintainers frequently update a list of good first issues and triage pretty much every issue that is opened.²

I ended up finding something that I thought would be relatively easy to tackle: giving focus to text control elements when the .select() method is called (#47753). For example:

<!doctype html>
<textarea id="Z">Testing</textarea>
<button id="Y">Click here</button>
<script>
  Y.onclick = () => Z.select();
</script>

Clicking the button should select all text in the <textarea>, along with giving it focus. The PR I wrote to implement this behavior was pretty simple.

However, I ran into failing tests as soon as submitted my PR. Like many browsers, Servo uses the WPT test suite to test compliance with Web standards. The failing WPT test looked something like this:

<!doctype html>
<textarea id="Z">Testing</textarea>
<script>
  Z.addEventListener("keydown", () => {
    Z.style.display = "none";
  });
  Z.select();
  sendKey("a");
  assert(Z.value == "Testing");
</script>

Essentially, this test tries to ensure that setting a <textarea> display to none makes it unwritable, even when it is hidden during a keydown event listener.

This test was passing before, so why is it failing with my change? The problem is that Z.select() is now focusing the element, whereas before it wasn't. So before, sendKey("a") wasn't actually sending anything to the <textarea>, which means that the test wasn't actually testing anything.

So my first PR to Servo uncovered a different issue, one that turned out to be slightly more complicated.

Text Editing

This new discovery led me to start exploring how text controls work in Servo. I eventually opened another PR to fix the bug I had found: #48165.

Getting both my original and induced PR was a bit of an arduous process, as I had to rebase my changes onto a large-ish refactor for text controls. But the benefit was that I got a solid understanding of how text editing in Servo works!³

This led me to a number of text-related contributions:

1. No longer allowing copy/cut in <input type="password"> (#48404)
2. Implementing the beforeinput event (#48519)
3. Move handling composition (IME events) into EditingAction (#48690)
4. Implementing CMD-delete for macOS (#48781)
5. Writing tools to unit test composition events (#48825)

These PRs were a lot of fun to write. Particular highlights for me were PRs numbered 3 and 5. Composition events and IME handling in general in Servo is far from complete. There are all sorts of fun problems to solve here. For example: Servo is designed to be emebeddable. What is an appropriate interface so that an embedder of Servo can communicate that IME events are happening? IMEs are inherently very platform-specific, so this is a difficult abstraction to make in practice.

My focus now, for the time being, will be to improve Servo's IME handling. Although it's difficult for me to test alone, since IME behavior varies wildly from OS to OS, there are certainly many improvements that I can make. My next post in this series will likely be on these improvements.

Servo's Community

Servo has one of the most approachable open source communities that I've worked in. There are immediately obvious things, such as the issue triaging that I mentioned earlier. But beyond the surface, Servo's maintainers have more robust ways for new contributors to get comfortable with the codebase, owing to the Zulip.

Servo's Zulip is very active and I found it to be a goldmine for understanding Servo. Even in conversations that I didn't contribute to (which is most), just reading various people's projects and issues was immensely helpful.

In addition, maintainer Martin Robinson holds weekly office hours! I've attended one office hours session so far, and got to ask some valuable questions.

I can't stress enough how smooth it felt to go from small bug fixes to larger, slightly more consequential PRs. For this reason, I believe that Servo is a great example of how productive and inclusive open source can be.

Conclusion

Working on Servo has been a joy so far. I hope to continue with these updates as I work on the codebase! In particular, I am working on designing an appropriate abstraction for embedder to WebView communication about IME events.

If you enjoyed this post, or have any feedback, please send me an email at mail@dzfrias.dev.