The ts-node app
I have a technical blog which is built on Docusaurus. When the Docusaurus build completes, a post processing script runs to do things like:
- update the
sitemap.xmlto include the
lastmoddate based on git commit date, and truncate the number of entries in the file
- patch the html files to use Cloudinary as an image CDN for open graph images
These scripts are implemented as a simple ts-node console app. For historical reasons it's called
trim-xml (it originally just truncated the
sitemap.xml file). It's not a particularly good name but I'm not going to change it now. As the blog is open source, you can see the [code of
What we're interested in, is porting this app from ts-node to Bun. The app has a few dependencies; so npm compatibility is important to us. Let's see how it goes.
I installed Bun on my Ubuntu machine using the following command:
curl -fsSL https://bun.sh/install | bash
Which resulted in the following output:
bun was installed successfully to ~/.bun/bin/bun
Added "~/.bun/bin" to $PATH in "~/.zshrc"
To get started, run:
I was a little weirded out by the inconsistent indentation in the output but I'm sure that's just a formatting issue. (I submitted a PR to fix this.) When I ran the suggested commands it looked like bun was happy and healthy.
Porting the install from yarn to bun
With bun in place I was ready to port the app. I opened up the (as I say, badly named)
trim-xml directory and triggered installation of the dependencies using
Output looked like this:
bun install v0.5.7 (5929daee)
5 packages installed [2.34s]
As well, a new
bun.lockb file had appeared in the directory alongside the
package.json. Although I can't find any documentation on it, I'm guessing that this is the Bun equivalent of
yarn.lock. It's a binary file, so you can't read it. I did find this project which allows you read bun.lockb files which looks like a useful way to solve that problem.
To avoid confusion, I also deleted the
yarn.lock file. Yay - I've installed things! And pretty fast! What next?
As I looked at the output for the install I realised that the
@types/node package had been installed. The
@types/node package is a package that contains TypeScript definitions for the Node.js runtime. Given we're moving to using Bun, it seemed likely that I didn't need these. But I likely did need something that represented the Bun runtime types. (Which incidentally, I would imagine to be pretty similar to the Node.js runtime types.)
I had a quick look at the Bun documentation and found the
bun/types package. I added it to my project, whilst removing
bun remove @types/node
bun remove ts-node
bun add bun-types
Output looked like this:
bun remove v0.5.7 (5929daee)
1 packages removed [3.00ms]
bun remove v0.5.7 (5929daee)
1 packages removed [843.00ms]
bun add v0.5.7 (5929daee)
1 packages installed [1.97s]
The docs also say:
Add this to your
// "bun-types" is the important part
I aligned my existing
tsconfig.json with the above. For my console app this meant the following changes:
- "target": "ES2022",
+ "target": "esnext",
- // "lib": ,
+ "lib": ["ESNext"],
- "module": "NodeNext",
+ "module": "esnext",
- // "types": ,
+ "types": ["bun-types"],
moduleResolution with Bun
I'd imagined that at this point I'd be able to run the app, but when I navigated around in VS Code I saw that I had a bunch of errors. I was getting errors like this:
The error message was suggesting I needed to explicitly state that I wanted to use the Node.js module resolution algorithm. Whilst we're using Bun, we're porting a Node app - so this made sense. So I made one more change to the
tsconfig.json to satisy this:
- // "moduleResolution": "node",
+ "moduleResolution": "Bundler",
With that in place, the module resolution errors were... resolved. (Sorry.)
File APIs with Bun
However, I was still getting errors. This time they were about the
fs.promises API. I was getting errors like this:
It looked like the version of bun I was using didn't support that API. As I dug through my code I realised that I was using the
fs.promises API in a few places. I was using it in the following ways:
- `await fs.promises.readFile`
+ `await Bun.file(path).text()`
- `await fs.promises.writeFile(path, content)`
+ `await Bun.write(path, content)`
There appeared to be no Bun equivalent for
fs.promises.readdir, so I used the sync Node.js API:
- `await fs.promises.readdir`
We now had code without any errors. (At least in VS Code as far as TypeScript was concerned. I had yet to run the app to see if it worked.)
I was tweeting about my findings as I wrote this, and Jarred Sumner (who works on Bun) was kind enough to share that the
fs.promises API is implemented but the types aren't as yet.
Running the app
I now needed to do one more thing:
- "start": "ts-node index.ts"
+ "start": "bun index.ts"
That's right; update the
start script in
package.json to use
bun instead of
ts-node. And now I was able to run the app with
Reducing 526 urls to 512 urls
The first positive thing about what I saw, was that we appeared to have running code. Yay! The program also appeared to be executing instantaneously, which seemed surprising. I was expecting Bun to be faster, but this seemed too fast.
Also, we seemed to be lacking many of the log messages I'd expect. I was expecting to see about 1000 log messages. Something wasn't right.
await and Bun
The issue was that my
main function was asynchronous. However, because support for top level
await wasn't available in Node.js when I originally wrote the code, I'd called the
main function synchronously. Fortunately Node didn't complain about that, and the program behaved in the way required.
However Bun looked like it was respecting the fact that
main was asynchronous. That's why it was apparently executing so quickly; it wasn't waiting for the
main method to complete before terminating.
To be honest, Bun's behaviour here is just right; the code as is didn't suggest that it was interested in waiting for the
main function to complete. But it turns out that waiting is exactly the desired behaviour. To bring things right, we could use top level
await. So I made the following change to my
+ await main();
And now I was getting the expected log messages; and the program appeared to be working as expected.
GitHub Actions and Bun
I was now able to run the app locally. But I wanted to run it in GitHub Actions. I just needed to add the
setup-bun action to my workflow, so bun was available in the GitHub Actions environment:
- name: Setup bun 🔧
Performance comparison; Bun vs ts-node
I was expecting Bun to be faster than ts-node. Let's take a run of our app in GitHub Actions with ts-node and compare it to a run of our app with Bun:
Post processing finished in 17.09 seconds
Done in 19.52s.
Post processing finished in 12.367 seconds
Done in 12.72s.
I haven't done any formal benchmarking, but it looks like Bun is about 50% faster than ts-node for this usecase. That's pretty good. It's also worth expanding on how this breaks down.
You'll notice in the logs above there's two log entries:
- The "Post processing" reflects the time taken to run the
- The "Done" reflects the time taken to run the
buncommand end to end.
What can we learn from this? First of all, running code in ts-node takes 17 seconds, compared to 12 seconds with Bun. So Bun is performing about 40% faster at running code.
The end to end is 19 seconds with ts-node, compared to 14 seconds with Bun. So Bun is performing about 50% faster end to end. There's two parts to this; the time taken to compile the code and the time taken to start up. We're doing type checking with ts-node; which if deactivated would make a difference.
However, when you look at the difference between the end to end runtime and code runtime with Bun, it's a mere 0.353 seconds. ts-node clocks in at 2.43 seconds for the same. So ts-node is about 6.5 times slower at starting up. That's a pretty big difference; it's unlikely that all of this is TypeScript compilation; Node.js is fundamentally slower at getting going than Bun is.
Moving from ts-node to Bun was a pretty easy process. I was able to do it in a few hours. I was able to run the app locally and in GitHub Actions. And I was able to run the app in less time.
This all makes me feel very positive about Bun. I'm looking forward to using it more in the future.