Pingback Podcast - Episode 3 - Sam Ruby

So I wanted something that translates my application, which has now got, I mean, it's a reasonable size application, to TypeScript or JavaScript, so it could actually run on the browser and offline.

And I had another project I started in 2013, Ruby2.js. It's still out there.

This is back when... jQuery and CoffeeScript are popular because JavaScript sucked and browsers sucked and these things helped.

And one of the nice things about JavaScript is people would write JavaScript on the server and the client and be able to write code that runs both places.

I wanted to run Ruby on both places, so I wrote a Ruby2JS transpiler.

And this was, again, 2013 long before LLMs was written by hand.

And it did the job when I used it and probably got a few dozen other people using it.

And after a bunch of years, JavaScript improved and I lost interest.

Another person picked it up, kept it going for another couple of years.

And then now I had the problem again.

So I gained interest again, trying to build my showcase application in JavaScript.

And I started building on top of the base I built 2013 on.

And found that it didn't quite work.

I had to take type inferencing seriously.

So I built a second version of it.

It's called Railcar.

It's out there, and it was built by LLMs, and it could do TypeScript.

And it could do Rust and C Sharp and all sorts of other languages.

And basically anything I asked the LM to do, it said, OK, sure.

But I took a look at it and it was basically, okay, here's a complete implementation for each language.

It didn't actually share any logic.

And that meant if I wanted to add a new feature, I'd have to add it to all of them.

So I took a look at that design and said, no, that's not the right design.

Let's build it like a real compiler.

So, Roundhouse was the third major iteration.