Hacker News Reader

Comment by dimitropoulos | Hacker News Reader

the real story here is an incredible team that managed to simultaneously keep two separate codebases alive for the most advanced type system known to mankind (yeahhh yeahh Hindley-Milner eat your heart out).

huge congrats to the team!

looking forward to the Rust rewrite ;)

View on HN

I am not sure a rust rewrite would be meaningful.

Go is great because it's fast to code.It's easy to reimplement typescript in go 1:1 just by looking at the code.

Rust on the other hand would take a lot longer to develop.

Maybe rust is 20% faster than go but overall the increase from typescript with go is good enough.

Maybe rust would yield a 14 times speedup over the 11 times in vscode but go is already good enough to make a huge difference.

A Rust rewrite would have an easy way to expose an API, something they're still debating how to do and deferring to 7.1.

But the team has already choose. They explained their reasoning and IMO it makes sense: they didn't want a rewrite, they wanted a bug-for-bug file-by-file translation. With a borrow checker and no GC, Rust sometimes forces you to structure things differently (especially in a compiler that usually has a lot of circular structures), so it was not worth it.

The benefit to Rust rewrite would be integration with the rest of the JS tooling ecosystem which is increasingly written in Rust rather than performance.

It probably won't ever happen though.

> It's easy to reimplement typescript in go 1:1 just by looking at the code.

That's also true of Rust if your codebase is written in a functional style. But apparently TSC had a lot of inheritance, which probably isn't a great fit for porting to Rust.

Can you elaborate why that's a factor? The tools are just binaries in how they're used, the language they're written in is no longer a factor then.

It's a good argument if you're talking about transferable skills though, I can imagine some contributors work on both TS and Biome, for example. This is why a lot of JS tools were initially written in JS, too.

> The tools are just binaries in how they're used

They are today, but the potential would be to expose something like the TypeScript compiler as a library. That is possible today with a lot of the JS tooling.

jokes aside, have you heard of the Jevons Paradox[1]? it feels like the "induced demand" effect to me with the whole "just one more lane" phenomenon you sometimes can see in roadways. when you increase the efficiency of a thing you thereby expand the set of things it can economically be used for, causing an overall increase in total consumption over time - not a decrease like you'd expect from just having made it much more efficient. "a smaller slice of a much bigger pie is still more pie" or something like that.

in TypeScript's case with the "pie" being compute time, things like HKTs (e.g. hotscript, hkt-toolbelt) that might not have made as much sense in the past suddenly become so much more feasible, but also are the very things that drag that hard-fought efficiency win back down into the mud. is it worth it? library authors will ultimately be the ones to decide the big chunks of that question by virtue of what they ship in their types.

[1] https://en.wikipedia.org/wiki/Jevons_paradox

Yes, I saw the YouTube video about Jevons paradox from Hank Green yesterday. :)

The Jevon's Paradox. And it is not a paradox :-)

It actually is a paradox, a veridical paradox. But I'm splitting hairs lol

*The

Fine, the Hank Green.

the difference is with roads you dont get a lot of good secondary effects, one lane is just like the next. benefits are linear with the cost so they balance out. but with typescript and software in general they can be exponential.

fast type inference unlocks brand new patterns that were too slow to be practical on the old checker. at least some of them will turn out to be useful for peoples projects. and its also great for legacy or less complex code bases that will get faster type checking for free.

Most complex, perhaps, but not "most advanced". I don't think there's necessarily a meaningful "correct" choice for that title, but surely one of the proof assistant languages would be a more likely candidate?

(I don't say this to be disparaging of TypeScript's type system, by any means — it's very interesting stuff!)

good points, let's get negated types and higher kinded types in there then you've got yourself a deal. maybe regex thrown in too for flavor

Steve Francia (author of Hugo and a bunch of other top Go projects) wrote up some thoughts of Go's fit in the agentic era:

https://spf13.com/p/go-the-agentic-language/

> in my experience, a change that stays local in Go ripples through lifetimes and trait bounds in Rust

imo extensive use of generics/trait bounds and explicit lifetimes in Rust is a huge code smell. Large projects should be making liberal use of trait objects and smart pointers to keep everything understandable and modular. Giving an agent a simple coding practice SOP for Rust should be enough to garuntee basically the same localized refactorability that Go has.

I'm reading a lot of style / flavor / quality / vibes in your reply, something that will be different between every developer (for the most part); the thing with Go is that there's a lot less of that in the wider ecosystem.

> for the most advanced type system known to mankind

Honest question, what do you mean by this?

It means they really like TypeScript and they hope to influence some juniors into investing themselves into it

I answered here https://news.ycombinator.com/item?id=48838629

Are you aware of languages like lean or idris? In that comment you mention doom in ts types.

nope! never heard of them! (note: I'm the guy that did the Doom in TS types thing) what type-level fun do they bring to the table?

Algorithm W is like undergrad level of sophistication. People who like HM more (and I am one) don't like it because it's "advanced" and to some extent exactly because it isn't. It's sound and fast and infers almost everything. TS seems to have one of those features now, so that's nice.

> most advanced type system known to mankind (yeahhh yeahh Hindley-Milner eat your heart out)

This TypeScript release is largely about performance. Isn't OCaml still at least twice as fast (and maybe even faster for incremental compilation on very large codebases)?

I don't think GP was referring to transpilation speed when they wrote "most advanced type system known to mankind".

The original poster was referring to the golang port of TypeScript which was done almost exclusively for performance reasons. They weren’t just making an unprompted comparison of two type systems.

I was referring to the type system in the formal "specification" sense, not the implementation details of the compiler, such as the compiler's runtime (Go, or, in the past, JavaScript/Node). it's too bad there's no specification for TypeScript the way there is for JavaScript and many other things (most?). in that sense, the performance of the type system is uncoupled from the semantics of the type system itself (as the Go port has thoroughly illustrated!).

I mentioned Hindley-Milner because I am under the belief that the HM system (as in OCaml) is, in the same formal/semantic/specification sense, perhaps more advanced. but, as is often with these things, the rubber meets the road on which one of them has been shown to actually run Doom, lol, to which TypeScript is currently the undisputed king.

They picked Go after meaningfully considering Rust (and others). I don't remember all the reasons for it but it was detailed in the original blog post.

There's the blog post (by Anders Hejlsberg, the author of Turbo Pascal, chief architect of Delphi, currently lead architect of C# and a core developer of TS; I'd say he knows his languages), but they also posted FAQs, here's some good reads:

The blog post: https://devblogs.microsoft.com/typescript/typescript-native-...

Why Go? https://github.com/microsoft/typescript-go/discussions/411

Why a port instead of a rewrite? https://github.com/microsoft/typescript-go/discussions/410

Reply on news.ycombinator.com