I understand that implementing the TypeScript compiler is not the same thing as implementing all Node.js APIs, but still, advertising "no runtime" and then requiring JS runtime (and a full local Rust setup to compile it) for something as basic as an Express web server makes the "no runtime" claim look like a slight exaggeration. I'm not saying that it's bad, it's just that the website is too optimistic.
Edit: as discussed in the thread below, the most likely reason for that is that Express is pure JS with types from @types/express, so the TypeScript compiler bails on it. Reasonable, but still frustrating.
Overall, it seems like every time I decide to try a vibe coded compiler I get this feeling like when you see a plate with fruits on a table but, coming closer, see that they are fake plastic fruits. No, I cannot use it to build a native binary of my project without V8 as easy as shown on the front page. Maybe some other project, yes, but not a real one.
Unrelated: if a project is called Perry, should the icon be a platypus in a hat, you know?
> Perry exposes a faithful subset of Node.js’s stdlib HTTP server modules on top of hyper + rustls + tokio-tungstenite. The whole shape — handler signature, IncomingMessage / ServerResponse properties + methods, TLS opts, ALPN-negotiated HTTP/2, WebSocket upgrade dispatch — works unmodified, so unmodified Node servers (Express / Koa / Polka / hono via @hono/node-server / etc.) compile and run natively[1]
It's pretty standard for "no runtime" to mean nothing on the device you install the compiled target app.
I think iOS development still needs Ruby for Pod installation but no one says Swift apps need a Ruby runtime for example.
Well, I did indeed spend some time playing with it before writing my comment. I first tried to compile the TypeScript project I'm working on, and it happens to be an Express server. After some minor unrelated fixes required (Perry does not understand importing "fs/promises", so I fixed it to import "fs" and then taking .promises) it said it needs JS runtime, and the smallest repro I found was
$ cat index.ts
import * as express from 'express';
const app = express();
which gives
$ perry index.ts
Collecting modules...
JS module: express -> /private/tmp/ex/node_modules/express/index.js
Error: build pulled in `perry-jsruntime` (QuickJS-based eval-equivalent runtime) via the following file(s):
- /private/tmp/ex/node_modules/express/index.js [express]
`perry-jsruntime` is treated as a privileged dependency on par with adding a JIT to the binary — it re-introduces arbitrary runtime code execution and defeats Perry's structural advantage over Node. Refusing to link by default. (#499)
To enable, set `perry.allowJsRuntime: true` in the host package.json, or pass `--enable-js-runtime` on the CLI for a one-off build. (Falls under `--lockdown` deny set when that flag ships — see #496.)
Maybe it's because Express is written in JavaScript with external types from @types/express, that would explain why it might need JS runtime, but it does not make things easier for me.
The website doesn't explain how it works in a lot of detail. I am the author of tsonic [1], a TS compiler that produces binaries via Clr NativeAOT (on Linux/Mac). The hardest parts were numbers (TS has no ints or shorts), Generics, and TS Utility Types. I've been on it for the last 6 months (almost every day); getting to near complete TS compatibility is a very long journey because of its expressiveness.
Add: A request is to explain how it works on the website. I did take a look at https://www.perryts.com/en/internals/ but are those techniques described really sufficient to express TS? Based on my experience, I must say I'm surprised. But the proof is in the pudding, and if it's compiling those examples it must be working somehow!
Hey there, I'm going to check out your project because the comments here have me a little worried that OP's project might have some quality issues.
Two things I found a little confusing from the docs though:
I couldn't easily find a page describing what it can't do yet. I saw that it only works with a "strict, deterministic subset of TypeScript", but is there a page showing what's included and not included in that subset?
Also, what's an "ambient surface" in this context? Is that a compiler term I'm just not familiar with?
> The hardest parts were numbers (TS has no ints or shorts)
The easy way to handle that would be to just treat "number" as 64-bit float, since that is semantically how Javascript defines them. But that can hurt performance.
Another option is to define your own integer types
Fascinating. I've written cross platform (WASM, iOS, Android) libraries with Rust before and had a good time but Rust can be a pain too. Cross-platform Typescript is a really interesting proposition.
That said, the more I think about it the more dubious I am. The site boasts no runtime dependencies but clearly it’s going to need things like a garbage collector, you can’t just magic that requirement away. At a certain point is it just doing what a JS engine’s JIT compilation does… except ahead of time?
Also doesn't inspire confidence that the text on the site is very clearly AI generated and the GitHub log shows an endless stream of AI powered commits. About 15 per hour, every hour? Doesn’t scream stability.
Perry uses NaN-boxing to preserve TypeScript's dynamic type system at runtime, the same approach as JavaScriptCore. The PERF_ROADMAP is honest about the cost: 1.86x behind Zig on image convolution, with 1.24 billion wasted instructions traced specifically to NaN-box unboxing. You cannot get C-level performance without dropping TypeScript semantics, and dropping them means you are no longer compiling TypeScript.
Typescript is a dynamic language. Without changing the language, there is fundamentaly no way to resolve at compile time decisions that can be made only at runtime (ie, they are data driven). Monomorphization helps pin down (some) dynamic types but the fundamental problem remains.
That's a JIT. Yes, you can do all sorts of optimizations in a JIT, because you do it at runtime using runtime information, and always keeping an escape hatch, so the static code bails when invoked with data it was not compiled to handle. This kind of hatch is used here with <any> wrapping.
JIT is a technique to accelerate dynamic languages at runtime to near machine performance while keeping dynamic ergonomics; but it can't transcend the AOT / runtime wall.
You're right. The typed buffer locals optimization keeps TypeScript semantics intact by exploiting the existing Buffer/Uint8Array type annotation to skip the NaN-unbox. It's not dropping types, it's using them. The floor I described applies to any-typed paths where static type info isn't available. For well-typed TypeScript, the roadmap shows the gap closes without semantic changes.
A very interesting project because I always thought TypeScript or at least some subset of it should be natively compiled.
It looks like others had a similar idea too, adding a "sound mode" to TypeScript, such as this project which is converting tsgo to Rust, also with LLMs.
> such as this project which is converting tsgo to Rust
If you'd like to follow, here's my attempt at converting tsgo to typescript (called tsts [1]). Admittedly there's AI involved, but it's a very mechanical job. Going from golang to ts is not a very difficult problem, the other way around would have been way harder. The plan is to then compile tsts to binary via tsonic.
Interesting, seems to be a very roundabout way of doing it. Have you tried compiling the current TypeScript implementation which is still in TypeScript, as tsgo is for TS 7? If so, what were the results?
I'm not against AI usage but the website, documentation, and even the comments the creator (proggeramlug) makes in response to questions are all very clearly AI-generated. Also, as someone else noticed, the pacing of the commits is eerily fast. That combined with the level of functionality makes me dubious how much accountability the creators have over the implementation.
Like you really built a backend that lowers to LLVM, integrated it with a generational gc, wrote a cross-platform reactive runtime, and built support for eleven different targets within like a year? Are you just prompting the model to tack on the next coolest thing or do you understand how these features work?
I worry how many of these kinds of projects will show up now. How do you guarantee stability? If there's a memory corruption error in the GC implementation, who's going to debug it?
Trust of a project long term always was and continues to be of concern when choosing a critical dependency .
The concern basically boils down to how large and serious is the team and what if they abandon the project in few weeks or months .
These were always the risks, many here have been burned by betting years of their career building against promising but what turned out to be weak projects
OP is alluding to the fact that today commit frequency, lines of code or how active the contributors in the issue trackers are no longer good signals to use as proxy.
When the underlying project to yours is few million lines of code written by machines only it is not going to be feasible fork and maintain or in-house it if the maintainers abandon it
To be clear users of a library or a tool aren’t owed anything when it available gratis and fully open source .
However not everyone has access to unlimited tokens to disregard the quality (in terms of history and usage ) or size of the underlying project completely
I think the primary value of a project like this is the demonstration that this is possible and a proof that it does not incur some unknown tradeoff you'll discover after spending resources doing it.
IMO the maintenance story is more or less solved if you can keep AI agents refactoring and improving it in a loop.
> However not everyone has access to unlimited tokens
Apologies. I did not consider this when writing my comment, being spoilt by unlimited 'free' AI.
Free in quotes because, presumably, training agents on AI usage from developers is worth more than the cost of providing free AI.
> IMO the maintenance story is more or less solved if you can keep AI agents refactoring and improving it in a loop.
That’s a weak argument, though, if the future of AI is totally unreliable when it comes to cost and quality. Right now I definitely wouldn’t want to depend on being able to infinitely access AI tools for such an important part of the toolchain.
Aside from that it’s just not attractive to trust a project made by one person.
I have used AI agents extensively for coding and my experience is that it's fine for prototypes, but in large projects like this there is risk that the codebase becomes unmaintainable.
Even C has one, regardless of how tiny it happens to be, or the possibility of freestanding deployment.
In compiler speak, a runtime provides all infrastructure required by the language for program startup, shutdown, infrastructure for the standard library execution.
> Traditional OOP runtimes use vtables for method resolution, adding a layer of indirection on every call. Perry resolves all method calls statically during compilation, turning interface method calls into direct jumps.
What? How is this possible, even with something as simple as:
interface Animal {
speak(): string;
}
class Cat implements Animal {}
class Dog implements Animal {}
function makeSound(animal: Animal) {
return animal.speak();
}
I only grepped "vtable" in the repository so this might be not the relevant piece of code, but from the comment in [1] it seems they fallback to building a vtable if the static resolution is impossible.
switch (animal.type)
case Cat: return cat_speak()
case Dog: return dog_speak()
The generated code has the functions resolved in compile time, there's no function pointer lookup in a table happening. I don't know if this is how this project does it, but this is the commonly used technique when you want to do this.
Perry definitely looks interesting, was just looking at getting one of these to include into my framework.
Would love to see more about it, or see more about the actual compiler docs.
While the UI framework part is neat, I prefer not to force everything into TS. Combining it means UI definitions and semantics get mixed into AST, making the unbundling of them a humongous task in itself.
Exactly the reason I built my own with pretty similar native UI semantics which supports Rust, Go, Kotlin and more (https://hypen.space) - would love to integrate Perry with it to compile TS apps directly into the runtime - but while the idea itself is great, looking at the documentation makes it hard to implement, and a lot of parts seem confusing.
Can I just use the compiler without the rest of the framework?
What is the architecture? What are the limits?
After digging through the documentation, I'm unfortunately just more confused honestly. There are dozens of packages and slop markdown files such as `BUG_STRING_COMPARISON.md` and or `PERRY_UI_IMPLEMENTATION.md` which is an instruction file left for the LLM that just makes me trust the project less.
So while the idea is cool and the performance seems cool, the AI slop presentation would definitely need improvement. Adding a human touch would make it much, much better, as one could actually understand what they are dealing with.
Curious where on spectrum compiling to wasm falls between art project & optimization potential. Should be able to make some nice interfaces between TS-wasm & TS-web
I understand that implementing the TypeScript compiler is not the same thing as implementing all Node.js APIs, but still, advertising "no runtime" and then requiring JS runtime (and a full local Rust setup to compile it) for something as basic as an Express web server makes the "no runtime" claim look like a slight exaggeration. I'm not saying that it's bad, it's just that the website is too optimistic.
Edit: as discussed in the thread below, the most likely reason for that is that Express is pure JS with types from @types/express, so the TypeScript compiler bails on it. Reasonable, but still frustrating.
Overall, it seems like every time I decide to try a vibe coded compiler I get this feeling like when you see a plate with fruits on a table but, coming closer, see that they are fake plastic fruits. No, I cannot use it to build a native binary of my project without V8 as easy as shown on the front page. Maybe some other project, yes, but not a real one.
Unrelated: if a project is called Perry, should the icon be a platypus in a hat, you know?
This seems either wrong or very uncharitable.
> Perry exposes a faithful subset of Node.js’s stdlib HTTP server modules on top of hyper + rustls + tokio-tungstenite. The whole shape — handler signature, IncomingMessage / ServerResponse properties + methods, TLS opts, ALPN-negotiated HTTP/2, WebSocket upgrade dispatch — works unmodified, so unmodified Node servers (Express / Koa / Polka / hono via @hono/node-server / etc.) compile and run natively[1]
It's pretty standard for "no runtime" to mean nothing on the device you install the compiled target app.
I think iOS development still needs Ruby for Pod installation but no one says Swift apps need a Ruby runtime for example.
[1] https://docs.perryts.com/stdlib/http.html
Well, I did indeed spend some time playing with it before writing my comment. I first tried to compile the TypeScript project I'm working on, and it happens to be an Express server. After some minor unrelated fixes required (Perry does not understand importing "fs/promises", so I fixed it to import "fs" and then taking .promises) it said it needs JS runtime, and the smallest repro I found was
which gives Maybe it's because Express is written in JavaScript with external types from @types/express, that would explain why it might need JS runtime, but it does not make things easier for me.Fair, but might have been worth including that in your initial comment because the docs don't mention that at all.
> It's pretty standard for "no runtime" to mean nothing on the device you install the compiled target app.
Only by layman that don't understand compilers.
First of all, congrats.
The website doesn't explain how it works in a lot of detail. I am the author of tsonic [1], a TS compiler that produces binaries via Clr NativeAOT (on Linux/Mac). The hardest parts were numbers (TS has no ints or shorts), Generics, and TS Utility Types. I've been on it for the last 6 months (almost every day); getting to near complete TS compatibility is a very long journey because of its expressiveness.
Add: A request is to explain how it works on the website. I did take a look at https://www.perryts.com/en/internals/ but are those techniques described really sufficient to express TS? Based on my experience, I must say I'm surprised. But the proof is in the pudding, and if it's compiling those examples it must be working somehow!
[1]: https://tsonic.org
Hey there, I'm going to check out your project because the comments here have me a little worried that OP's project might have some quality issues.
Two things I found a little confusing from the docs though:
I couldn't easily find a page describing what it can't do yet. I saw that it only works with a "strict, deterministic subset of TypeScript", but is there a page showing what's included and not included in that subset?
Also, what's an "ambient surface" in this context? Is that a compiler term I'm just not familiar with?
> The hardest parts were numbers (TS has no ints or shorts)
The easy way to handle that would be to just treat "number" as 64-bit float, since that is semantically how Javascript defines them. But that can hurt performance.
Another option is to define your own integer types
> Another option is to define your own integer types
This is what I did. Most int usage is inferred, but if they had to define it explicitly, I make them import { int } from "@tsonic/core/types.js";
Does tsonic have debugging support of any kind and how fast is the compiler?
Fascinating. I've written cross platform (WASM, iOS, Android) libraries with Rust before and had a good time but Rust can be a pain too. Cross-platform Typescript is a really interesting proposition.
That said, the more I think about it the more dubious I am. The site boasts no runtime dependencies but clearly it’s going to need things like a garbage collector, you can’t just magic that requirement away. At a certain point is it just doing what a JS engine’s JIT compilation does… except ahead of time?
Also doesn't inspire confidence that the text on the site is very clearly AI generated and the GitHub log shows an endless stream of AI powered commits. About 15 per hour, every hour? Doesn’t scream stability.
how do one tell when text is ai generated - honest question. What are the tell tale signs?
vibrant colours that don't match
X. Y. SUPER Z. heading
X. Y. SUPER Z. in subheading
excessive purple and gradients
--> arrows
cards, cards, cards, cards
doesn't just X, emdash, it [SUPER Y]
more cards
ridiculous awful contrast in copy that makes things unreadable (grey on black etc)
Not X, not Y, just Z.
The whole site is very jarring to read.
In regards to the site, all of them follow almost the same templates.
Once you see a few, it becomes obvious
if /comprehensive/ then ai
tbf Rust also can spit out pretty big binaries for small programs
Agreed. You can optimize things a fair amount with the Rust compiler, at least.
Calling a couple million lines of ai written Rust "stable software" is a bold statement
vibeware :)
Perry uses NaN-boxing to preserve TypeScript's dynamic type system at runtime, the same approach as JavaScriptCore. The PERF_ROADMAP is honest about the cost: 1.86x behind Zig on image convolution, with 1.24 billion wasted instructions traced specifically to NaN-box unboxing. You cannot get C-level performance without dropping TypeScript semantics, and dropping them means you are no longer compiling TypeScript.
I think you mean you can't get that performance without monomorphization. When you know the types you can...
...wait, I went and looked up that file.
"The Three Optimizations That Would Close the Gap"
You're presenting the data from there in an extremely misleading way! They in no way need to drop any Typescript semantics to go faster.
Typescript is a dynamic language. Without changing the language, there is fundamentaly no way to resolve at compile time decisions that can be made only at runtime (ie, they are data driven). Monomorphization helps pin down (some) dynamic types but the fundamental problem remains.
Why don't JITs preserve previous work across runs of the same code?
If you encounter code with the same hash as last time, load up the previously generated binary and run that... or is that already happening?
Julia?
That's a JIT. Yes, you can do all sorts of optimizations in a JIT, because you do it at runtime using runtime information, and always keeping an escape hatch, so the static code bails when invoked with data it was not compiled to handle. This kind of hatch is used here with <any> wrapping.
JIT is a technique to accelerate dynamic languages at runtime to near machine performance while keeping dynamic ergonomics; but it can't transcend the AOT / runtime wall.
You're right. The typed buffer locals optimization keeps TypeScript semantics intact by exploiting the existing Buffer/Uint8Array type annotation to skip the NaN-unbox. It's not dropping types, it's using them. The floor I described applies to any-typed paths where static type info isn't available. For well-typed TypeScript, the roadmap shows the gap closes without semantic changes.
Put the LLM down and talk to humans as a human.
A very interesting project because I always thought TypeScript or at least some subset of it should be natively compiled.
It looks like others had a similar idea too, adding a "sound mode" to TypeScript, such as this project which is converting tsgo to Rust, also with LLMs.
https://news.ycombinator.com/item?id=48189156#48189573
https://tsz.dev/
It already exists for years, it is part of Microsoft's MakeCode efforts for kids.
https://makecode.com/language
https://www.microsoft.com/en-us/research/publication/static-...
Much saner than a vibe coded compiler.
> such as this project which is converting tsgo to Rust
If you'd like to follow, here's my attempt at converting tsgo to typescript (called tsts [1]). Admittedly there's AI involved, but it's a very mechanical job. Going from golang to ts is not a very difficult problem, the other way around would have been way harder. The plan is to then compile tsts to binary via tsonic.
[1]: https://github.com/tsoniclang/tsts
Interesting, seems to be a very roundabout way of doing it. Have you tried compiling the current TypeScript implementation which is still in TypeScript, as tsgo is for TS 7? If so, what were the results?
Yep. Too dynamic.
Uhm... You're trying to port Typescript's Typescript-to-Go port back to Typescript? Am I missing something?
Yes. :) With the goal of compiling it to native code.
It's mostly a mechanical port. Hoping to do a Show HN maybe in a month.
I'm not against AI usage but the website, documentation, and even the comments the creator (proggeramlug) makes in response to questions are all very clearly AI-generated. Also, as someone else noticed, the pacing of the commits is eerily fast. That combined with the level of functionality makes me dubious how much accountability the creators have over the implementation.
Like you really built a backend that lowers to LLVM, integrated it with a generational gc, wrote a cross-platform reactive runtime, and built support for eleven different targets within like a year? Are you just prompting the model to tack on the next coolest thing or do you understand how these features work?
I worry how many of these kinds of projects will show up now. How do you guarantee stability? If there's a memory corruption error in the GC implementation, who's going to debug it?
where exactly do you see comments by proggeramlug?
This is how software development works now. We have to live with it.
The models are good enough that this works.
You can keep disagreeing for a while, but know that almost all the code in the industry is written like this now.
Trust of a project long term always was and continues to be of concern when choosing a critical dependency .
The concern basically boils down to how large and serious is the team and what if they abandon the project in few weeks or months .
These were always the risks, many here have been burned by betting years of their career building against promising but what turned out to be weak projects
OP is alluding to the fact that today commit frequency, lines of code or how active the contributors in the issue trackers are no longer good signals to use as proxy.
When the underlying project to yours is few million lines of code written by machines only it is not going to be feasible fork and maintain or in-house it if the maintainers abandon it
To be clear users of a library or a tool aren’t owed anything when it available gratis and fully open source .
However not everyone has access to unlimited tokens to disregard the quality (in terms of history and usage ) or size of the underlying project completely
I think the primary value of a project like this is the demonstration that this is possible and a proof that it does not incur some unknown tradeoff you'll discover after spending resources doing it.
IMO the maintenance story is more or less solved if you can keep AI agents refactoring and improving it in a loop.
> However not everyone has access to unlimited tokens
Apologies. I did not consider this when writing my comment, being spoilt by unlimited 'free' AI.
Free in quotes because, presumably, training agents on AI usage from developers is worth more than the cost of providing free AI.
> IMO the maintenance story is more or less solved if you can keep AI agents refactoring and improving it in a loop.
That’s a weak argument, though, if the future of AI is totally unreliable when it comes to cost and quality. Right now I definitely wouldn’t want to depend on being able to infinitely access AI tools for such an important part of the toolchain.
Aside from that it’s just not attractive to trust a project made by one person.
Not at all, I can assert that the Spring code on my current project is classical programming.
In many places AI tools aren't even allowed to touch customer repos.
I have used AI agents extensively for coding and my experience is that it's fine for prototypes, but in large projects like this there is risk that the codebase becomes unmaintainable.
This is a very popular opinion that is sort of obsolete now in my opinion.
It was a valid concern last year. We have seen tremendous progress on this in the last 4-6 months.
Even if your initial prototypes are unmaintainable slop, the state of the art models are fairly good at refactoring and fixing things.
the claim of "no runtime" is a bit dubious... you're telling me that you're statically linking a full, modern UI library into every app?
A runtime is needed for GC, unless you're fundamentally changing Javascript. Even golang has a runtime.
Even C has one, regardless of how tiny it happens to be, or the possibility of freestanding deployment.
In compiler speak, a runtime provides all infrastructure required by the language for program startup, shutdown, infrastructure for the standard library execution.
Its documentation mentions a garbage collector.
https://docs.perryts.com/language/supported-features.html#ga...
They probably are, given that's how Flutter and other Rust GUI frameworks work too.
> 0 ms Startup time
Is that true? It just goes right into the code with no initialization of any other libraries needed?
> Traditional OOP runtimes use vtables for method resolution, adding a layer of indirection on every call. Perry resolves all method calls statically during compilation, turning interface method calls into direct jumps.
What? How is this possible, even with something as simple as:
I only grepped "vtable" in the repository so this might be not the relevant piece of code, but from the comment in [1] it seems they fallback to building a vtable if the static resolution is impossible.
[1]: https://github.com/PerryTS/perry/blob/39d5ba2e9db5adf7f7fc90...
You can just generate the 'vtable' as code :)
The generated code has the functions resolved in compile time, there's no function pointer lookup in a table happening. I don't know if this is how this project does it, but this is the commonly used technique when you want to do this.Hmm yeah good point. I didn't think of it. It might even be cheaper to do this when the list of possible types are closed and few.
I am still inclined to believe AI just made up the documentation though, because this has its own tradeoffs.
> might even be cheaper to do this
Oh yeah, very often. Especially if it's a loop and resolves the same way very often.
Modern CPUs just blaze through code like this, after three decades optimizing for object oriented and dynamic languages.
It's not possible, it reduces to the halting problem, unless you start restricting the Javascript/Typescript you allow.
Otherwise the best you can do is be solid on common special cases. Which is what a modern (non-static) JS/TS runtime is.
https://github.com/PerryTS/perry
The screenshots in the showcase look goofy
Perry definitely looks interesting, was just looking at getting one of these to include into my framework.
Would love to see more about it, or see more about the actual compiler docs.
While the UI framework part is neat, I prefer not to force everything into TS. Combining it means UI definitions and semantics get mixed into AST, making the unbundling of them a humongous task in itself.
Exactly the reason I built my own with pretty similar native UI semantics which supports Rust, Go, Kotlin and more (https://hypen.space) - would love to integrate Perry with it to compile TS apps directly into the runtime - but while the idea itself is great, looking at the documentation makes it hard to implement, and a lot of parts seem confusing.
Can I just use the compiler without the rest of the framework? What is the architecture? What are the limits?
After digging through the documentation, I'm unfortunately just more confused honestly. There are dozens of packages and slop markdown files such as `BUG_STRING_COMPARISON.md` and or `PERRY_UI_IMPLEMENTATION.md` which is an instruction file left for the LLM that just makes me trust the project less.
So while the idea is cool and the performance seems cool, the AI slop presentation would definitely need improvement. Adding a human touch would make it much, much better, as one could actually understand what they are dealing with.
Curious where on spectrum compiling to wasm falls between art project & optimization potential. Should be able to make some nice interfaces between TS-wasm & TS-web
Why go through LLVM at all?
There are a bunch of tools that JIT somewhat optimized assembly.
Yeah, a missed opportunity. But I'm sure such a compiler will eventually arise.