Building a Minimalistic Virtual Machine | Hacker News Reader

Building a Minimalistic Virtual Machine

pointersgonewild.com

151 pointsingve61 comments

This is awesome.

Thank you for sharing.

My ancient ruby code and nodejs code all broke because I didn't pin dependencies. As a result I've got software that is unrunnable.

More software shall be unrunnable as time goes on, I don't know many trends that prevent software from being unbuildable and unrunnable due to change except maybe repeatable builds and hermetic builds.

Given platform toolchains complexity and libc versions and complexity of static Vs dynamic linking, I suspect preserving software is very difficult.

It seems doing + - ×÷ on numbers is not the difficult part of computers but arranging information into the right places in order to do it.

Logistics and package management are difficult to get right.

I think Java got something right. Bytecode is longlasting. Can rewrite the JVM for new platforms and architectures.

I am writing my own language and it is implemented as an assembly interpreter and a compiler for that interpreter. This lets me get development speedier.

What the hard thing I think is more interesting than bytecode or virtual machines is INTEROP.

The Amd64 SysV Binary interface of registers for C calling interface and the System call Interface of Linux.

Mozilla abandoned XPCOM extensions, part of the reason was performance of the interop between JavaScript and C++.

If I could run a virtual machine and interop with modern code that would mean the software was useable for longer.

> My ancient ruby code and nodejs code all broke because I didn't pin dependencies. As a result I've got software that is unrunnable.

You can simply remove the ^ symbol from your versions listed in package.json and it will use exactly that version you originally added.

Besides the issue of whether you pin or don't pin your dependencies, the problem is that node packages can depend on external native code. You can have several more layers of dependencies in there. If, for any reason, those native packages won't install/run on your machine, your dependencies can still break under you, even if you pin them. Python and Ruby have the same vulnerability when it comes to dependencies breaking.

The JVM is an interesting use case.

Indeed, at a compiled level, byte code is mostly long lasting.

I haven’t fired up a 10 year jar file recently but it would not surprise me if it Just Worked.

The success of that is twofold. First is simply that whatever changes are being made to the JVM, they’re mostly forward looking and don’t deprecate running code.

The other is that the conventional packaging mechanic is, essentially, a static binary. A “fat jar” is the term of art, with all of the dependencies bundled in.

But there’s still potential problems. They’ve been removing large subsystems from the JDK as of late. XML, web services, Java FX are poster examples. So legacy binaries depending on those will fail outright.

These can be added back to the Java runtime, but still “one more thing”.

Of course from the source code side, Java suffers dependency hell and code rot along with the best of them. Network based dependencies up and vanish. Long standing projects may not publish 10 year old jar files any more.

Also, Java has had other clods dropped into its churn. Oracle shutting down the java.net website was a huge sudden black hole in the community consciousness of Java. Overnight thousands of articles, blog posts, forum entries, and other artifacts vanished like Keyser Soze. Leaving behind a debris field of dead links across the internet.

So, to be fair, the JVM is a boon. I really like Java, and it’s still going strong. The VM architecture and the comprehensive nature of the Java runtime has made the moving of running code across systems much easier. As someone with an enterprise Java background, used to deploying WAR and EAR files, I got to mostly avoid the entire Docker and other such family of infrastructure. Install a JDK, install an App Server, all fairly trivial to isolate, and the system is ready to go.

But in the large, it takes more than a VM to get things accomplished. There’s always an eco-system at play.

And one primary characteristic of eco-systems is they evolve. Time marches on, and waits for no one.

Thank you for your thoughtful and interesting comment.

I too like Java a lot and think it's a great technology.

I can see how a runtime for a compiler (such as the JVM) doesn't need to change much over time: if the compiler produces the right code in 2000, then the code is probably still right in 2020, it just could use more features of the ISA that were introduced since then.

On the other hand, the design of platform, library code and framework code is an ecosystem that is desperate to transform over time as better approaches to solving technical and business problems are found.

That’s one of the many advantages of building for the web. It won’t ever be unsupported (for some reasonable definition of never).

You get text rendering, canvas, audio, webgl… it’s a pretty wide platform.

Except that's just not true. It's already broken/incompatible in many places across browsers. You may not notice if you're just doing basic HTML/CSS, but if you do anything slightly more dynamic, you're going to notice. An ever-expanding set of complex APIs also makes it more and more likely that bugs will go unseen and unfixed. See two examples I detailed in the blog post.

Sorry, I was reading the comments before reading your article.

Getting around pointer events inconsistencies is a lot easier than building your own cross-platform VM, of course. But the project looks awesome and seems like a great initiative.

I imagine there will also be differences in the way macOS, Linux and Windows handle graphics, IO, audio, etc, that will eventually leak to UVM, it's just the nature of the challenge.

> Getting around pointer events inconsistencies is a lot easier than building your own cross-platform VM

For sure, and I did. I wrote some code that makes an invisible fullscreen div appear just at the right time to prevent pointer events being triggered when they shouldn't be #cleancode. It's just frustrating that things like that need to be done, and how often they might need to be done.

> I imagine there will also be differences in the way macOS, Linux and Windows handle graphics, IO, audio, etc, that will eventually leak to UVM, it's just the nature of the challenge.

At the moment you can create a window with one function call, and you have another function call to copy one frame's worth of pixels into the window. The pixel format is in BGRA byte order, 32-bits per pixel, and that's the only option. I'm going with really basic, low-level APIs like that because they're harder to get wrong.

Audio is going to be equally simple. There could be cross-platform differences in things like the amount of latency to write audio, but I'll do my best to make the APIs extremely portable and hard to get wrong.

Hi! I can’t help but see a big similarity to the JVM, both in terms of design goals and the semantics of the byte codes - only “not having dynamic linking” being an exception.

Could you expand on what you don’t find sufficient in JVM byte code, when it is arguably one of the best platform for backwards compatibility, has easy support for bringing up a canvas and start painting, etc.

For one the JVM is a huge piece of software. Large enough that only a large corporation could realistically reimplemented or maintain it. It also exposes many APIs with a large surface area. Then there's the issue of Oracle and how you feel about them as a company.

UVM has obviously nowhere near the ecosystem, but you can draw pixels to a frame buffer with two function calls, and your UI will be guaranteed to look the same everywhere.

JVM is a specification which can be implemented (and has been plenty of times, completely independently of each other) by a single developer in like half a years tops. It is a simple stack-machine with 100+ basic instructions, a simple exception mechanism and a heap. A GC is not even necessary if we are talking about minimalistic approaches, but a basic tracing GC is also not hard.

It is only a (huge) plus that it can be run everywhere with top of the line performance thanks to OpenJDK (which is mostly developed by Oracle, but is big enough that an insane amount of companies critically depend on it and several one could single-handedly finance the future of the platform if anything were to happen, which won’t because it has the same license as Linux).

Implementing a good GC is incredibly hard. The JVM may have just "100+ basic instructions", but it also has classes, objects, arrays, and a whole set of APIs it provides. Your JVM is kind of useless if it doesn't ship with all of the user interface primitives (and other APIs/classes) people expect, for instance. Otherwise what you have is not what people expect to find in a JVM.

I'm also under the impression that building a good JIT for a JVM would be a massive undertaking. It literally took over a decade for the Sun/Oracle JVM's JIT to become mature enough.

I've designed UVM in a way that I believe it will be possible to design a good JIT with relatively little effort

> Implementing a good GC is incredibly hard

If you are interpreting instructions a simple one will be more than enough. Classes are its primitives, you just create a basic runtime representation for them with name, superclass, implemented interfaces and the methods’ bytecodes. Then an object can be as simple as a header containing a pointer to the class’s representation and then a listing of its fields, which can all be 64bits. And an array can have the exact same representation as well, with the first element being its size.

All APIs are just classes with some methods that may be “native”, which are linked to a native implementation (basically just a function pointer). This is how file access and the like becomes possible.

> Otherwise what you have is not what people expect to find in a JVM

You can just say that it is a partial implementation that doesn’t support the whole of the Java standard lib. It’s not unheard of (e.g. Java ME is a subset that runs on every SIM and bank card).

> I'm also under the impression that building a good JIT for a JVM would be a massive undertaking

Well, then just go with an okayish JIT. With all due respect, you ain’t going to beat the JVM with your UVM’s JIT compiler, not even close. Why do you think creating a similarly good JIT compiler to a very similar design would be any easier in case of UVM?

But don’t get me wrong, I just ask these questions because I dislike NIH syndrome and I believe there are useful lessons to be learned from the past. But you should be able to answer why the thing you do is any different (unless it is for learning). And the JVM spec is a surprisingly good read, and you can sure take great ideas from that.

> With all due respect, you ain’t going to beat the JVM with your UVM’s JIT compiler, not even close.

I think I may be able to get very close to native performance. I don't want to sound like an asshole by appealing to authority, but you aren't talking to a teenager writing an interpreter from their parent's basement. I have 21 years of programming experience, a PhD in compiler design and multiple published papers. I have some idea what I'm talking about.

> Why do you think creating a similarly good JIT compiler to a very similar design would be any easier in case of UVM?

The design is superficially similar to the JVM but it's also quite different. UVM's bytecode is untyped. It maps fairly directly to the x86-64 and ARMv8 instruction sets. If you want an idea of how a simple JIT compiler for a bytecode like that can perform, you should look at the performance of Apple's Rosetta. But, I actually think I can build something that yields better performance than that :)

I eagerly await your results, and didn’t want to sound condescending at all, sorry if it came across like that. I was just genuinely interested in a - to me - more understandable difference.

Also, what does “native performance” even mean here? Only removing the interpreter overhead?

Thanks for clarifying. Tone is sometimes ambiguous via text.

At the moment I'm in no rush to actually write the JIT compiler because I think it's faster to iterate with an interpreter. I want to flesh out the VM and its APIs, test the hell out of everything and develop the system a bit more first.

The interpreter runs at something ~400 million instructions per second on my laptop, which is probably close to the performance of an old school Pentium 2 chip, so it's actually fast enough to run a lot of non-trivial software. With even a really basic JIT I should be able to hit 10x that throughput. I've benchmarked code out of GCC and it runs about 27 times faster (on a microbenchmark).

You say this stuff is incredibly hard. Kaba says it's impossible and don't try. I say it's easy. I whipped up a JIT to make my virtual machine go 50x faster. I never expected Blinkenlights to go faster than Bochs. Now all the sudden it's outperforming Qemu for many of my use cases. People are doing stuff with it I never expected, like booting the Linux Kernel and running Alpine Linux on Cygwin. Garbage Collection is easy too. I wrote a GC using the NSA POPCNT instruction for an experimental LISP dialect I wrote last Winter called Plinko. It ran faster than any other LISP interpreter I've seen, as measured by the GC-intensive binary trees benchmark game. The only thing faster was SBCL with JIT which was only a hair faster than Plinko using just an interpreter. NIH is awesome because the truth is, when you're focused on your own needs, outperforming the big official things is like shooting fish in a barrel. Technologies like the JVM aren't great because they're better. They're great because they've carefully crafted the long tail of edge cases and compromises that enables it to be good enough for the largest group of people. Generalized software is at a huge disadvantage because bloat fills caches and it can't use special case algorithms. For example, people publish papers all the time bragging about how they beat the performance of the C++ STL at some given thing and that impresses the people who never tried, but it honestly isn't that hard if you consider the burdens that the STL is required to carry.

It's not close to as easy as you represent. To start with, there are 204 instructions, some of them, far more complex than what you term "basic," such as invokedynamic. The exception mechanism is also far from "simple," -- it's simple conceptually but extremely difficult to get exactly right when it involves finally clauses both in the exception handler and the original excepting code. There are many subtleties that can lead to completely wrong results if not designed very, very carefully. It's far from "simple."

Out of those 204 plenty are the exact same functionality for different types though.

Sure, there is invokedynamic/static/virtual that is a bit more complicated (they basically do runtime linking at first run), but I have implemented them and it is not harder than other pieces of a runtime.

Every method has an exception handler description, which is basically a series of instruction address ranges — if the thrown exception came from there, it jumps to the handler specified by the first match. If not, it propagates up. “Finally” clause is syntactic sugar only.

Sure, these are hard to get right, but that is inherent in the domain to a degree. You need many many “integration” tests - I wrote a test runner that runs the same program with OpenJDK and my implementation and compared their outputs.

The folks at the Jacobin JVM [0] project (a JVM written in Go) are working on the issue of size and the ability to have a fully functional JVM maintained by a small group of developers. Right now, per the latest post [1], they can run simple classes and expect to complete the interpreter in the next few months.

[0] jacobin.org [1] http://binstock.blogspot.com/2023/02/jacobin-jvm-at-18-month...

> or that I should base my system on an existing processor architecture and work on something like Justine Tunney’s blink instead.

If she did then she'd be very welcome as a Blink developer.

You seem to be doing just fine without me :)

Blink is a very impressive project. Mad props.

I don't think it's a bad idea, but having looked into this sort of thing pretty deeply, I do think it's a "now you have two problems" idea when started by centering an ideal VM.

Foundational software is very slippery stuff - it's analogous to having a yeast culture. Before the software was there we had machine language and the specific hardware, but gradually we built a stack of compiler tech, common protocols, etc. And from there we went and started targeting the end of further abstracting it, making it perfectly portable and so forth, but that's like trying to engineer a perfect yeast: while some might get a better benchmark on some metric, the overall metric is a pass/fail: "Can I bake bread with this?"

Of course, a lot of people give up along the way and buy something off the shelf, because they want bread now. And that's fine - they're eating off it, and the world keeps turning.

But if you do try, you can "do more with less" by carefully limiting the kinds of computation and I/O you're doing and describing that as a protocol, then developing software around the protocol. At that point, the software's specification has an insulation barrier from its dependencies, and you don't necessarily care about the whole of the dependency because you've selected a subset whose specific features and codepaths can be vetted in more depth.

And the odd thing about this is that it's the kind of thing that, at its outset, doesn't lead towards complexity, because you intentionally started with a system that does less than its dependent parts: but once you have that, you can knock out and replace the dependencies to "do one thing well," ship of Theseus style. At that point it can become larger again, and evolve. And that's roughly how we ended up with the giant soup of stuff that is "Web tech". At every point along the way, it did bake bread. But when it was first specified, it hardly did anything at all: it bundled "download remote files, present files as documents".

Nice idea! Some of the goals remind me of CHIP-8 [0] ("the hello world of emulators" [1]) but presumably not constrained to 8 bits and with a bit more than just the ISA and graphics support?

[0] http://www.cs.columbia.edu/~sedwards/classes/2016/4840-sprin...

[1] https://www.reddit.com/r/EmuDev/comments/9ezrvt/easiest_emul...

I have not ever studied POSIX academically (curricularly) but:

> a small set of minimalistic APIs that remain stable over time

but from what I've gathered (over years, by 'osmosis'). Isn't this the essential idea behind POSIX?

Yes, but POSIX doesn't provide APIs for graphics or audio, for example, even though these things are necessary to build a lot of end-user software.

See also Turing Drawings http://maximecb.github.io/Turing-Drawings/ (One of my favorite things on the Internet!)

Very interesting for embedded or kiosk. Or in general, if the VM is run as a server inside a host: encryption in a box, instead of a library. Curious about what kind of multitask will be added.

I could actually use some feedback when it comes to the design of the parallelism model for UVM. I have a few ideas but it's not my area of expertise, so I would welcome feedback and suggestions.

I have experience as a user, implementation is way over my head. Depending on the minimum host supported, you could only use lightweight threads or write a scheduler. IIRC, JVM started using its own thread implementation and later used the one provided by the host OS, when available.

Best of luck!

At the moment I have an event-driven system where you can set up callbacks and timers. It's not green threads but it makes it easy to have multiple different update events running at different rates, for example: https://github.com/maximecb/uvm/blob/main/ncc/examples/attac...

For parallelism I'd expect something like Cilk (or Rayon, which was inspired by it), where there's some syscalls for spawning and then joining on tasks, but the VM handles starting and managing the thread pool.

For concurrency it's a lot harder to make a good interface

> For concurrency it's a lot harder to make a good interface

Can you elaborate?

I wonder how this compares to LLVM. That project had building a directly interpretable bytecode and a virtual machine as an initial goal.

LLVM is more heavyweight. Has a lot of analysis and optimization passes for static compilation. UVM is currently very lightweight, will be JIT compiled. Crucially UVM will provide graphics, audio and networking primitives.

That's what LLVM is today, but IIRC their goals were the same. It might be worth it to look at what made it turn away from the minimal idea and into the heavyweight that it is now.

If I had to guess, I would say it probably comes down to wanting to go with static, ahead of time compilation rather than JIT. There are real and valid advantages to AOT compilation. There's downsides with the LLVM approach too. For example, LLVM makes it explicit that they provide zero guarantees when it comes to the stability of their bitcode format. That somewhat limits what you can do with it.

LLVM also has a strong emphasis on performance and interoperability with existing code and is willing to introduce arch- and platform-specific features for that purpose.

This is fantastic! Just what the doctor ordered!

I think this is a dead end. The world doesn't need beautiful soultions, it needs ugly solutions that work tomorrow.

i use qemu for that. bypass all speed and virtualization improvements via flags and you have a rather isolated system

Reminds me of the Dis VM and Inferno OS

Arguably more interesting is UXN: https://100r.co/site/uxn.html

A small personal computing stack, with a plethora of examples ready to go. Built by two they/them hackers who live on a boat and basically have bootstrapped everything about their vessel, their computing, their engineering, etc. It's very, VERY old-school hacker-y.

They would've made excellent phreaks back in the good ol' days.

I was about to mention uxn. It's almost as if people with enough motivation to build and use this are also people who will never agree to use and contribute to one another's work. Not to say any of this work is a waste of course

Creator of Uxn here, Maxime is a dear friend of mine and we've jammed extensively on the design of UVM together.

welcome to the world of Forth/Lisp

It is somewhat sadly ironic that to preserve software one has to actually create software that other people care about preserving. And, as we've seen with emulators, if you have software people care about then there are no roadblocks that will stand in the way of someone figuring out how to get your ancient code to run. I take the whole 100R thing as more an interesting art project than a serious goal. More of a "what if".

With that said, I feel most devs are the same. How often do you come across a lead or senior dev that becomes passive aggressive over people modifying "their" code? Quite often, in my experience. Any code that I didn't write sucks and any code I have to touch that I didn't write is the worst code ever written. Code that I didn't write is old obsolete dog turds and code I wrote is "modern". Every dev slaps "modern" on their open source project today.

Author here. The creator of UXN is a friend of mine and we chat semi-regularly about our VMs.

I have a lot of respect for uxn and credit it as an inspiration, but the goals of each project are different. UXN is a 16-bit system with 64KB of RAM accessible. It will also probably always remain interpreted. These design restrictions are seen as tools to foster creativity.

UVM is a 32/64-bit VM. It's currently interpreted, but I've designed the instruction set with JIT compilation in mind. I have a PhD in compiler design and I'm fairly confident that I can make a fast JIT for UVM in a relatively short amount of time, when I feel the design is mature/stable enough.

At the moment, UVM is relatively immature, but I want it to be a small/minimalistic VM that you can still build "real" or modern software in that takes good advantage of the capabilities and performance of your machine.

Another difference is that IMO, UVM is more approachable. UXN's assembly language is fairly esoteric IMO. It doesn't look like any other assembly language I've ever seen. That doesn't make it bad, but it does potentially make it harder to learn and harder to leverage an existing base of programming skills. UVM's assembly is designed to not be surprising if you've ever programmed in assembly and know the basic ideas about how a stack machine works. I also have a WIP C compiler that's already usable to write simple programs. See my little snake game for a fun toy example: https://github.com/maximecb/uvm/blob/main/ncc/examples/snake...

Assembly syntax example: https://github.com/maximecb/uvm/blob/main/vm/examples/factor...

I'll point to the fact that there is almost no boilerplate necessary to start drawing some pixels on a 2D canvas, which IMO makes it a fun platform to develop for. Like I said, it's immature, but I'll iron out all the bugs I can find and keep making it better.

Awesome project! I've been on the lookout for such projects ever since discovering uxn, I'll definitely have a look and keep an eye on uvm.

>The creator of UXN is a friend of mine and we chat semi-regularly about our VMs.

Does the discussion happen in a public place? If yes I'd be extremely happy to join in since I also got started with making my own system around a month ago, and it feels a bit lonely going on such an endeavor at times.

It's extremely early and I haven't really shared it anywhere yet, but I feel there is already the possibilty to play around with the custom editor I made, try to make little graphical programs etc.. If you manage to build it that is (I develop mostly on OpenBSD and also try to make it build under Ubuntu with gcc sometimes).

The source is hosted here for now: https://git.blazebone.com/pochi/

The README (in the about tab) should give a rough explanation of what it is, I also have a bit of documentation already.

>Another difference is that IMO, UVM is more approachable.

Very interesting choice, I did away with such assumptions and ran the other way, my system might feel quite alien/esoteric since I went for something that draws a lot of inspiration from Chuck Moore's work with ColorForth as well as his F18 chip.

> Does the discussion happen in a public place? If yes I'd be extremely happy to join in since I also got started with making my own system around a month ago, and it feels a bit lonely going on such an endeavor at times.

I'm happy to discuss anything in the GitHub discussions for UVM: https://github.com/maximecb/uvm/discussions

> Very interesting choice, I did away with such assumptions and ran the other way, my system might feel quite alien/esoteric since I went for something that draws a lot of inspiration from Chuck Moore's work with ColorForth as well as his F18 chip.

If you're building a system for fun, or to explore new ideas, then it seems fine to make it as esoteric as you want. However, in my experience, making esoteric choices when designing programming languages for instance, can really alienate potential users. Especially if you could have obviously gone with some more traditional and familiar choices but you went with something more esoteric that doesn't have any clear value added.

IMO it's a bit like when it comes to terminology. If there's a commonly accepted way to refer to something, use it. Don't make up your own nomenclature, you'll just create extra confusion for no reason.

Could you explain what you did to build in mind for JIT?

I feel WASM has the opportunity to create a truly audiovisual API for interacting with computers.

I go into some of the design decisions I made to make JIT optimizations easier here: https://github.com/maximecb/uvm/blob/main/doc/design.md

You mention parallel computation and being open for discussion.

My favourite area of computing is parallel computing and multithreading.

My toy multithreaded interpreter in Java can communicate integers between threads with message passing. I never got around to communicating complicated objects because I'm not sure how to solve the garbage collection problem with compound data structures/object graphs AND sending objects between threads. I'm currently relying on Java garbage collection at this time but I have played with a garbage collector written by Matthew Plant (http://maplant.com/gc.html), so if I were to implement my language in C I could also be inspired by Pony's reference capabilities.

Since my interpreter is a simple imaginary assembly interpreter I have instructions for "sendcode" "receivecode" which tell a thread to do a remote jump. There is also a "send" and "receive" instruction for sending data between threads in a thread safe manner. This uses actor style mailboxes behind the scenes.

I was thinking something like actors, or independent processes sending messages would be nice. Just because it's very safe and predictable. Less error-prone than threads.

The thing that kind of gets me is it seems difficult to have safe shared memory with actors? You ideally want to be able to share memory if you want things to be efficient, but if you have shared memory, then you get into issues with atomic writes and things being observed in different orders, etc.

I assume you would be communicating pointers with actor mailboxes, so the only copy is a pointer. I believe Erlang copies data itself into other actor's heaps, that simplifies garbage collection since GC can be done per process and there is only one owning reference to a processes' data.

You might find Pony's ORCA interesting which is how they implement garbage collection between actors.

https://www.imperial.ac.uk/media/imperial-college/faculty-of...

It might be worth linking the instruction set in the sources (there doesn't seem to be a doc yet?) - it gives a fairly decent picture of what kind of VM we're talking about:

https://github.com/maximecb/uvm/blob/main/vm/src/vm.rs

One thing I didn't quite grok there. It seems that the operand stack is also used directly for locals and hence is indexable, with what looks like an implicit frame base register? But the stack is not addressable. Is the language runtime expected to manage a separate in-memory stack for addressable locals and dynamically allocated arrays, like wasm?

the jvm already exists and you can run ancient compiled class files with it.

The level of cynicism on HN is sometimes really depressing.

> the jvm already exists and you can run [some] ancient compiled class files with it [but many won't work correctly].

FTFY.

I’ve been toying with the idea of taking an ancient bootstrapping JVM implementation, forget which now dead project it was written for, dusting it off and doing something similar to your project. None of these new millennium fancified features — compute like it’s 1996.

Though I’m mostly interested in compiler tech, as an amateur, and don’t really want to delve too deeply into VM theory.

Time is the main factor holding me back though…sigh.

If you want to compute like it's 1986, I think it would be fun to build a BASIC interpreter for UVM. Bonus points if you do your own text rendering on a blue background and you add simple primitives for 2D graphics.

https://github.com/maximecb/uvm/issues/7

ok, but not everything wants the baggage that comes with the JVM.

Reply on news.ycombinator.com