Lua beats MicroPython for embedded devs(embedded.com) |
Lua beats MicroPython for embedded devs(embedded.com) |
The ease of embedding Lua, even with a C++ wrapper, is incredible. With little effort, I now have something I consider “ready”.
Not to mention, it’s a very lightweight VM.
Generally, I use pybindgen to get the basic module and then hack on that by hand. The main problem is most C(++) libraries aren't designed to interoperate with managed memory languages so most of the work is figuring that out. Don't get me wrong, I've tried to work within the binding library (with pybindgen at least) but the amount of work for anything even slightly complicated isn't really worth it.
For a project with a major python API (like blender) you're better off crafting your own python class generator (again, like blended does, and even that has some major issues around object lifetime management). Best would be to design the underlying library/application with python integration in mind but that's not always possible if you want to include other libraries.
I can say I did mess around with using lua as an embedded scripting language within an application years ago and it wasn't too difficult from what I remember. It was only ever a proof-of-concept and didn't go too far so I never ran into the inevitable edge case as one always does with these sorts of things.
I find it difficult to take any writing seriously when it uses phrases like this.
MicroPython can be equally readable, but in practice, many projects end up with blurred layers between system code and scripting. That creates a maintenance burden as projects grow.
Yeah, right. Even if this is the case (I find it hard to belive the author has really seen 'many' sort of professional MicroPython projects), where's the proof the language used was the deciding factor in that. And not the project management for instance. Or simply the dev's architecturing abilities.
While it's true you can't have multiple MicroPython interpreters running concurrently (or at least not easily; it's not that the design makes this impossible, it's just that all in all MicroPython is fairly young and development focus has been put elsewhere), it is possible to embed MicroPython. Not completely out of the box, needs some glue code etc. See for example https://github.com/micropython/micropython/tree/master/ports....
Lua ranks higher than MicroPython in this list
https://www.farginfirmware.com/home/lua-vs-micropython
Github account "SkipKaczinksi" thinks Lua is generally faster
https://hackaday.com/2020/11/14/micropython-on-microcontroll...
Hackaday commenter "Michael Polia" suggests Lua, smaller, faster
https://www.cnx-software.com/2021/11/28/toit-open-source-lan...
Toit scripting language claims to be 30x faster than MicroPython
Toit founder was "responsible for initial development of V8"
Python does have the 'there should be one, preferably only one, way to do it' mantra, but to me it utterly fails at that, and is in fact a bit of a 'kitchen sink' or 'armchair' language.
That is it's strength in some ways, it's easy and approachable, and has more libraries than perhaps any other language, so you can usually get something working fairly quickly.
But it's not so suited to sparse environments. You can't easily turn a plump armchair with automatic footrests and telescoping side-tables into a plywood Eames.
The problem of "easy" is that it implies hidden complexity for its magic. The problem of "simple" is that it requires more work from its users.
You use micrpython when you have lots of horsepower and need something fairly robust on the network.
You use C/C++ if you need precise control over power, memory or CPU. Even though if you're doing network stuff its much harder to do quickly and securely. (THere might be better embedded TLS support now)
Lua is frankly just sparkling C. Sure if someone has create a bunch of libraries for you, then great, if not, you've now go to support lua toolchain, and your own microcontrollers toolchain, and port what ever control lib the manufacturer provides yourself.
Or, as this is a marketing page, pay https://realtimelogic.com/products/xedge/ to do it for you.
It runs on a 2350.
But I'd like to try some other compiled language someday because I'm not a big fan of C++. Any recommendations for something that works well with a Raspberry Pi Pico?
I'm also intrigued by Zig. I haven't used it for anything yet but the language looks fun and I believe platform.io supports it
Raspberry Pis are beefy enough that you could also get away with less systems-y languages. I like Kotlin. By default Kotlin needs a JVM but I think it's usable if you build native executables. However if you want to fiddle with GPIOs you might have to do it by manually setting things on/off in the filesystem (edit, sorry, just read you're using a PICO. Not sure how well Kotlin is supported)
https://micropython.org/resources/code-coverage/
('py' folder: 99.2% line, 88.7% branch coverage)
Testing the port-specific code has been a much more challenging problem.
Which is why in the past year or two there's been a lot of energy put in to HIL testing. There are now a few dozen boards that are automatically tested with an increasingly rich suite of hardware tests. Both the number of boards and tests are increasing rapidly.
It's not perfect but it's getting pretty darn good. If you want to help out please do reach out.
They need to test the whole ecosystem, that is the usual interaction with typical peripherals (I2C, I2S, SPI, Serial, interrupts...) and for every bug fixed there should be a unit test to make sure future changes don't cause regressions.
I had this conversation with them but they take every opportunity to justify/ask for more funding.
They are also prioritizing adding support for new chips but they quickly get eaten alive by chaos, regressions and platform fragmentation.
gdscript is so awesome
Some concerns are valid: Arduino doesn't have as much flexibility for digging really deep into things like pin and memory assignments and what happens when a microcontroller starts up. Also, the quality and documentation of Arduino support can vary from one MCU family to another.
There's a concern about the quality of libraries and code.
It doesn't support hardware debugging.
Granted, the embedded community has good reasons for being conservative, especially for critical applications.
All of these products are still alive today, actively supported and making my customers good money.
Some things come very natural to Lua: Lua <=> C interfacing is a breeze, and while some modern languages are still struggling to figure out how to do proper async, Lua has been able to do this for decades. The language itself is minimal and simple but surprisingly powerful - a few smart constructs like coroutines, closures and metatables allow for a lot of different paradigms.
For new projects at this scale, I would still choose Lua + C/C++ as my stack. Over the last few years I have been visiting other ecosystems to see what I'm missing out on (Elixir, Rust, Nim), and while I learned to love all of those, I found none of them as powerful, low-friction and flexible as Lua.
Most of the time you have less than 1MB of ram on MCU
Every so often I have a need for a small cheap device interoperating with a larger system that I'm developing. Like something that sits on MODBUS and does a simple task when signalled. I've taken the RP2040 and Pico board and spun it into a gizmo that can do whatever I want with Micropython, and it's an order of magnitude cheaper and faster than trying to spin it up in STMCube.
On typing, there's only a few main ones you need to worry about—strings, functions, tables and numbers. I don't think it does weird things like JS where it converts between them without asking. Luau adds some type hinting if it's a big point of concern but I haven't really looked into it much.
I know it’s probably and overreaction, but this was a compete non-starter for me.
https://typescripttolua.github.io/
I’m personally a fan of YueScript which is basically an evolution of MoonScript (but it’s not typed).
LuaJIT has ridiculously easy C interop.
The dev experience for lua is f-ing awful.
The language is small, but not “simple”; it’s stuck in 1985.
The tooling is nearly non-existent and the stuff that does exist is a joke.
The few libraries that exist are awful and barely maintained; the maintained libraries are maintained by neckbeards with a god-complex.
The community is “reimplementing everything by hand is the right way to do it AND ALSO you’re an idiot doing it wrong” toxic. There are a million good reasons why it’s only has a foothold in nginx and Roblox territory.
It’s not a joke to say that it’s f-ing terrible all the way around.
Even "modern" languages often don't have features like first-class functions, closures, proper tail calls, etc that Lua has had for a very long time now. LuaJIT also trades blows with JS almost certainly making it the fastest or second-fastest dynamic language around.
There's a lot to like about the language (aside from array indexes starting at 1), but I think you are right about the ecosystem and probably right about most of community.
There are a LOT of tools with embedded Lua scripting capabilities.
I thought this was common knowledge. Just look at the non-empty list of removed things in every minor release changelog. If you have enough code to maintain something is going to affect you directly or indirectly.
Because it's a large, widely-used project with relatively few core developers. Many other projects have similar numbers of issues, that's not unique to MicroPython! For example, I admire Zephyr's organisation, yet they have double the issues and 5x the PRs.
> They need to test the whole ecosystem...and for every bug fixed there should be a unit test...
Sure, that's a great ideal to strive for. That was - and is - the case for the language and interpreter. It's becoming more common for port-specific code as the tools to allow automated hardware testing are implemented. But it doesn't happen overnight and it's a large surface area.
> I had this conversation with them but they take every opportunity to justify/ask for more funding.
Ah, I presume you're referring to this discussion?
https://github.com/orgs/micropython/discussions/13436
No-one asked for funding. Some noted that such testing infrastructure doesn't come without effort or for free.
BTW I'm not sure if that was you, but that post had a pretty unhelpful, disrespectful tone. If that was you, I suggest you check your tone and get in and help create some tests. We'd appreciate the help!
> They are also prioritizing adding support for new chips...
That's often paid work. Recent paid work has helped fund HIL testing. Further, a careful expansion of the number of ports is necessary to avoid becoming irrelevant.
> ...get eaten alive by chaos, regressions and platform fragmentation
I use MicroPython commercially, daily...and I don't share your experience with chaos or regressions. Sure, sometimes issues occur, particularly if you're on the bleeding edge - but they're pretty well controlled and rapidly fixed. YMMV.
As for platform fragmentation, I struggle to understand your concerns; it has been improving for a long time now as developers have focused on ensuring compatibility between ports. Again, yes, there are some differences - but given how radically different the micros can behave, the consistency between ports is pretty remarkable.
We don't implement unit tests because it is more effort, we implement them because in the long run is less effort and we catch issues before they happen.
Up to you.
However, it turns out that MicroPython has a simple and efficient GC - and once you eliminate code that gratuitously fragments memory it behaves quite predictably. We've tested devices running realistic scenarios for months without failure.
Some of our products use MicroPython though we also use a whole host of other technologies. Some of our devices are proof-of-concept (often designed to progress a theory) but we also deliver up to Class B solutions.
https://benchmarksgame-team.pages.debian.net/benchmarksgame/...
makes it look fairly slow, competing with Ruby, rather then JS.
https://programming-language-benchmarks.vercel.app/lua-vs-ja...
LuaJIT used to have a performance comparison page, but it was removed a couple years ago apparently. It shows that LuaJIT is an average of around 16x faster than standard Lua (geomean). Very impressive considering it was written by just one guy for years (Mike Pall) and even more impressive when you consider that the JIT is only around 0.5mb (compare that with v8 which comes in at somewhere around 28mb IIRC).
https://web.archive.org/web/20230605114058/http://luajit.org...
You have no idea.
> He removed LuaJIT
LuaJIT supports Lua 5.1
The benchmarks game shows Lua 5.4.7
Assuming your flash allows XIP (execute in place) so all that memory is available for your lua interpreter data, you should at least be able to run some code, but don't expect to run any heavy full applications on that. I don't know Berry but it sounds like a better fit for the scale of your device.
But sure, why not give it a try: Lua is usually easy to port to whatever platform, so just spin it up and see how it works for you!
The experience I had there might be your best bet for something productive. That board came with a 'limited C-like compiler' (took a mostly complete subset of C syntax and transcribed it to ASM).
You'll probably be doing a lot of things like executing in place from ROM, and strictly managing stack and scratch pad use.
The 64MB of RAM and 8MB (I assume that's 64Mbit) of ROM allow for highly liberating things like compressed executable code copied to faster RAM, modify in place code, and enough spare RAM otherwise to use scripting languages and large buffers for work as desired.
If you compare this with what MicroPython uses, its requirements are well over an order of magnitude larger.
Viper Bytecode emitter.
From what I understand, LuaJIT was removed before Lua 5.2 was released, so that wasn't the reason.
https://luajit.org/extensions.html
LuaJIT supports most of the features of Lua versions after 5.1 with the major missing feature being 64-bit integers, but like modern JS JITs, it actually uses 31/32-bit ints internally most of the time. Even in Lua 5.4 code, you are using implicit rather than explicit ints 99% of the time.
I haven't run the code to see, but I'm willing to bet that you can copy all the current benchmark code into LuaJIT and it'll run just fine.
> You have no idea.
I know with certainty that deoptimizations were applied to at least some scripts. Here's three examples for Common Lisp, StandardML, and Haskell over some time.
https://zerf.gitlab.io/ComputerLanguageBenchmarksGame2018Arc...
https://github.com/lemire/ComputerLanguageBenchmark/blob/fbe...
https://hackage.haskell.org/package/ajhc-0.8.0.4/src/example...
Here's a C example from Mike Pall (presumably the same guy who created LuaJIT) that also got the deopt treatment by Isaac Gouy.
https://github.com/lemire/ComputerLanguageBenchmark/blob/fbe...
It's not a question of if this happens -- only if it affects Lua (I've never checked).
On that page, what words do you think support your claim "was removed before Lua 5.2 was released".
> I know with certainty
I made those code changes. I wrote de-optimized as a joke.