Parallel Perl – Autoparallelizing interpreter with JIT
perl.petamem.comWHOA. Talk about burying the lede... Look from the beginning of the slide show, he made a super cool geothermal project! Look at the size of this hole!! https://perl.petamem.com/gpw2026/perl-mit-ai-gpw2026.html#/1... His cad drawings are great too!
Basically he wanted home automation in Perl to control his geothermal/solar house, and ended up reimplementing Perl with AI. That's some yak shaving...
Standing in that hole without shoring... no thanks. Impressive project nonetheless though
> and ended up reimplementing Perl with AI.
Man, I would have just learned Ruby.....
The guy here.
Project page is: https://perl.petamem.com/
Deep-linking into a reveal slideshow from the presentation - which is meant to be navigated by keyboard arrows ONLY - is suboptimal.
Yes, standing in such a hole is normally not recommended without shoring - safety first - but you do not know the soil specifics. It's like concrete, the excavator digging that hole stood on its very edge after 2 days of rain no problem at all. Disadvantage of such soil is that the percolation rate goes against zero.
Yes, doing a "Perl Interpreter" (it's way more than that) even with the help of the most advanced AI on the planet is PITA. The coding agents do fake, lie or are way out of their depth, but when you are used to limited AI since the 90ies - like me - you know how to handle it. Good News is that in 2 years you will probably tell your AI to create you - just for the laughs - a Perl interpreter AND smarthome system before you go to bed. You will have them ready for breakfast.
As for the maturity of the project, it's really too soon. I thought the German Perl Workshop would be in May, but mixed that up with last years' date, so I presented what I had. In about two months this should be nice(r).
And one final remark: Everyone knows Torvalds for the Linux kernel. Most don't know or ignore he did git too. Here, I presented two things: WHIP and pperl. WHIP being a smarthome solution way above and beyond what is available on the market today, but that seems to somehow evade peoples minds when they see the slides.
It's a very interesting project (even if I always avoided Perl and 'officially don't care'). And so it sucks you got a mediocre response because dum slideshow UI issues. Maybe write up a blog post and try again later (just make sure its not too chatgpt-ish).
I'm interested, but can't navigate the website. The down-arrow in the lower-right is unclickable, maybe covered by some semi-transparent chrome of my browser, not sure. And no idea why there need to be 4 directional arrows.
That's Reveal.js / Slides.com format. It became very popular in 2010s. The idea behind the 2-d navigation is that you can use left-to-right to move between chapters, and move down to dive into a specific chapter. This allows you to skip chapters due to time constraints. Or hide gnarly details about something so that these specific slides do not break the flow of presentation but still having them available for the audience online. Or, having slides announcing demos, but if demos do not work the down slide would have a video demonstrating how the demo is supposed to work. Many possibilities like this. Also the slides are produces using Markdown, so the format was appealing to many authors.
However, doing chapters well turned out to be tricky. Ideally you want them to be of similar size and have 3 to 7 of them in the talk, but many presentations aren't structured like this. The rise of Slideshare and SpeakerDeck for sharing slides in mid 2010s caused this 2-d navigation to go out of favor: those services only support linear static slides. This is also a reason why people use fewer animations in slides nowadays and why tools like Prezi didn't catch on (that was another presentation tool with non-standard navigation that went out of favor very quickly).
Many people still use Reveal.js to make their slides but they stick to left-to-right nav only.
I have the same problem with the mouse (little page marker overlay covers the down arrow).
But using keyboard arrow keys work for me.
Going to the link and just hitting the spacebar worked for me. Next slide, and so on. Firefox/Linux.
Homepage: https://perl.petamem.com
In case HN shows its user hostility again by cutting off the URI fragment, the intended deep-link was presentation slide #/4/1/1
Ugh, deep links should be part of the path, and anchor should be where on the page to scroll. Very annoying slide software. If the content weren't so good I simply wouldn't bother.
HTML+JavaScript-based statically hostable apps (eg. presentations) can't use paths as deep links, since there's no standard for simple static hosting or URL rewriting (even 30 years later). Oh well.
You should be able to use the query part of the URL (after ?). You can get at it with Javascript, but it doesn't influence which static HTML page is served.
I'm not sure why people hate using the query string so much.
They absolutely can generate the file tree so that each slide has its own url.
They also could use the query part on the url rather than anchor.
Lastly statically hosted doesn't mean no URL rewriting, they could again catch links to parts easily.
The poor UX of these tools is just a lack of will, not a technical limitation.
Then again hacker news should probably not blanket delete the hash in URLs either.
This looks like a huge project, even with AI help... I have a sweet spot for perl but I'm honestly not sure if the current community has the bandwidth and interest to sustain an alternative implementation. At the very least it should be ported to MacOS too. Breaking with XS is a bold decision. Best of luck though!!
I'm kind of curious how it compares to some of the current concurrency solutions for Perl because there is a bunch of them now:
"Auto-Parallelization - Automatic parallel map, grep, for, while loops via Rayon work-stealing"
Given any kind of "for" loop, how can it know that there is no synchronization required ? That no mutual exclusion is required ? No concurrent access of some kind ? Offloading some work to another process/thread is expensive, too
If the inner body of the loop is a pure-function, then that's easy (except for the performance part, which may require heuristics or something). But if the body is not pure .. ? I cannot see how this can work reliably with any random code
The slides got stuck at https://perl.petamem.com/gpw2026/perl-mit-ai-gpw2026.html#/6. The right arrow disappeared. The down arrow was flashing, but did not respond to any clicks. I tried different browsers on my mac. None worked.
Keyboard arrows worked.
The down arrow doesn't respond because of the overlay page number. Only when clicking a little bit left of the overlay, it will work.
I can't help but giggle at the fact that AI written project doesn't seem to get its home page right.
It's a kind of crappy slide deck, not a proper home page. Even worse, the link drops you into the middle of the deck. (TBF, it wouldn't be so bad if you know that it's a slide deck when you load the page.)
Try using the arrow keys to navigate. It took me multiple tries to get it figured out.
Use up/down to navigate within a chapter/topic. Use left/right to switch between topics.
The docs are better than the slides.
The project relies on Rayon [1] for scheduling parallel tasks and Cranelift [2] to JIT the hot loops.
There are plenty of other interesting features like auto-FFI, bytecode caching (similar to Python's .pyc files), and "daemonize" mode (similar to mod_perl or FastCGI).
Where's the codebase?
I had to build a Perl implementation of the Chaskey mac algorithm. ChatGPT spat out a working Perl prototype based on a C file for Arduino. It quite slow with not very much to optimize, so I made it write it with XS. A hour later I have a working XS implementation that compiles and tests cleanly.
So the AutoFFI thing is super interesting. The .plc also.
Code is hidden so far. He ships only binaries.
The autoffi thing is nothing new, I did that with cperl a decade ago. Added native types also, which he doesnt have yet.
Awesome to see a perl JIT. I love perl, and it's exciting to see something that tries to offer good-enough compatibility to run most perl code.
Impressive, ambitious work.
I wonder how long he waited for the CPAN nologin case. I remember requesting a CPAN account 3 years back and it took ~2 months for someone to look at and accept.
I'm too scared to check how good llms are in writing perl.
Very good, actually. But you have to nudge them slightly. Tell them you prefer the modern version of the language, with gradual typing† and function signatures, and you'll get very good results. Perl interpreter comes standard on modern OSes and due to permissive licensing and impeccable backwards compatibility you can always assume you deal with very modern versions of Perl.
I write Perl scripts that are 10-100 lines of code, and at this size Perl is a Strictly Better Bash: better syntax, some type checking, better text support, and still effortless calls to external processes: essentially you put a command with arguments in backticks, and you get it's output. Ruby can do it too, but not all systems have it. Python is another obvious choice but calling external commands in it is annoying. I also use Perl for some one-liners as a better `sed` for text replacements.
† Perl nowadays have TypeScript-style type checking for function parameters. So, while the syntax is wild sometimes, the language is much better than it used to be.
> Perl nowadays have TypeScript-style type checking for function parameters.
I can't believe that.
TS code, compile time error "TS2345: Argument of type 'null' is not assignable to parameter of type 'number'."
Perl code:function foo(x: number): void {}; foo(null);
This code passes CHECK (perl -c), but should not if you are correct.use Kavorka qw(fun); use Types::Standard qw(Num); fun foo(Num $x) {} foo(undef);I invite you to prove the claim. Rewrite this with any module you like.
Because perl5 Types::Standard went the broken python way of type hints.
cperl types worked, and actually made it faster.
What are you using for parameter type checking? I switched to native function signatures, native try/catch and might look into the new class system soon, but I don't recall native type checking...
Are you talking about perl 5 or perl 6?
A few years ago; perl 6 renamed itself to 'raku', so the perl 5 folks can continue to improve/maintain the original 'perl'.
5 has this. There are modules that get you to function signatures and type constraints. It's all opt-in and, as was said, you have to nudge LLMs to use it, but they can and the results are indeed better.
What kind of performance impact does it have? Obviously it depends on the specific program, but let's say the worst case scenario, something like a recursive implementation of the factorial function.
> What kind of performance impact does it have?
Minor. Faster unpacking of @_, but it's not a huge win until you have a lot of arguments. The conventional Perl 5 interpreter has no JIT to leverage the benefits of stronger types, inline functions, unroll loops, etc. A factorial function has few arguments, so the unpack gain will be small to nothing.
Codex for some reason sometimes runs Perl instead of Python to work with local files
Why can't I scroll on this website ;___; come on now please
Not to be confused with GNU parallel[1], written in Perl.
When will perl 7 be released?