This talk is so good that I won’t spoil the twist for you. Yeah, this talk has a twist, believe it or not. Just so if you doubted that one of the inventors of Scheme and Java was not a super genius.
I really despise how Java was ‘twisted’ into this corporate shit of the language that we got today. Sad, really. At least Scheme is still one’s best bet to knock his teeth on a compiler.
I recently came across an awesome Scheme compiler known as Cyclone. It’s really nice. Scheme is truly the best LISP.
Cyclone is better than Racket even. Except, I wished they would put their shit up on package managers. Because you need a bootstrap to install the damn thing. I don’t wanna release a project where people have to go through hell to install it. At least, Cyclone translates to C, not Assembly, so you can just translate it yourself and put it on a tarball. Bonus if you put a backdoor there.
54 minute video, is there a paper? Video is an OOPSLA talk, “Growing a language”.
It might be this: https://homepages.inf.ed.ac.uk/wadler/documents/steele-oopsla98.pdf
The advice offered by Steele in this video no longer applies. It’s still a bit more up-to-date than Kernighan’s talk of a similar title. The fun of this video is in how he twists the English language. He’s truly an erudite man.
The reason this advice no longer applies is, that I, as a person trying to enter the world of langdev, at least personally, see no reason on defining a new language. I think we should find new ways to describe languages we already have i.e. implement them.
I am currently making a C compiler in OCaml, besides some other languages. I just began work on the AST tags. I somehow decided to use SSA versioning here.
But descring a new compiler for C, it’s blaze. I only do it because C is easy, at least non-intrinsicaly. For example, there’s no automatic GC. There are no first-class functions, or function literals (aka ‘lambas’, although this term is massively ‘disused’ — function literals are one thing, lambda literals are one thing, lambda expressions are one thing, function expressions are another thing and so on and so forth – and I don’t have classical education, I just managed to understand that imperative languages abuse the term ‘lambda’ to a dangerous degree. They have named their literals after the concept which it derives from ,it’s like calling binary computers, not the concept, the literal thing, 'Neumann machines, right? Like, go to Best Buy and say ‘Give me this Neumnan machine to play games on’! Maybe that’s because I am too uneducated that I think that, anyways).
Besides that, there are:
1- Too many old languages that could use a new veneer, like a SML compiler that uses MLIR or LLVM. 2- There are too many interpreted languages that could use being jitted. Like Ruby. Not sure if there’s jitted Ruby, but I just discoevered how sweet it is and I like a faster version 3- We could dig a mass grave and bury every Python user alive, after torturing them (I’m kidding! lol)
So I’m not very educated, I brute-force. I rely on ChatGPT models to spit facts at me, or give me validation on my work because I kinda need a ‘college simulator’. Like, I figure, I don’t have nay peers so let’s make this bot my peer.
In the realm of DSLs, let’s look at a successful example of ‘re-description’: Fish.
Fish is truly a marvel. Ever since I switched to it, you can’t beleive how faster I work. I don’t know if there were interactive-friendly shells before Fish, but Fish is ‘Friendly’ you know?
I am implementing my own shell too.
I dunno man. I’m just rambling.
Thanks.