Running WebAssembly from my hobby compiler!

Code: github.com/contextfreeinfo/rio
0:00 Intro
0:53 Internal & Wasm representation
1:58 Support scripts & export example
3:15 Wasm generation
3:43 Stages & type representation
5:30 Compiler internals
6:08 Type inference strategy
8:38 Core module
9:02 Main function and friends
10:12 Keep on trucking
11:35 Running junk
12:25 Prep for tooling
12:56 Summary

[@dercoder015]

very cool, i was also interested to compile to wasm but getting into it seeemed daunting. Will definitely use your project as inspiration

[@contextfree]

Definitely review output from other compilers for ideas. That helps me a lot.

[@UliTroyo]

I dig the syntax :)

[@1vader]

There are some wasm optimizers like binaryen which I'd assume would optimize stuff like variables that are stored and immediately loaded again and then never used again. So maybe you could just run that after your compilation to get some speedups. Although I guess it might add some annoying dependencies and performance probably isn't exactly a priority.

[@contextfree]

Yeah, I've used wasm-opt from binaryen, and I think that's my fallback. Worst case, people can follow up with that after a non optimizing compiler, if/when that matters to them. Maybe I should include it as part of my overall test script, though, so I can compare processing times and output sizes and contents.

[@irubberduck]

Really interesting series. A completely different perspective compared to languages trying to be mature or production-ready. Especially the wild file / program.
Is wasm limited to a 32-bit address space? Otherwise, why do your strings get unwrapped into 32-bit address and length parameters?

[@AtomicalYT]

WebAssembly in its core spec uses 32-bit integers for pointers. There is a separate proposal to have 64-bit memory addresses

[@donutloop]

Cool

[@0marble8]

As im writing one of those parser generator programs for my thesis, ive been wondering, do real language designers actually use them? It seems like a lot of projects i looked at, all write their parser by hand

[@contextfree]

I've tried antlr and others in the past, and I think they're good for some things, but yeah, sometimes easier to be flexible with hand rolled. Or maybe I'm just stubborn, even if other people have good reasons to roll their own.

[@casperes0912]

What was the difference between print message and print(message)? (and possibly print (message))?

[@contextfree]

Nothing really. I mostly want to support `print message` so constructs like `if whatever be ...` don't need parens, and those constructs get parsed the same as any function call, even if processed differently in the compiler (effectively macros although I'm not sure yet if I plan to expose macro definition support to user code). But when nesting function calls, not needing parens at every level might also be prettier sometimes, so I'll have to think about recommended style some. And I'm also thinking more on the `print (message)` thing. I was considering allowing tuples earlier, although as I mentioned in the video, I'm leaning away from that now. If there were tuples, then `something (a, b)` would pass in a single tuple as the first argument, whereas `something(a, b)` would have two separate arguments. I didn't change that comment yet because I want to think about it a bit more. But if I really don't have tuples, the space might as well be irrelevant, unless I'm forgetting something. Sometimes I only notice details when I start implementing. Anyway, sorry for being so longwinded in my reply!

[@contextfree]

And yeah, I have some plans, but I'm also winging a lot.

[@casperes0912]

@@contextfree right. Makes sense. I think each style, by association, will encourage different styles. Print message would make me write more functional as my brain would initially group the language with ML languages like OCaml. Print(message) puts me in a procedural mindset. At least initially.

[@contextfree]

I'm definitely aiming for functional-first attitudes in Rio, but not aiming for purity.

[@familyshare3724]

I imagine a statically typed language with only two primitive cardinals: non-null value and non-null list. The "nullable" value can only be a list with zero elements.

[@contextfree]

Any variety in the non-null value or do you imagine using Peano numbers or some such?

[@familyshare3724]

@@contextfree oh yes, full typing, definable class types. Just no null. Anything "nullable" is just the empty set of that type. Consider a deserialized JSON with no null. Empty string same as [] (if not predefined). Arrays and objects are simply different types of a collection.

[@familyshare3724]

@@contextfree i would hope it would be simpler and faster than Lua, with types, and of course 0 indexed.

[@contextfree]

Thanks for the further explanation! I guess now would be a bad time to say that I often flirt with 1-based indexing ...

[@atiedebee1020]

So... Lisp?