Wasm GC: What Exactly Is It

(and Why Should I Care)

Ivan Mikushin, VMware

Who am I

  • Staff Engineer @ VMware

  • Recognized Contributor @ Bytecode Alliance

  • Working on Wasm GC support in wasm-tools
    (as a side-project)

What got me here

Curiosity ๐Ÿ˜บ

What got me here

  • ๐Ÿ•ถ๏ธ I've been watching WebAssembly since 2018
  • ๐Ÿ”ฎ Recent Interest: Zero-Knowledge Cryptography

Zero-Knowledge Cryptography

Verifiable Computation

Example: Circom

pragma circom 2.0.0;

template A(n){
   signal input in;
   signal output out;
   var array[n];

   out <== array[in];
}

component main = A(10);

Error: Non-quadratic constraint was detected statically, 
using unknown index will cause the constraint to be 
non-quadratic

General Purpose Verifiable Compute

  • ๐Ÿ’ก Any program execution should be provable
    (not just ones in circuit languages)

  • ๐Ÿ’ก In one language (Wasm)

  • ๐Ÿ’ก Just compile everything to Wasm!

  • ๐Ÿ˜พ Problem: most popular languages require GC

# TODO

  • What is Wasm GC

  • Why should I care

Why I care

Most popular programming languages

  • ๐Ÿ”ฎ will be compiled to Wasm

  • ๐Ÿ˜พ require GC

... ๐Ÿ”ฎ will be compiled to Wasm

  • Empirical law

    • What can be done in JavaScript
    • will be done in JavaScript

  • will (soonโ„ข๏ธ) read:

    • What can be compiled to Wasm
    • will be compiled to Wasm

... ๐Ÿ˜พ require GC

  • Java (and JVM-based)
  • JavaScript (and compiled to JS)
  • Python
  • Ruby
  • Go
  • ... (many more)

# TODO

  • What is Wasm GC

  • โœ… Why should I care

What is GC

Garbage Collection

Per Wikipedia:

  • A form of automatic memory management.

  • Garbage collector attempts to reclaim allocated memory, that is no longer referenced.

  • Invented by John McCarthy around 1959 to simplify manual memory management in Lisp.

What is Wasm GC?

https://github.com/WebAssembly/gc

"proposal to to add garbage collection (GC) support to WebAssembly"

What is Wasm GC not?

  • A garbage collector specification

    • Web-based runtimes already have
      high performance GCs

    • Just need heap types to make use of them

  • Seamless interoperability between multiple languages

So, what is Wasm GC?

  • A bunch of new types:
    New heap types and new ways to define types.

  • A bunch of new instructions:
    ~30 new instructions to manipulate reference types.

MVP vs Post-MVP

(MVP โ€” Minimum Viable Product)

MVP

  • Reference Types

  • Heap Types

  • Subtypes

  • Recursive Types

  • New Instructions

Post-MVP

(in the future, maybe)

  • Type parameters (polymorphism, generics)

  • Threads and shared references

  • Weak references

  • (a ... ton more)

Current scope (MVP)

  • Reference Types

  • Heap Types

  • Subtypes

  • Recursive Types

  • New Instructions

Kinds of Memory in Wasm

  • Stack
  • Linear memory
  • Heap โ€” added with GC

Heap

A pool of memory used to fulfill memory allocation requests (Wikipedia)

Disjoint hierarchies of heap types

  • function types โ€” classify functions (within the module)
  • external types โ€” external references (possibly owned by the embedder)
  • aggregate types โ€” dynamically allocated managed data (such as structures, arrays, or unboxed scalars)

Type Hierarchy

(from Chromium's

Wasm GC proposal: Spec for Prototype Implementation v.6)

Already existing heap types

Indexed Types

Examples:

  (type $A (struct (field (mut i32)) (field i64)))

  (type $B (array (mut i32)))

// Go

type A struct {
  f0 int32
  f1 int64
}

type B []int32

Reference Types

WAT syntax: (ref null? <heap-type>)

Examples:

  (type $A (struct)) ;; index = 0

  (func $f1 (param (ref null $A)))
  (func $f2 (result (ref 0))) ;; same as `(ref $A)`

// Go

type A struct {}

func f1(a *A)  {/* */}
func f2() *A {/* */}

Shorthands

  • funcref == (ref null func)
  • externref == (ref null extern)
  • anyref == (ref null any)
  • nullref == (ref null none)
  • nullexternref == (ref null noextern)
  • nullfuncref == (ref null nofunc)
  • eqref == (ref null eq)
  • structref == (ref null struct)
  • arrayref == (ref null array)
  • i31ref == (ref null i31)

Subtypes

Examples:

  (type $A (struct))
  (type $B (sub $A (struct)))
  (type $C (sub final $A (struct)))

// Java

class A {}
class B extends A {}
final class C extends A {}

Recursive Type Groups

Examples:

  (rec
    (type $A (struct (field $b (ref null $B))))
    (type $B (struct (field $a (ref null $A))))
  )
  
  (type $C (struct (field $f0 i32) (field $c (ref null $C))))

// Go

type A struct {
  b *B
}
type B struct {
  a *A
}
type C struct {
  f0 int32
  c  *C
}

New Instructions

~30 new instructions:

  • i31.* -- creating and manipulating "unboxed scalars" (31-bit integer values).
  • array.* and struct.* -- creating and manipulating arrays and structs.
  • ref.* -- casting and testing values of reference types.
  • br_on_* -- conditional branching depending on values of reference types.
  • extern.* -- external reference conversion.

# TODO

  • โœ… What is Wasm GC

  • โœ… Why should I care

A prediction ... ๐Ÿ”ฎ

Compilers to Wasm with GC support are coming

  • Go
  • JavaScript and JS-compilable languages
  • Java and JVM-based languages
  • Python
  • Ruby
  • ... (more)

WebAssembly Core Specification

Release 2.0 + tail calls + function references + gc
(Draft 2023-08-28)

https://webassembly.github.io/gc/

https://docs.wasmtime.dev/contributing.html

https://docs.wasmtime.dev/stability-wasi-proposals-support.html

Thank You! ย ๐Ÿ˜ป

@imikushin ย โ€” ย on ๐• (Twitter), GitHub, LinkedIn

Special Thanks to:

  • Nick Fitzgerald (Fastly, Bytecode Alliance)
  • Daniel Lopez Ridruejo (ex VMware, Bytecode Alliance)
  • David Wong (zkSecurity)