Add WebAssembly support to the language and to dmd

[PetarKirov]

Description

• Implement toolchain libraries for working with WebAssmebly:
  • Assembling WAT / WAST (WebAssembly Text format) files to WASM (WebAssembly Binary format)
  • Disassembling WASM to WAT/WAST
  • Builder API for generating WASM modules programatically
  • Linker - link multiple .wasm files into a single .wasm file
• Implement WebAssembly compiler backend in D and integrate it with dmd
  • This backend should focus on:
  • implementation quality - unlike the current x86 dmd backend, which was written in old-style C, the new wasm backend should be written in modern D-style and should follow software engineering best practices, like unit testing
  • features - see https://webassembly.org/features/; perhaps the most relevant are:
    • Sign-extension operations
    • bulk memory operations
    • multiple return values
    • reference types
    • SIMD
    • exception handling
  • fast compilation
  • Backend optimizations are not important, as there are ahead of time compilers for WebAssembly that can sophisticated optimization on the unoptimized wasm output from dmd
• Port druntime to WebAssembly + WASI
• Port phobos to WebAssembly + WASI

What are rough milestones of this project?

The project can be split in two:

• adding WebAssembly backend to dmd
• porting Druntime and Phobos to WebAssembly and using LDC for D->Wasm compilation

and each one can be worked on to some extent independently from the other (e.g. the webassembly backend will compile programs in "BetterC" mode until druntime and phobos are ported.

How does this project help the D community?

It expands the range of platforms and uses cases that D supports

Recommended skills

• Bytecode-based virtual machines
• Compiler development

Point of Contact

References

<NG discussions, GitHub PRs, Bugzilla issues, ...>

[Herringway]

I can definitely see value in druntime support (#108 may be related), but what value does DMD support have? I don't think fast compilation is guaranteed anymore with the major backend changes required for that.

Would this cover the emscripten platform as well? A number of pre-WASI projects (such as SDL) out there only support it, and it may be worth considering for compatibility.

[mdparker]

WebASM support for DRuntime has filtered up our project list. We had a discussion about it in a recent planning session. There's no appetite for doing anything in dmd for it right now.

[maxhaton]

Merging the druntimes of DMD and LDC would go a long way. There's a lot of wasm crap floating around that should really be upstreamed.