Skip to content

Latest commit

 

History

History
141 lines (103 loc) · 4.46 KB

ROADMAP.md

File metadata and controls

141 lines (103 loc) · 4.46 KB

Roadmap

This is a list of all the research and work I'd like to do within the context of LAM sort of organized in small chunks of work that are more or less self contained.

There is no promise that I'll address these sequentially or completely, as I'm mainly guiding the work by building utilities like cat, or grep.

Milestones

  • 2020 Q4 - Milestone 1: BEAM Binaries ✔️
  • Milestone 2: Single Process Programs 🔨
  • Milestone 3: Concurrency-Oriented Programming 🔨
  • Milestone 4: Multi-Core Scheduling 🔨
  • Milestone 5: Native Extensions 🔮
  • Milestone 6: WebIDL API 🚀
  • Milestone 7: Native Graphics 🚀

2020 Q4 - Milestone 1: BEAM Binaries ✔️

The initial goal here is to package a bunch of .beam files into a single relocatable binary for a specific architecture (eg, only x86_64-unknown-linux-gnu), that can run:

  • Natively
  • On a WASI enabled runtime
  • On a browser through WebAssembly

There should be a small cli, that we can run:

$ lam build *.beam -o bin.exe -t native
$ ./bin.exe
Hello, Joe!

$ lam build *.beam -o bin.wasm -t wasm
$ wasmtime bin.wasm
Hello, Joe!

$ lam build *.beam -o js.wasm -t web
$ parcel serve index.html
...

# and in a broswer going to the url we see in the JS console
> Hello, Joe!

Milestone 2: Fast Single-process Programs 🔨

For this milestone, I'd like to be able to write arbitrary single-process Erlang programs.

The goal here would be to be able to create any sort of values that are valid in Erlang, and call functions in any way valid in Erlang as well, so that we can build some command line tools.

Some things that should be possible are:

  • run forever:

    • f() -> f(). should just run and run and run ✔️

  • run fast:

    • no-op boot time should be under 10ms ✔️

  • use common data types:

    • atoms ✔️
    • integers ✔️
    • floats
    • binary strings ✔️
    • lists ✔️
    • tuples 🔨
    • maps

  • first class functions:

    • creating lambdas and applying them ✔️
    • passing lambdas around as values ✔️

Milestone 3: Concurrency-Oriented Programming 🔨

Here the goal should be to enable building solutions that scale by orchestrating hundreds of thousands of processes that communicate with each other.

We'll need to support:

  • spawning processes ✔️
  • terminating processes 🔨
  • monitor / demonitor processes
  • link / unlink processes
  • sending messages ✔️
  • selectively receive messages ✔️

Some things that should be possible are:

  • spawn millions of processes in constant time (shouldn't get slower to spin up new processes)
  • core OTP modules like supervisor or gen_server should have all the instructions needed to run (give or take small changes)

Milestone 4: Multi-Core Scheduling 🔮

The goal here is to Implement symmetric multi-processing to spread the work across several cores by means of either job-stealing or some other configurable policy.

Stuff that needs doing:

  • maintain and run as many schedulers as cores there are ✔️
  • load-balancing / work stealing
  • message routing

We should be able to run things like:

  • map-reduce patterns, like a word count program, across all cores without modifications

Milestone 5: Native Extensions 🔨

To support building useful native and web applications, we need to figure out how to allow people to link custom code that interfaces with the runtime, and expose it in a transparent way.

For example, natively, we may want to open a socket to make an HTTP request. On a browser, we may want DOM manipulation. On other runtimes (like wasi), we may want other things to be available.

But regardless of the runtime, the calls should look completely transparent. Calling json:parse(String). should be the exact same for a consumer. Whether the implementation is fulfilled by bytecode running on LAM, or by native platform code, should be irrelevant.

Milestone 6: WebIDL API 🚀

To make it more useful to build web things, we should generate the bytecode necessary to interface with the web platform from the WebIDL description. This is how the web-sys crate currently does it.

Milestone 7: Native Graphics 🚀

From the work I've done before on the erlang-gui project, it should be possible for us to generate native extensions to Vulkan and Skia to build native applications.