Popcorn: Run Elixir in WASM

Popcorn: Run Elixir in WASM(popcorn.swmansion.com)

145 points by clessg 1 year ago | 27 comments

Jump3r 1 year ago |

Hey, Kuba from team working on Popcorn here. I wasn't expecting it to be posted on hn but here we are. Feel free to ask me any question.

pesnk 1 year ago |

Congratulations on the project! It works great, until we need to handle the best part of Elixir, that's creating multiple actors.

This Task code, for example doesn't work.

  Enum.map(0..10, fn(_) ->
    Task.async(fn -> IO.puts("new process") end)
  end) |> Task.await_many()

mathek 1 year ago | |

Tasks seem not to work indeed, but spawning processes works. Try

  Enum.map(1..10, fn _i ->
    spawn(fn -> IO.puts("new process") end)
  end)

Also, there are two scenarios: compiling code in the browser and running it, and running precompiled code in the browser. In the latter case more things work, for example the 'game of life' example uses GenServers, Supervisors and Registry:

https://popcorn.swmansion.com/game_of_life/

https://github.com/software-mansion/popcorn/tree/main/exampl...

But yes, it's still unstable. Improving the stability is our main focus right now.

Lord_Zero 1 year ago |

I'm honestly surprised at how slow WASM is moving. As a very experienced web dev, when I first learned about WASM I was sure people would be building production UIs in Python and Golang and other traditionally server-side languages.

afavour 1 year ago | |

That’s actually what I feared the most and I’m glad we haven’t seen it happen.

Python and Go are not small languages, making users download entire runtimes to do something that can be done fine without them would be a huge shame. The performance problem with web UI is the DOM, not JavaScript.

camdenreslink 1 year ago | | |

I kind of thought popular runtimes would just be bundled with the browser if we could get an "official" enough source (e.g. Python Software Foundation or Google). Then users wouldn't need to download a million different versions and sources of Python for each website.

lylejantzi3rd 1 year ago | | |

You can mitigate some of that with web workers, but it's a shame that multiple websites can't share a python runtime the same way they can on a shared linux box. You have to download a separate one for every website that uses it. But, if you follow that train of thought to the end, you wind up with the browser itself bundling those runtimes, just like it did with Flash back in the day.

I'm not sure any of this would be an improvement over what we have now.

hinkley 1 year ago | | |

At least with Go static analysis should allow you to tree shake the bits you aren’t using.

The problem with the CDN solution was always that it assumed that everyone would be on a couple versions and that never happens. With success comes more contributors and with more contributors come more point releases and more users who are not on the latest version.

So soon you could have five versions for five sites you visit.

eterm 1 year ago | |

Well there's blazor, which does that rather well, but it's treated with the same suspicion that most MS frameworks are. The fear that it'll be killed for poor adoption, leading to poor adoption.

The blazor adoption probably isn't even that bad, but it's hard for MS to shake this stigma since so many people got burned on Silverlight and don't ever want to make the same mistake.

pier25 1 year ago | | |

Blazor wasm is just too heavy for most use cases.

In terms of speed, it's not even close to anything else in JS:

https://krausest.github.io/js-framework-benchmark/current.ht...

throwawaymaths 1 year ago | | |

I thought blazor failed because people didn't like websites that take minutes to load up all the websites assets. I'm not sure why, it could be:

- Technical issue with blazor performance or blazor makes perf regressions hard to fix

- blazor technical framework encourages programming style that is bad for perf

- blazor or blazor ecosystem attracts programmers that can't deal with perf issues

Muromec 1 year ago | |

Wasm doesn help much with UI, its more about moving parts of the server into the client as blacboxes with no access to dom.

The blackbox can of course return the state of UI to be rendered/pathced, but that doesnt unlock much (if any) interesting capabilities for amount of overhead it adds

jcmontx 1 year ago | |

That's what the .NET folks have been doing for many years now. Blazor is .NET on the client-side

IshKebab 1 year ago | |

I really hope nobody plans to make web front-ends in Python. Jesus.

codetrotter 1 year ago |

For anyone else wondering the GitHub repo is here:

https://github.com/software-mansion/popcorn

Couldn’t find any link to it on the article

Muromec 1 year ago |

AtomVM is something like 1m when compiled, isn't it?

mathek 1 year ago | |

Compiled to WASM it's ~700K, gzipped only 190K. But there's also BEAM code - the whole standard library is ~3M gzipped. We plan to do tree shaking to only include the used parts of the stdlib - then it should get way smaller for most use cases.

Muromec 1 year ago | | |

That's a lot of wasm and beam. I tried to do it the other way -- compile beam straight to wasm and it seems to work in straightforward cases, producing wasm binaries in kilobytes sizes. With wasm fx it's even possible to do processes natively. You can find the link in bio.

Cool thing would be to somehow wire it up to liveview to make it possible to run some parts on the frontend side transparently.

I'm not doing any active development on it right now, but it looked pretty doable when I tried last time.

nelsonic 1 year ago |

Very promising. But still not stable. One to watch.