Skip to content

patte/gleam-playground

Repository files navigation

gleam chat

Exploring Gleam. Svelte frontend using types and functions from Gleam. OTP actors to simulate a chat server and test concurrency.

Development

Setup

Use asdf to asdf install gleam and other dependencies defined in .tool-versions.

Easy

gleam run

Open localhost:3000/test.html in your browser.

Svelte Frontend

# run backend
gleam run

# build shared types (optional for first run)
cd src/shared
gleam build --target javascript
# copy only js and ts files to frontend
rsync -av --delete --include '*/' --include '*.mts' --include '*.mjs' --include 'gleam_version' --exclude '*' build/dev/javascript/ ../../frontend/src/generated/
cd ../..

# run frontend
cd frontend
bun dev

Open: localhost:3000 in your browser.

Features

Types and code sharing between Gleam and Svelte

This repo contains 3 projects:

./               # gleam module hello_world (backend)
├─ src/shared/   # gleam module shared (shared types)
├─ frontend/     # svelte (frontend)

The folder src/shared is part of the be project and also it's own project that is built to typescript and copied to the frontend project. This organisation comes from the limitation that gleam can not have different entrypoints for the targets erlang and javascript but nested projects seem to work.

The shared module contains type definitions and encode and decode functions for the messages between the frontend and backend. It is compiled into shared.mjs shared.d.mts. See how the fe consumes the types at onMessage and sendMessage. End to end types are working 🎉

Currently, everytime the shared code is edited it needs to be built and copied to the frontend manually with the commands noted above. Whether these files should be checked into version control is debatable, it's done here for this example to be explorable without running it.

OTP Actors

To test out erlang OTP and scheduling of tasks, the backend provides a simple chat server with OTP actors. One single room_actor, one websocker_actor per connection, and one calculator_actor on demand per websocket actor.

The room actor is created on startup. On connection a new websocket actor is created which registers itself (a subject) at the room actor. When the websocket actor receives a message from a client it parses it, sends it to the room actor, which sends it to all registered websocket actors (including the sender), which then send it to the client. This ensures that when you see your own message, it has been sent to the server and back to you.

Calculator

The calculator actors role is to provide for an example of a long running, blocking, task. It is created on demand by the websocket actor when it receives a message that is a calculation. The websocket actor then waits for the result before sending it to the room actor. This can be used to block one websocket actor and test that other actors are continuing to work.

OOM: The calculator actor solves factorials (3! = 3 * 2 * 1 = 6) and doesn't have an upper limit. For big inputs it will run out of memory and crash... the complete beam vm. Out of memory errors in erlang are not caught by the actor, so the complete beam vm crashes. I tried to find a solution to this "problem", but it doesn't seem to be possible easily also because it's an artificial problem: In the real world, APIs don't tend to offer unbouded calculations. For example image CDNs limit the maximum size of an image, which would by estimating the required memory, allow for a queue actor that doesn't start more calculation actors than the system can handle. Still it's not what I would have expected. It seems OOM is one of the few things that can crash the beam vm. I would be very interested in input on this topic.

Svelte Frontend

The svelte frontend is a simple chat client that connects to the backend and sends and receives messages. It uses the shared types to encode and decode messages. The code is currently a bit messy.

Ping

The client shows the round trip time in ms for messages. It's only possible to determine the RRT for messages sent by me because the clocks of clients are not synchronized.

Sources

Without these sources this project would not have been possible. Thank you to the authors! Sources:

Screenshot

Screenshot

x264Screen.Recording.2024-10-12.at.00.04.59.mov