info_tfgrid/collections/developers/internals/rmb/rmb_coding_guides.md
2024-06-13 15:50:44 -04:00

5.1 KiB

Coding Guides

Table of Contents


Introduction

We present coding guides for the reliable message bus (RMB).

This document will always be work in progress. Read the official docs for updates.

Module Structure

In Rust there are multiple ways to create a (sub)module in your crate

  • <module>.rs A single file module. can be imported in main.rs or lib.rs with the keyword mod
  • <module>/mod.rs A directory module. Uses mod.rs as the module entrypoint always. Other sub-modules can be created next to mod.rs and can be made available by using the mod keyword again in the mod.rs file.

We will agree to use the 2nd way (directory) but with the following restrictions:

  • mod.rs will have all traits and concrete types used by the traits.
  • <implementation>.rs file next to mod.rs that can include implementation for the module trait. z

Example

Following is an example of animal module.

animal/
   mod.rs
   dog.rs
   cat.rs

File names are always in snake_case but avoid the _ as much as possible because they basically look ugly in file tree. For example we prefer the name dog.rs over dog_animal.rs because we already can tell from the module name that it's a dog animal. Hence in the identity module for example the name ed25519.rs is preferred over ed25519_identity.rs because that's already inferred from the module name. The mod.rs file then can contain

pub mod dog;
pub mod cat;
pub use dog::Dog;

pub trait Food {
    fn calories(&self) -> u32;
}

pub trait Animal<F>
where
    F: Food,
{
    fn feed(&mut self, food: F);
}

The dog.rs file then can contain

use super::{Animal, Food};

pub struct DogFood {}
impl Food for DogFood {
    fn calories(&self) -> u32 {
        1000
    }
}
pub struct Dog {}

impl Animal<DogFood> for Dog {
    fn feed(&mut self, food: DogFood) {
        println!("yum yum yum {} calories", food.calories());
    }
}

A user of the module now can do

use animal::dog::{Dog, DogFood};

For common implementation that are usually used in your modules, a pub use can be added in mod.rs to make it easier to import your type. For example

// dog is brought directly from animal crate
use animal::Dog;
// cat i need to get from the sub-module
use animal::cat::Cat;

Naming Conventions

Following the rust guide lines for name

  • file names are short snake case. avoid _ if otherwise name will not be descriptive. Check note about file names above.
  • trait, struct, enum names are all CamelCase
  • fn, variables names are snake case

Note, names of functions and variables need to be descriptive but short at the same time. Also avoid the _ until absolutely necessary. A variable with a single word name is better if it doesn't cause confusion with other variables in the same context.

The name of the variable should never include the type.

error Handling

We agreed to use anyhow crate in this project. Please read the docs for anyhow

To unify the practice by default we import both Result and Context from anyhow

Others can be imported as well if needed.

use anyhow::{Result, Context};

fn might_fail() -> Result<()> {
    // context adds a `context` to the error. so if another_call fails. I can tell exactly failed when i was doing what
    another_call().context("failed to do something")?; // <- we use ? to propagate the error unless you need to handle the error differently

    Ok(()) // we use Ok from std no need to import anyhow::Ok although it's probably the same.
}

fn might_fail2() -> Result<()> {
    if fail {
        // use the bail macro fom anyhow to exit with an error.
        bail!("failed because fail with set to true");
    }
}

> NOTE: all error messages starts with lowercase. for example it's `failed to ...` not `Failed to ...`

logs

logging is important to trace the errors that cannot be propagated and also for debug messages that can help spotting a problem. We always gonna use log crate. as

log::debug!(); // for debug messages
log::info!(); // info messages

Note only errors that can NOT be propagated are logged.

NOTE: All log messages start with lowercase.

Function Signatures

For function inputs (arguments) generic types are preferred if available over concrete types. This most obvious with string types. depending on the function behavior

Examples

This is bad:

fn call1(key: String);
fn call2(key: &str);

It is preferred to use:

// in case function will need to take ownership of the string.
fn call1<k: Into<String>>(k: K);
// inc ase function will just need to use a reference to the string.
fn call2<K: AsRef<str>>(k: K);

// this will allow both functions to be callable with `&str`, `String`.