ruci 1.0.2

A UCI (Universal Chess Interface) crate.
Documentation

tests clippy coverage docs.rs crates.io

ruci

Rust Universal Chess Interface.

This crate is a full implementation of the UCI protocol using shakmaty for relevant types. The UCI protocol is the most widely used way for chess GUI's to communicate with engines and vice versa.

Includes a struct to manage the I/O with an engine, but does not offer the same I/O management for engines themselves. However, engines can still take advantage of this library to add strong typing to strings, as is demonstrated in the engine example.

#![no_std] compatible.

See also the examples. You can run each one with cargo run --package <example-name>.

Comparison

There's two other crates that I'm aware of which serve a similar purpose; vampirc-uci and shakmaty-uci. shakmaty-uci is basically an improved version of vampirc-uci, so I'll only cover shakmaty-uci. Anyways, it:

  • Doesn't separate the two types of messages (engine, GUI) and specific messages. It has one big enum which mostly uses enum fields for message data. This is inconvenient because you can't represent specific messages, just the whole UciMessage enum.
  • Doesn't provide IO communication with an engine.
  • Is 3 or more times slower than ruci when deserializing, 2 or more times slower when serializing. Benches don't cover that many types but there's a trend. Results available at tigerros.github.io/ruci/bench, or you can view a more compact version in the summary of the latest Bench workflow run. Sources at benches.
  • Has more tests, but I don't know about the coverage.
  • Uses owned versions of non-copy data. This makes for a cleaner API compared to using Cows (which is what ruci does), but it hurts performance. Converting a message to a string doesn't require owned data, and seeing as that is half of the functionality of this crate, it would be wasteful to force ownership.
  • Uses nom for parsing, whereas ruci doesn't pull in anything.

Benches

ruci has more benches per scenario than the other two. This is because it uses Cows to allow for borrowed or owned data. So, the suffix borrowed in bench names means that it is using a statically borrowed type rather than the owned type, e.g. &'static str instead of String. The flip side is the suffix owned.

Features

  • default: no features are enabled by default.
  • engine-sync: adds the Engine struct for communicating with an engine. Requires std.
  • engine-async: enables the engine-sync feature and adds async versions of all functions using tokio.
  • tokio-process: enables the engine-async and tokio/process features and adds a tokio version of Engine::from_process.
  • serde: enables serde support for most types. All implementations are derived with no parameters. Requires std.