../

Bevy Jam #1 data exploration

Published: 31 March, 2022, updated: 10 April, 2022

Intro

The first Bevy Jam offers an opportunity to look at how different people use Bevy (at least in the context of a game jam) - which features they use, which crates they use to augment Bevy, and also which asset types they use. This can help highlight commonly used features or dependencies that might encourage further development, or infrequently used or avoided features that may need better documentation, examples, or improvements.

Some summary highlights

Dependencies

These are the direct expressed dependencies of the game project, representing what the game developer directly interacted with. I somewhat-arbitrarily decided what were “game development-related” vs. general Rust or infrastructural crates.

Bevy-specific dependencies

These are 38 unique dependencies built to work directly with Bevy. Debugging (perhaps unsurprisingly?) was the most common category, and egui was a clear favourite. bevy_kira_audio was used for 9 of the 11 entries in the top 25 entries that had audio assets, showing a clear preference for skipping the built-in audio capabilities.

Game development related dependencies

Unsurprisingly, most entries used randomness, and most (43/55) used the rand crate. Perhaps surprisingly, many entries had no non-Bevy dependencies beyond that crate.

Bevy feature usage

Plugins and features

Cameras

This one is a bit of a stretch - just looked at whether PerspectiveCameraBundle or OrthographicCameraBundle were present (preferring perspective), and if ::new_2d() or ::new_3d() were used (preferring the latter).

Assets

Asset Categories

Unsurprisingly, most entries used fonts and images. Fewer than half of games used audio. Using external level editors was infrequent, as well as imported shaders.

Font asset formats

Fonts were the most frequently used asset category, and roughly 50 unique font faces were used over all entries, while the top 25 entries used roughly 25 different font faces. Fira Sans (13) and Fira Mono (9) were the only faces used by more than 2 entries.

Image asset formats

PNG dominates for plain images/textures, while Aseprite is the infrequently used sprite format.

Audio asset formats

Model asset formats

The binary GLTF variant is preferred, although not many entries used 3D model assets

Level asset formats

Not many projects used level assets authored externally, but LDtk was slightly preferred over Tiled.

Shader asset formats

Game dependency and asset data

Methodology

This ended up being a single Rust binary that does all the work, with caching of external data to both be a good citizen and to speed up iteration. There are two main phases of collection - fetching information from itch.io, and then looking at the source code - and then reporting.

itch.io data

Itch.io was the platform used to host the jam, and while it doesn’t have official APIs for getting jam data. Luckily, the entries page is actually built client-side, and fetches the data for the entries via JSON - https://itch.io/jam/316945/entries.json. Note that this URL has the jam ID in it (316945, not the vanity name (bevy-jam-1). This entries page (and the data that backs it) doesn’t include rankings, which is on the results page. The results page renders server-side, but there happens to be an easily-guessed JSON file with the data - https://itch.io/jam/316945/results.json. Both of these could be fetched without authentication, so they could be fetched from code, and deserialized using serde_json. For all URL retrieval, reqwest was used, with a simple on-disk cache to prevent fetching the same resource more than once, and a sleep between fetches (both of which are probably easy to do with middleware).

The Bevy Jam encouraged, but did not require, source code to be provided for the software. Unfortunately, Itch.io doesn’t appear to provide a way to structurally denote things like a link to the source code, and so none of the internal APIs provide any such thing. There are two URLs available from here per entry - the entry page specific to this jam, and the general game page. The entry page doesn’t appear to have any useful information on it, but the general game page has freeform information from the game’s developers, as well as potentially download links. Using the scraper crate, all the links in the downloaded game pages could be examined for potentially being a link to a code hosting platform. GitHub appeared to be the only platform used for the jam. (It’s possible some entries used download links, but a spot-check showed these were largely used for binaries.)

Source data

With homogenous usage of Git repositories for code, the git2 crate could be used to fetch the repositories (with a simple on-disk cache). The cargo-metadata crate could be used to extract the dependencies from the Cargo.toml files (with a simple fix-up for one case where the file was named the lower case cargo.toml). Asset data could be found using the walkdir crate to traverse the source code, focusing on any assets folders. Finally, the grep crate from ripgrep could be used to find simple regex patterns in files to discover use of particular cameras, plugins, and so forth.

Reporting

This was largely done with plain text, although for the main table, the comrak crate was used to generate the Markdown table in Markdown.

Further work

In an ideal world, something like Tree Sitter could be used to examine the source code to be able to answer questions like “what are all the structs used from bevy modules by this repository”, or look at initialization for particular structs to tell if they’re configured for 2D or 3D usage.

There was another Rust game jam using the Itch.io platform, the Rusty Game Jam, which could probably be evaluated using very similar approaches. Trying to do this for something like Ludum Dare would require a different approach - besides being a different platform, only a small subset of entries use Rust, and it may be hard to discover this. Perhaps this could be attempted by using a compiled list of Rust games from a source like Reddit or Twitter.