CONTRIBUTING.md

Contributing to Bevy

Hey, so you're interested in contributing to Bevy! Feel free to pitch in on whatever interests you and we'll be happy to help you contribute.

Check out our community's Code of Conduct and feel free to say hi on Discord if you'd like. It's a nice place to chat about Bevy development, ask questions, and get to know the other contributors and users in a less formal setting.

Read on if you're looking for:

• The high-level design goals of Bevy.
• Conventions and informal practices we follow when developing Bevy.
• General advice on good open source collaboration practices.
• Concrete ways you can help us, no matter your background or skill level.

We're thrilled to have you along as we build!

Getting oriented

Bevy, like any general-purpose game engine, is a large project! It can be a bit overwhelming to start, so here's the bird's-eye view.

The Bevy Engine Organization has 4 primary repos:

1. bevy: This is where the engine itself lives. The bulk of development work occurs here.
2. bevy-website: Where the official website, release notes, Bevy Book, and Bevy Assets are hosted. It is created using the Zola static site generator.
3. bevy-assets: A collection of community-made tutorials, plugins, crates, games, and tools! Make a PR if you want to showcase your projects there!
4. rfcs: A place to collaboratively build and reach consensus on designs for large or controversial features.

The bevy repo itself contains many smaller subcrates. Most of them can be used by themselves and many of them can be modularly replaced. This enables developers to pick and choose the parts of Bevy that they want to use.

Some crates of interest:

• bevy_ecs: The core data model for Bevy. Most Bevy features are implemented on top of it. It is also fully functional as a stand-alone ECS, which can be very valuable if you're looking to integrate it with other game engines or use it for non-game executables.
• bevy_app: The api used to define Bevy Plugins and compose them together into Bevy Apps.
• bevy_tasks: Our light-weight async executor. This drives most async and parallel code in Bevy.
• bevy_render: Our core renderer API. It handles interaction with the GPU, such as the creation of Meshes, Textures, and Shaders. It also exposes a modular Render Graph for composing render pipelines. All 2D and 3D render features are implemented on top of this crate.

What we're trying to build

Bevy is a completely free and open source game engine built in Rust. It currently has the following design goals:

• Capable: Offer a complete 2D and 3D feature set.
• Simple: Easy for newbies to pick up, but infinitely flexible for power users.
• Data Focused: Data-oriented architecture using the Entity Component System paradigm.
• Modular: Use only what you need. Replace what you don't like.
• Fast: App logic should run quickly, and when possible, in parallel.
• Productive: Changes should compile quickly ... waiting isn't fun.

Bevy also currently has the following "development process" goals:

• Rapid experimentation over API stability: We need the freedom to experiment and iterate in order to build the best engine we can. This will change over time as APIs prove their staying power.
• Consistent vision: The engine needs to feel consistent and cohesive. This takes precedence over democratic and/or decentralized processes. See our Bevy Organization doc for more details.
• Flexibility over bureaucracy: Developers should feel productive and unencumbered by development processes.
• Focus: The Bevy Org should focus on building a small number of features excellently over merging every new community-contributed feature quickly. Sometimes this means pull requests will sit unmerged for a long time. This is the price of focus and we are willing to pay it. Fortunately Bevy is modular to its core. 3rd party plugins are a great way to work around this policy.
• User-facing API ergonomics come first: Solid user experience should receive significant focus and investment. It should rarely be compromised in the interest of internal implementation details.
• Modularity over deep integration: Individual crates and features should be "pluggable" whenever possible. Don't tie crates, features, or types together that don't need to be.
• Don't merge everything ... don't merge too early: Every feature we add increases maintenance burden and compile times. Only merge features that are "generally" useful. Don't merge major changes or new features unless we have relative consensus that the design is correct and that we have the developer capacity to support it. When possible, make a 3rd party Plugin / crate first, then consider merging once the API has been tested in the wild. Bevy's modular structure means that the only difference between "official engine features" and "third party plugins" is our endorsement and the repo the code lives in. We should take advantage of that whenever possible.
• Control and consistency over 3rd party code reuse: Only add a dependency if it is absolutely necessary. Every dependency we add decreases our autonomy and consistency. Dependencies also have the potential to increase compile times and risk pulling in sub-dependencies we don't want / need.
• Don't re-invent every wheel: As a counter to the previous point, don't re-invent everything at all costs. If there is a crate in the Rust ecosystem that is the "de-facto" standard (ex: wgpu, winit, cpal), we should heavily consider using it. Bevy should be a positive force in the ecosystem. We should drive the improvements we need into these core ecosystem crates.
• Rust-first: Engine and user-facing code should optimize and encourage Rust-only workflows. Adding additional languages increases internal complexity, fractures the Bevy ecosystem, and makes it harder for users to understand the engine. Never compromise a Rust interface in the interest of compatibility with other languages.
• Thoughtful public interfaces over maximal configurability: Symbols and apis should be private by default. Every public API should be thoughtfully and consistently designed. Don't expose unnecessary internal implementation details. Don't allow users to "shoot themselves in the foot". Favor one "happy path" api over multiple apis for different use cases.
• Welcome new contributors: Invest in new contributors. Help them fill knowledge and skill gaps. Don't ever gatekeep Bevy development according to notions of required skills or credentials. Help new developers find their niche.
• Civil discourse: We need to collectively discuss ideas and the best ideas should win. But conversations need to remain respectful at all times. Remember that we're all in this together. Always follow our Code of Conduct.
• Test what you need to: Write useful tests. Don't write tests that aren't useful. We generally aren't strict about unit testing every line of code. We don't want you to waste your time. But at the same time:
  • Most new features should have at least one minimal example. These also serve as simple integration tests, as they are run as part of our CI process.
  • The more complex or "core" a feature is, the more strict we are about unit tests. Use your best judgement here. We will let you know if your pull request needs more tests. We use Rust's built in testing framework.

The Bevy Organization

The Bevy Organization is the group of people responsible for stewarding the Bevy project. It handles things like merging pull requests, choosing project direction, managing bugs / issues / feature requests, running the Bevy website, controlling access to secrets, defining and enforcing best practices, etc.

Note that you do not need to be a member of the Bevy Organization to contribute to Bevy. Community contributors (this means you) can freely open issues, submit pull requests, and review pull requests.

Check out our dedicated Bevy Organization document to learn more about how we're organized.

Classifying PRs

Our merge strategy relies on the classification of PRs on two axes:

• How controversial are the design decisions.
• How complex is the implementation.

Each label has a prefix denoting its category:

• A: Area (e.g. A-Animation, A-ECS, A-Rendering).
• C: Category (e.g. C-Breaking-Change, C-Code-Quality, C-Docs).
• D: Difficulty (e.g. D-Complex, D-Good-First-Issue).
• O: Operating System (e.g. O-Linux, O-Web, O-Windows).
• P: Priority (e.g. P-Critical, P-High).
• S: Status (e.g. S-Blocked, S-Controversial, S-Needs-Design).

PRs with non-trivial design decisions are given the S-Controversial label. This indicates that the PR needs more thorough design review or an RFC, if complex enough.

PRs that are non-trivial to review are given the D-Complex label. This indicates that the PR should be reviewed more thoroughly and by people with experience in the area that the PR touches.

When making PRs, try to split out more controversial changes from less controversial ones, in order to make your work easier to review and merge. It is also a good idea to try and split out simple changes from more complex changes if it is not helpful for them to be reviewed together.

Some things that are reason to apply the S-Controversial label to a PR:

1. Changes to a project-wide workflow or style.
2. New architecture for a large feature.
3. Serious tradeoffs were made.
4. Heavy user impact.
5. New ways for users to make mistakes (footguns).
6. Adding a dependency.
7. Touching licensing information (due to level of precision required).
8. Adding root-level files (due to the high level of visibility).

Some things that are reason to apply the D-Complex label to a PR:

1. Introduction or modification of soundness relevant code (for example unsafe code).
2. High levels of technical complexity.
3. Large-scale code reorganization.

Examples of PRs that are not S-Controversial or D-Complex:

• Fixing dead links.
• Removing dead code or unused dependencies.
• Typo and grammar fixes.
• Add Mut::reborrow.
• Add Res::clone.

Examples of PRs that are S-Controversial but not D-Complex:

• Implement and require #[derive(Component)] on all component structs.
• Use default serde impls for Entity.

Examples of PRs that are not S-Controversial but are D-Complex:

• Ensure Ptr/PtrMut/OwningPtr are aligned in debug builds.
• Replace BlobVec's swap_scratch with a swap_nonoverlapping.

Examples of PRs that are both S-Controversial and D-Complex:

• bevy_reflect: Binary formats.

Some useful pull request queries:

• PRs which need reviews and are not D-Complex.
• D-Complex PRs which need reviews.

Prioritizing PRs and issues

We use Milestones to track issues and PRs that:

• Need to be merged/fixed before the next release. This is generally for extremely bad bugs i.e. UB or important functionality being broken.
• Would have higher user impact and are almost ready to be merged/fixed.

There are also two priority labels: P-Critical and P-High that can be used to find issues and PRs that need to be resolved urgently.

Closing PRs and Issues

From time to time, PRs are unsuitable to be merged in a way that cannot be readily fixed. Rather than leaving these PRs open in limbo indefinitely, they should simply be closed.

This might happen if:

1. The PR is spam or malicious.
2. The work has already been done elsewhere or is otherwise fully obsolete.
3. The PR was successfully adopted.
4. The work is particularly low quality, and the author is resistant to coaching.
5. The work adds features or abstraction of limited value, especially in a way that could easily be recreated outside of the engine.
6. The work has been sitting in review for so long and accumulated so many conflicts that it would be simpler to redo it from scratch.
7. The PR is pointlessly large, and should be broken into multiple smaller PRs for easier review.

PRs that are S-Adopt-Me should be left open, but only if they're genuinely more useful to rebase rather than simply use as a reference.

There are several paths for PRs to be closed:

1. Obviously, authors may close their own PRs for any reason at any time.
2. If a PR is clearly spam or malicious, anyone with triage rights is encouraged to close out the PR and report it to Github.
3. If the work has already been done elsewhere, adopted or otherwise obsoleted, anyone with triage rights is encouraged to close out the PR with an explanatory comment.
4. Anyone may nominate a PR for closure, by bringing it to the attention of the author and / or one of the SMEs / maintainers. Let them press the button, but this is generally well-received and helpful.
5. SMEs or maintainers may and are encouraged to unilaterally close PRs that fall into one or more of the remaining categories.
6. In the case of PRs where some members of the community (other than the author) are in favor and some are opposed, any two relevant SMEs or maintainers may act in concert to close the PR.

When closing a PR, check if it has an issue linked. If it does not, you should strongly consider creating an issue and linking the now-closed PR to help make sure the previous work can be discovered and credited.

Making changes to Bevy

Most changes don't require much "process". If your change is relatively straightforward, just do the following:

1. A community member (that's you!) creates one of the following:
   • GitHub Discussions: An informal discussion with the community. This is the place to start if you want to propose a feature or specific implementation.
   • Issue: A formal way for us to track a bug or feature. Please look for duplicates before opening a new issue and consider starting with a Discussion.
   • Pull Request (or PR for short): A request to merge code changes. This starts our "review process". You are welcome to start with a pull request, but consider starting with an Issue or Discussion for larger changes (or if you aren't certain about a design). We don't want anyone to waste their time on code that didn't have a chance to be merged! But conversely, sometimes PRs are the most efficient way to propose a change. Just use your own judgement here.
2. Other community members review and comment in an ad-hoc fashion. Active subject matter experts may be pulled into a thread using @mentions. If your PR has been quiet for a while and is ready for review, feel free to leave a message to "bump" the thread, or bring it up on Discord in an appropriate engine development channel.
3. Once they're content with the pull request (design, code quality, documentation, tests), individual reviewers leave "Approved" reviews.
4. After consensus has been reached (typically two approvals from the community or one for extremely simple changes) and CI passes, the S-Ready-For-Final-Review label is added.
5. When they find time, someone with merge rights performs a final code review and queue the PR for merging.

Complex changes

Individual contributors often lead major new features and reworks. However these changes require more design work and scrutiny. Complex changes like this tend to go through the following lifecycle:

1. A need or opportunity is identified and an issue is made, laying out the general problem.
2. As needed, this is discussed further on that issue thread, in cross-linked [GitHub Discussion] threads, or on Discord in the Engine Development channels.
3. Either a Draft Pull Request or an RFC is made. As discussed in the RFC repo, complex features need RFCs, but these can be submitted before or after prototyping work has been started.
4. If feasible, parts that work on their own (even if they're only useful once the full complex change is merged) get split out into individual PRs to make them easier to review.
5. The community as a whole helps improve the Draft PR and/or RFC, leaving comments, making suggestions, and submitting pull requests to the original branch.
6. Once the RFC is merged and/or the Draft Pull Request is transitioned out of draft mode, the normal change process outlined in the previous section can begin.

How you can help

If you've made it to this page, you're probably already convinced that Bevy is a project you'd like to see thrive. But how can you help?

No matter your experience level with Bevy or Rust or your level of commitment, there are ways to meaningfully contribute. Take a look at the sections that follow to pick a route (or five) that appeal to you.

If you ever find yourself at a loss for what to do, or in need of mentorship or advice on how to contribute to Bevy, feel free to ask in Discord and one of our more experienced community members will be happy to help.

Battle-testing Bevy

Ultimately, Bevy is a tool that's designed to help people make cool games. By using Bevy, you can help us catch bugs, prioritize new features, polish off the rough edges, and promote the project.

If you need help, don't hesitate to ask for help on GitHub Discussions, Discord, or reddit. Generally you should prefer asking questions as GitHub Discussions as they are more searchable.

When you think you've found a bug, missing documentation, or a feature that would help you make better games, please file an issue on the main bevy repo.

Do your best to search for duplicate issues, but if you're unsure, open a new issue and link to other related issues on the thread you make.

Once you've made something that you're proud of, feel free to drop a link, video, or screenshot in #showcase on Discord! If you release a game on itch.io we'd be thrilled if you tagged it with bevy.

Teaching others

Bevy is still very young, and light on documentation, tutorials, and accumulated expertise. By helping others with their issues, and teaching them about Bevy, you will naturally learn the engine and codebase in greater depth while also making our community better!

Some of the best ways to do this are:

• Answering questions on GitHub Discussions, Discord, and reddit.
• Writing tutorials, guides, and other informal documentation and sharing them on Bevy Assets.
• Streaming, writing blog posts about creating your game, and creating videos. Share these in the #devlogs channel on Discord!

Writing plugins

You can improve Bevy's ecosystem by building your own Bevy Plugins and crates.

Non-trivial, reusable functionality that works well with itself is a good candidate for a plugin. If it's closer to a snippet or design pattern, you may want to share it with the community on Discord, Reddit, or GitHub Discussions instead.

Check out our plugin guidelines for helpful tips and patterns.

Fixing bugs

Bugs in Bevy (or the associated website / book) are filed on the issue tracker using the C-Bug label.

If you're looking for an easy place to start, take a look at the D-Good-First-Issue label, and feel free to ask questions on that issue's thread in question or on Discord. You don't need anyone's permission to try fixing a bug or adding a simple feature, but stating that you'd like to tackle an issue can be helpful to avoid duplicated work.

When you make a pull request that fixes an issue, include a line that says Fixes #X (or "Closes"), where X is the issue number. This will cause the issue in question to be closed when your PR is merged.

General improvements to code quality are also welcome! Bevy can always be safer, better tested, and more idiomatic.

Writing docs

Like every other large, rapidly developing open source library you've ever used, Bevy's documentation can always use improvement. This is incredibly valuable, easily distributed work, but requires a bit of guidance:

• Inaccurate documentation is worse than no documentation: prioritize fixing broken docs.
• Bevy is remarkably unstable: before tackling a new major documentation project, check in with the community on Discord or GitHub (making an issue about specific missing docs is a great way to plan) about the stability of that feature and upcoming plans to save yourself heartache.
• Code documentation (doc examples and in the examples folder) is easier to maintain because the compiler will tell us when it breaks.
• Inline documentation should be technical and to the point. Link relevant examples or other explanations if broader context is useful.
• The Bevy book is hosted on the bevy-website repo and targeted towards beginners who are just getting to know Bevy (and perhaps Rust!).
• Accepted RFCs are not documentation: they serve only as a record of accepted decisions.

docs.rs is built from out of the last release's documentation, which is written right in-line directly above the code it documents. To view the current docs on main before you contribute, clone the bevy repo, and run cargo doc --open or go to dev-docs.bevyengine.org, which has the latest API reference built from the repo on every commit made to the main branch.

Writing examples

Most examples in Bevy aim to clearly demonstrate a single feature, group of closely related small features, or show how to accomplish a particular task (such as asset loading, creating a custom shader or testing your app). In rare cases, creating new "game" examples is justified in order to demonstrate new features that open a complex class of functionality in a way that's hard to demonstrate in isolation or requires additional integration testing.

Examples in Bevy should be:

1. Working: They must compile and run, and any introduced errors in them should be obvious (through tests, simple results or clearly displayed behavior).
2. Clear: They must use descriptive variable names, be formatted, and be appropriately commented. Try your best to showcase best practices when it doesn't obscure the point of the example.
3. Relevant: They should explain, through comments or variable names, what they do and how this can be useful to a game developer.
4. Minimal: They should be no larger or complex than is needed to meet the goals of the example.

When you add a new example, be sure to update examples/README.md with the new example and add it to the root Cargo.toml file. Run cargo run -p build-templated-pages -- build-example-page to do this automatically. Use a generous sprinkling of keywords in your description: these are commonly used to search for a specific example. See the example style guide to help make sure the style of your example matches what we're already using.

More complex demonstrations of functionality are also welcome, but these should be submitted to bevy-assets.

Reviewing others' work

With the sheer volume of activity in Bevy's community, reviewing others work with the aim of improving it is one of the most valuable things you can do. You don't need to be an Elder Rustacean to be useful here: anyone can catch missing tests, unclear docs, logic errors, and so on. If you have specific skills (e.g. advanced familiarity with unsafe code, rendering knowledge or web development experience) or personal experience with a problem, try to prioritize those areas to ensure we can get appropriate expertise where we need it.

When you find (or make) a PR that you don't feel comfortable reviewing, but you can think of someone who does, consider using Github's "Request review" functionality (in the top-right of the PR screen) to bring the work to their attention. If they're not a Bevy Org member, you'll need to ping them in the thread directly: that's fine too! Almost everyone working on Bevy is a volunteer: this should be treated as a gentle nudge, rather than an assignment of work. Consider checking the Git history for appropriate reviewers, or ask on Discord for suggestions.

Focus on giving constructive, actionable feedback that results in real improvements to code quality or end-user experience. If you don't understand why an approach was taken, please ask!

Provide actual code suggestions when that is helpful. Small changes work well as comments or in-line suggestions on specific lines of codes. Larger changes deserve a comment in the main thread, or a pull request to the original author's branch (but please mention that you've made one). When in doubt about a matter of architectural philosophy, refer back to What we're trying to build for guidance.

Once you're happy with the work and feel you're reasonably qualified to assess quality in this particular area, leave your Approved review on the PR. If you're new to GitHub, check out the Pull Request Review documentation. Anyone can and should leave reviews ... no special permissions are required!

It's okay to leave an approval even if you aren't 100% confident on all areas of the PR: just be sure to note your limitations. When maintainers are evaluating the PR to be merged, they'll make sure that there's good coverage on all of the critical areas. If you can only check that the math is correct, and another reviewer can check everything but the math, we're in good shape!

Similarly, if there are areas that would be good to fix but aren't severe, please consider leaving an approval. The author can address them immediately, or spin it out into follow-up issues or PRs. Large PRs are much more draining for both reviewers and authors, so try to push for a smaller scope with clearly tracked follow-ups.

There are three main places you can check for things to review:

1. Pull requests which are ready and in need of more reviews on bevy.
2. Pull requests on bevy and the bevy-website repos.
3. RFCs, which need extensive thoughtful community input on their design.

Not even our Project Leads and Maintainers are exempt from reviews and RFCs! By giving feedback on this work (and related supporting work), you can help us make sure our releases are both high-quality and timely.

Finally, if nothing brings you more satisfaction than seeing every last issue labeled and all resolved issues closed, feel free to message the Project Lead (currently @cart) for a Bevy org role to help us keep things tidy. As discussed in our Bevy Organization doc, this role only requires good faith and a basic understanding of our development process.

How to adopt pull requests

Occasionally authors of pull requests get busy or become unresponsive, or project members fail to reply in a timely manner. This is a natural part of any open source project. To avoid blocking these efforts, these pull requests may be adopted, where another contributor creates a new pull request with the same content. If there is an old pull request that is without updates, comment to the organization whether it is appropriate to add the S-Adopt-Me label, to indicate that it can be adopted. If you plan on adopting a PR yourself, you can also leave a comment on the PR asking the author if they plan on returning. If the author gives permission or simply doesn't respond after a few days, then it can be adopted. This may sometimes even skip the labeling process since at that point the PR has been adopted by you.

With this label added, it's best practice to fork the original author's branch. This ensures that they still get credit for working on it and that the commit history is retained. When the new pull request is ready, it should reference the original PR in the description. Then notify org members to close the original.

• For example, you can reference the original PR by adding the following to your PR description:

Adopted #number-original-pull-request

Contributing code

Bevy is actively open to code contributions from community members. If you're new to Bevy, here's the workflow we use:

1. Fork the bevyengine/bevy repository on GitHub. You'll need to create a GitHub account if you don't have one already.
2. Make your changes in a local clone of your fork, typically in its own new branch.
   1. Try to split your work into separate commits, each with a distinct purpose. Be particularly mindful of this when responding to reviews so it's easy to see what's changed.
   2. Tip: You can set up a global .gitignore file to exclude your operating system/text editor's special/temporary files. (e.g. .DS_Store, thumbs.db, *~, *.swp or *.swo) This allows us to keep the .gitignore file in the repo uncluttered.
3. To test CI validations locally, run the cargo run -p ci command. This will run most checks that happen in CI, but can take some time. You can also run sub-commands to iterate faster depending on what you're contributing:
   • cargo run -p ci -- lints - to run formatting and clippy.
   • cargo run -p ci -- test - to run tests.
   • cargo run -p ci -- doc - to run doc tests and doc checks.
   • cargo run -p ci -- compile - to check that everything that must compile still does (examples and benches), and that some that shouldn't still don't (crates/bevy_ecs_compile_fail_tests).
   • to get more information on commands available and what is run, check the tools/ci crate.
4. When working with Markdown (.md) files, Bevy's CI will check markdown files (like this one) using markdownlint. To locally lint your files using the same workflow as our CI:
   1. Install markdownlint-cli.
   2. Run markdownlint -f -c .github/linters/.markdown-lint.yml . in the root directory of the Bevy project.
5. When working with Toml (.toml) files, Bevy's CI will check toml files using taplo: taplo fmt --check --diff
   1. If you use VSCode, install Even better toml and format your files.
   2. If you want to use the cli tool, install taplo-cli and run taplo fmt --check --diff to check for the formatting. Fix any issues by running taplo fmt in the root directory of the Bevy project.
6. Check for typos. Bevy's CI will check for them using typos.
   1. If you use VSCode, install Typos Spell Checker.
   2. You can also use the cli tool. Install typos-cli and run typos to check for typos, and fix them by running typos -w.
7. Push your changes to your fork on Github and open a Pull Request.
8. Respond to any CI failures or review feedback. While CI failures must be fixed before we can merge your PR, you do not need to agree with all feedback from your reviews, merely acknowledge that it was given. If you cannot come to an agreement, leave the thread open and defer to a Maintainer or Project Lead's final judgement.
9. When your PR is ready to merge, a Maintainer or Project Lead will review it and suggest final changes. If those changes are minimal they may even apply them directly to speed up merging.

If you end up adding a new official Bevy crate to the bevy repo:

1. Add the new crate to the ./tools/publish.sh file.
2. Check if a new cargo feature was added, update cargo_features.md as needed.

When contributing, please:

• Try to loosely follow the workflow in Making changes to Bevy.
• Consult the style guide to help keep our code base tidy.
• Explain what you're doing and why.
• Document new code with doc comments.
• Include clear, simple tests.
• Add or improve the examples when adding new user-facing functionality.
• Break work into digestible chunks.
• Ask for any help that you need!

Your first PR will be merged in no time!

No matter how you're helping: thanks for contributing to Bevy!