wit-breaking

Detects breaking changes between two versions of a WebAssembly Component Model WIT interface — the equivalent of buf breaking (protobuf) or oasdiff (OpenAPI), for WIT.

Background, for anyone new to this

If you don't work with WebAssembly components day to day, here's the context that makes the rest of this README make sense.

What WIT is. The WebAssembly Component Model lets you compile a library in Rust, publish it as a .wasm component, and have someone else load and call it from Go, Python, or JavaScript — without either side knowing what language the other is written in. For that to work, both sides need to agree on a contract: what functions exist, what arguments they take, what they return. That contract is written in WIT (Wasm Interface Type), a small interface-description language. A WIT interface block looks like this:

interface key-value {
  get: func(key: string) -> option<string>;
  set: func(key: string, value: string);
}

Any component that exports this key-value interface promises those two functions exist with those exact signatures. Any component that imports it is trusting that promise.

Why "breaking changes" are a real problem here. Components get rebuilt and republished over time, same as any library. If a new version of a component silently changes set's signature — adds a required parameter, changes a return type — every other component that imports the old key-value interface now fails to compose with it, possibly not until someone tries to link them together and gets a confusing error far from the actual change. This is exactly the problem protobuf and OpenAPI solved years ago with dedicated tools (buf breaking, oasdiff) that diff two versions of a schema and tell you, in plain language, what would break. WIT didn't have an equivalent — this project is that tool for WIT.

What it actually does

You give it two WIT documents — normally "this interface before my change" and "this interface after my change" — and it walks every interface and every function in both, and classifies each difference:

What changed	Verdict	Why
A function was removed	Breaking	Anyone calling it now gets a link/runtime error.
A function's parameters changed	Breaking	Existing callers pass the old argument shape.
A function's return type changed	Breaking	Existing callers expect the old shape back.
A whole interface was removed	Breaking	Everything that imported it breaks.
A function was added	Safe	Nobody was calling it before, so nobody can be broken by it existing now.
A whole interface was added	Safe	Same reasoning, one level up.

npx wit-breaking before.wit after.wit

BREAKING  key-value.set: function "set" parameters changed: "(key: string, value: string)" -> "(key: string, value: string, ttl-seconds: u32)"
BREAKING  key-value.delete: function "delete" was removed
safe      key-value.list-keys: function "list-keys" was added

2 breaking, 1 safe.

Exit code 1 if anything breaking is found, 0 otherwise. Where do the two .wit files come from? If you already hand-write WIT, you have them. If you're diffing two versions of a compiled .wasm component, extract each one's interface with the official Bytecode Alliance tool: wasm-tools component wit my-component.wasm > interface.wit, then diff the two extracted files with this tool.

Use it as a GitHub Action

The fastest way to get real value from this: add it to your CI so a breaking WIT change fails the build automatically, the same way a broken test would.

- name: Extract WIT from both builds
  run: |
    wasm-tools component wit ./before/my-component.wasm > before.wit
    wasm-tools component wit ./after/my-component.wasm > after.wit

- name: Check for breaking interface changes
  uses: jayblast-spec/wit-breaking@main
  with:
    before: before.wit
    after: after.wit
    # fail-on-breaking: 'false'   # uncomment to only annotate, never fail the build

The action exposes two outputs (breaking — "true"/"false", and summary — the full text report) if you want to post the result somewhere yourself instead of just failing the step.

Honest design tradeoffs

Every choice below was made deliberately, and each one trades some completeness for something else that mattered more. Worth reading before you trust this in CI.

It reads WIT text, not .wasm binaries — on purpose, not because it was easier. The obvious way to extract a component's interface is through the Bytecode Alliance's own JS tooling (@bytecodealliance/jco). I actually installed it and inspected its API while building this, and its dependency tree currently pulls in a critical-severity transitive vulnerability — a Zip Slip archive-extraction bug, several layers down, used only at that toolchain's own build time, but still not something I'm willing to add to your node_modules for a tool meant to sit in your CI pipeline. Since extracting WIT from a binary already has a canonical, official, actively-maintained tool (wasm-tools component wit), duplicating that logic here would only add risk for no real benefit. So wit-breaking takes WIT text as input and has zero runtime dependencies — pipe the official extractor's output into it.

It diffs signatures textually, not a full type system. If a function's parameter list changes at all — a type renamed, reordered, widened, narrowed — this tool flags it as breaking, full stop. It does not attempt to understand, for instance, that widening a variant to accept one more case is actually backward-compatible while narrowing it isn't. Building that correctly means implementing WIT's real type system, which is a much bigger and riskier undertaking than a v1 warrants. The practical effect: this tool will occasionally flag a change as breaking when a deeper analysis would call it safe, but it will never miss a real break. For a CI gate, that's the right side to err on.

Install

npm install --save-dev wit-breaking

Library usage

import { parseWit, diffWit } from "wit-breaking";

const result = diffWit(parseWit(beforeSource), parseWit(afterSource));
if (result.breaking) {
  for (const change of result.changes.filter((c) => c.severity === "breaking")) {
    console.error(change.description);
  }
  process.exit(1);
}

Run the annotated demo:

npx tsx examples/demo.ts

Not in v1, and why

• world import/export diffing. A WIT world bundles interfaces together into what a whole component imports/exports; interfaces are the primary contract surface underneath that, so they came first. Worlds are a real, planned extension — not something silently half-supported here today.
• Package version / semver policy enforcement. This tool tells you what changed. Deciding what your version-numbering scheme should do in response (bump major, block the release, just warn) is a policy decision specific to your project, not something a diffing tool should impose.

Development

npm install
npm run typecheck
npm test
npm run build

License

MIT

Part of a series

This is one of 8 independent open-source infrastructure tools, each built against a real, researched gap:

• embedguard -- catches silent embedding-model swaps under a vector index
• agora -- propose/vote/veto/quorum for multi-agent systems
• mcp-versionbridge -- bridges old and new MCP protocol handshakes
• yjs-lens -- verifies Yjs convergence and summarizes causal history
• gatewayproof -- conformance-tests OpenAI-compatible LLM gateways
• automerge-lens -- explains Automerge merge conflicts and verifies convergence
• pglite-diff -- diffs migration query results on isolated Postgres

---

Built by ArkNet Digital.