@efffrida/vitest-pool

A vitest pool that runs test files inside a frida agent, so tests execute in the target process instead of in node.

Example usage

import { createFridaPool } from "@efffrida/vitest-pool";
import { defineConfig } from "vitest/config";

export default defineConfig({
    test: {
        include: ["test/**/*.test.ts"],
        pool: createFridaPool({
            device: { connection: "local" },
            attach: { spawn: ["/usr/bin/sleep", "infinity"] },
        }),
    },
});

Each worker writes an agent into a temporary .vitest-frida-pool-* directory under the project root, compiles it with frida's compiler, and loads it into the attached session. Worker initialization is serialized, so only one agent is compiled and attached at a time.

Vitest patch

The agent imports vitest's test runner (startTests and collectTests from vitest/internal/browser), which lives in vitest/dist/chunks/run.BPv-BBRv.js. That chunk contains the regex /\$([$\p{ID_Continue}.]+)/gu used to interpolate test.each titles, and frida's JS runtime can't parse unicode property escapes (see frida/frida-gum#1195), so loading the agent fails with a syntax error.

patches/[email protected] replaces it with /\$([$\w.]+)/g. The only behavior change is that $name interpolation in test.each titles matches ascii identifiers only.

The patch targets a content hashed chunk name, so it has to be regenerated whenever vitest is bumped, even if the regex itself is unchanged. Run pnpm patch vitest@<version>, apply the same change to whichever run.*.js chunk now contains the regex, then pnpm patch-commit and update patchedDependencies in pnpm-workspace.yaml. Projects outside this repository that use the pool need to apply the same patch.