esm-merge-coverage

Diff against the merge base to list incoming files, then run the repo's cjs-to-esm codemod on server sources leaf-to-root, followed by the migrations codemod and import-convention normalizer. Finish by running the lint, build, and dist-resolution verification gates.

A git merge copies source verbatim and never runs codemods, lint, or build steps, so pre-ESM files with require and module.exports land unconverted. They only surface later as type-check failures, ERR_MODULE_NOT_FOUND, or initialization errors at runtime.

ESM static hoisting exposes cycles that CJS lazy require hid, throwing Cannot access before initialization at boot. The fix is factory DI: services and models must receive the Crowi instance as an argument instead of importing the Crowi class, or split shared types into an interfaces file.

Migrations must use named exports, export async function up and down, because migrate-mongo reads migration.up and migration.down off the loaded module. export default leaves migration.up undefined and module.exports throws under type module.

Keep a config as .cjs only when it is consumed by a CJS-only CLI such as migrate-mongo, paired with a hand-written .d.cts sibling and .cjs specifiers in importers. Configs consumed by ESM-capable tooling should use .mjs instead.

Do not use it for net-new code authored on an ESM branch, which is already covered by the normal lint gates and ESM authoring rules. It is specifically for imported code from branches that predate the v8 ESM migration.