Bench-node should load the same entry points in all workers

[jdmarshall]

Here is an attempt by me to build worker-based tests with the current implementation, modulo the current issue with async tests:

cobblers-children/faceoff@main...workers

That's a lot of heavy lifting. Just way too much for anyone else to want to do (I'm willing because I only have to do it once and lots of people will hopefully get to use it as-is).

I think it would be much simpler if bench-node ran all of the new Suite and suite.add() functions checking whether we are in a Worker thread or not and making different decisions when it is.

Including having suite.run() wait for a message and then only running specific tests as the result of that call.

This will avoid all problems with tests trying to keep any state outside of the function under test, (example of a test that broke:
cobblers-children/faceoff@main...workers#diff-0d458569bd1c86b86a54927f40fa1fc49fa9862cc6b0a10adbfcd780e9286327
)

and drastically simplify the setup and teardown workflow, which is just gonna be harsh to deal with. Things might get a little weird for large test suites (like prom-client) but it's already pretty weird as things stand.

[RafaelGSS]

I think it would be much simpler if bench-node ran all of the new Suite and suite.add() functions checking whether we are in a Worker thread or not and making different decisions when it is.

I'm not sure this would work as you expect. I also believe this might be a bad UI/UX for non-familiarised Node.js users.

A more friendly approach would be passing the benchmark module as string and loading it in a worker thread. For instance:

const { Suite } = require('bench-node')
const path = require('node:path')

const suite = new Suite({ useWorkers: true })

suite.add('fn1', path.join('./my-bench.js'))

suite.run()

module.exports = (fn) => {
  // ...
}

---

I suggest opening an new issue as a Worker Thread roadmap, listing the current limitations and then we can iterate/discuss on it.

[jdmarshall]

Yeah, I thought about that as well. That might be the best that can be managed.

And/or, if the worker-runner used a file, any tests that are registered as functions could be put into a temp file and sent as a path, which would facilitate setting breakpoints in the transferred code if it calls out.