Fix in-process race condition

[davidmarkclements]

This is related to #23 but it's doesn't solve the multiprocess race condition.

This might have only started occurring in later node versions, but when calling get-port a lot sometimes the same port will be returned and then port already bound error occurs when trying to bind to it.

You can extend this to solve #23 by replacing Set with an ondisk store (maybe leveldb?).

The reason for using an old/young sets approach is to avoid setting one interval per port (which would be overkill imo) while also ensuring that a port isn't forgotten for at least the sweep time, but up to double the sweep time.

[sindresorhus]

Thanks for working on this 🙌

Can you add a test?

[davidmarkclements]

thanks @sindresorhus for the feedback, should be all addressed now 👍

oops - except for test, incoming

[davidmarkclements]

@sindresorhus ok test added, also realised that we only need one setInterval and it's lazily initialised

[sindresorhus]

The code looks good. Can you mention the behavior in the readme? That ports are locked up, how long, why, and how it helps prevent race-conditions.

[davidmarkclements]

@sindresorhus description of behaviour has been added to the readme 👍

[sindresorhus]

Excellent. Thanks for working on this :)

[davidmarkclements]

thanks @sindresorhus !