ReplaySubject windowTime behavior is maybe misinterpreted

[hansschenker]

Issue Summary

There is a potential mismatch between the documented and implemented behavior of ReplaySubject's windowTime parameter and the intuitive expectation set by Erik Meijer's original definition of reactive subjects.

see Erik Meijer describing ReplaySubject Behavior: https://youtu.be/WKore-AkisY?si=0THPHML0KjGY6iJK&t=1687

Expected Behavior (Erik Meijer / Intuitive)

• bufferSize (count): Should keep the last n values.
• windowTime: Should keep all values emitted within the last windowTime milliseconds when a new subscriber subscribes.

Actual RxJS Behavior

• bufferSize works as expected (last n values).
• windowTime keeps each value in the buffer for up to windowTime ms after it was emitted. Buffer is trimmed only on emission or subscription.
• This means that a new subscriber may not receive all values emitted in the last windowTime ms, but only those whose expiration (emitTime + windowTime) is after the subscription time and that haven't already been trimmed.

Example

Suppose values are emitted at t=0, t=1000, t=2000, t=3000, and windowTime=2000. If a subscriber arrives at t=3500:

• Intuitive expectation: Subscriber should get values emitted at t=2000 and t=3000 (those within the last 2000 ms).
• RxJS implementation: Subscriber gets values still in the buffer whose expiration is after t=3500. If buffer hasn't been trimmed since last emission, expired values may remain until next emission/subscription triggers trimming.

Request

• Please clarify the documentation for ReplaySubject and its windowTime parameter to explicitly state this behavior.
• Consider referencing the difference from the intuitive/Erik Meijer definition so users know what to expect.
• Optionally, provide an example in the docs.

---

Thanks for considering this clarification to help future users avoid