[RFC] Add overloads for cache() that take a size

[mttkay]

We find the cache method to be quite useful, however, it's quite memory inefficient for caching single items since even when passing a capacity of 1, it will create an unbounded replay subject rather than one that actually has size 1.

Memory pressure being a major concern on Android, how would everyone feel about adding more overloads/alternatives to cache that take a size, not just a capacity? e.g.:

Observable.cacheWithSize(1)

which would call ReplaySubject.createWithSize internally, using a BoundedState?

[mttkay]

Actually, nevermind, I think I misread ReplaySubject. This only seems to have an effect on the eviction strategy, so using an "unbounded" replay subject with capacity 1 should be what we want?

The terminology is really confusing here.

[benjchristensen]

We added the cache(int) overload in 0.20.0-RC3 to provide a hint to the underlying ReplaySubject to optimize memory allocation.

The pull request was: #1303

The idea is that it should allocate only the capacity up front but can grow if more items are received.

We could add a cacheWithSize(int) and/or cacheLast() variant that uses ReplaySubject.createWithSize: https://github.com/Netflix/RxJava/blob/master/rxjava-core/src/main/java/rx/subjects/ReplaySubject.java#L157 if that is shown to be more efficient that the more flexible cache which grows if necessary.

A cacheLast() variant would be similar to doing takeLast().cache(1).

Currently the idea for cache is that if only 1 item is wanted, take(1).cache(1) should efficiently allocate memory.

[benjchristensen]

@mttkay Is the available functionality sufficient or would you like to discuss this further?

[mttkay]

I think it's sufficient, thanks. You can close this out unless someone else feels strong about it.