1.x: scan with an initial factory callback

[akarnokd]

This PR adds an overload to scan which allows specifying an initial value factory. The initial value factory is invoked for each incoming Subscriber individually and provides a non-shared starting point for the scan operation.

I propose the scan(Func2 accumulator, Func0 initialFactory) parameter order because some Java 8+ compilers can't properly disambiguate with scan(R initialValue, Func2 accumulator), thinking that R can be made Func0.

Related: #3667

[abersnaze]

Why add the overload if it can be done with defer and scan then?

[akarnokd]

Yep, it can be done. A decision has to be made if this convenience should make it into the library or not, and notify @NiteshKant the outcome in #3667.

[davidmoten]

I'd like to see this overload in, thanks.

[zsxwing]

I'm not strong against adding it but not like the idea of the new parameter order. I would give +1 if giving it a better name instead of changing the parameter order. By the way, there is already a OperatorScan(Func0<R> initialValueFactory, Func2<R, ? super T, R> accumulator) for this overload so you don't need to use defer.

Probably @benjchristensen should take a look since you removed it before.

[benjchristensen]

It was removed due to ambiguity as shown in #1881 (comment)

As long as that issue is not recreated, I have nothing against the functionality pursued here. I too don't like solving the overload ambiguity by moving parameter order around. But there is no good solution, it's either that or a new name.

[akarnokd]

As an alternative, I suggest extending scan's javadoc with the example of the defer trick.

[akarnokd]

The argument order bothers me as well. I'll close this and post a javadoc update instead.