Add pMapIterable

[Richienb]

Main differences from pMap():

• Skipped values are simply never yielded
• stopOnError is not available because this can now be done by the user themselves
• signal is also not available for the same reason
• Values that are yielded from the iterator, are immediately passed to the mapper and the returned values are yielded from the function in the order that they originally were yielded

If you're ok with this, I'll add docs and tests

Fixes sindresorhus/promise-fun#21

[sindresorhus]

What if nextItem throws? It will cause an unhandled rejection.

[sindresorhus]

Bump :)

[sindresorhus]

What's left to do here?

[Richienb]

What's left to do here?

docs

[sindresorhus]

Why do we need the separate backpressure option?

[Richienb]

Why do we need the separate backpressure option?

Values returned by the async iterator can only be consumed one at a time. However, we can run the mapper function concurrency times in parallel. This means there will end up being a backlog of values that are waiting to be consumed. We limit the size of the backlog using backpressure such that we don't allow further calls to the mapper function if doing so would cause it to exceed the backpressure.

[Richienb]

Might be a good idea to try re-implement pMap using pMapIterable to simplify the code and see where its holes are.

[sindresorhus]

Values returned by the async iterator can only be consumed one at a time. However, we can run the mapper function concurrency times in parallel. This means there will end up being a backlog of values that are waiting to be consumed. We limit the size of the backlog using backpressure such that we don't allow further calls to the mapper function if doing so would cause it to exceed the backpressure.

👍

[Richienb]

What should happen when the mapper function returns an error?
pMap simply throws the error because it only returns the values when everything finishes. However, pMapIterable streams returned values. Should it just prevent future values down the iterable from running and let the ones before it finish?

[sindresorhus]

An error should generally make things stop as soon as possible, so I don't think it should even let the current ones finish.

[Richienb]

Ok I guess I'll throw out values that arrived out of order after an error.

[Richienb]

@sindresorhus I did stuff

[sindresorhus]

Nice! Big improvement.

[sindresorhus]

Nice work 🙏