[ruby-core:99270] [Ruby master Feature#17016] Enumerable#scan_left

[finch.parker@gmail.com]

Issue #17016 has been updated by parker (Parker Finch).


matz (Yukihiro Matsumoto) wrote in #note-17:
> I don't see any real-world use-case for `scan_left`. Is there any?


I think there are real-world use cases!

Would you consider converting a history of transactions into a history of account balances a valid use-case? That can be done easily with a scan. For example, if you have `transactions = [100, -200, 200]` then you can find the history of account balances with `transactions.scan_left(0, &:+) # => [0, 100, -100, 100]`.

I have described our current use case of `scan_left` [here](https://medium.com/building-panorama-education/scan-left-a-lazy-incremental-alternative-to-inject-f6e946f74c00). We use an event-sourced architecture where we scan over a lazy stream of events, building up the state as we go. It is important for our use case that we are able to access past states in addition to the final state.

[This post](https://medium.com/beauty-date-stories/algorithms-how-prefix-sums-can-help-improving-operations-over-arrays-b1f8e8141668) shows how it can make repeated subarray operations more efficient -- the example there is that if you have an array of values representing changes over time periods, then you can easily aggregate those into changes over different time periods using a `scan`. This post does not use `scan`, but instead has a workaround because `scan` doesn't exist:
```ruby
sums = [0]
(1..gains.length).each do |i|
  sums[i] = sums[i - 1] + gains[i - 1]
end
```
could, if `scan` was introduced, be replaced with:
```ruby
sums = gains.scan_left(0, &:+)
```

Do those use cases seem sufficient?

---

> In addition, the term `scan` does not seem to describe the behavior of keeping the past sequence of accumulation.
> Although I know the name from the Haskell community, I don't agree with the name.

I agree the name is not ideal. It is easy to get it confused with the idea of a string scanner. As you mentioned, `scan` is the name used by Haskell, but it is also used by [Scala](https://www.scala-lang.org/api/current/scala/Array.html), [Rust](https://doc.rust-lang.org/std/iter/trait.Iterator.html#method.scan), and [C++](https://en.cppreference.com/w/cpp/algorithm/inclusive_scan). So it is a widely-used term, even if it's not the best one, which makes me think that it might be good to use it.

Alternatively, what do you think about the name `accumulate`, which [Wolfram uses](https://reference.wolfram.com/language/ref/Accumulate.html)? I think it gets the idea across better than `scan`.

Are there other names you think should be considered?

----------------------------------------
Feature #17016: Enumerable#scan_left
https://bugs.ruby-lang.org/issues/17016#change-86657

* Author: parker (Parker Finch)
* Status: Open
* Priority: Normal
----------------------------------------
## Proposal

Add a `#scan_left` method to `Enumerable`.

(The name "scan_left" is based on Scala's scanLeft and Haskell's scanl. It seems like "scan_left" would be a ruby-ish name for  this concept, but I'm curious if there are other thoughts on naming here!)

## Background

`#scan_left` is similar to `#inject`, but it accumulates the partial results that are computed. As a comparison:
```
[1, 2, 3].inject(0, &:+) => 6
[1, 2, 3].scan_left(0, &:+) => [0, 1, 3, 6]
```

Notably, the `scan_left` operation can be done lazily since it doesn't require processing the entire collection before computing a value.

I recently described `#scan_left`, and its relationship to `#inject`, more thoroughly in [this blog post](https://medium.com/building-panorama-education/scan-left-a-lazy-incremental-alternative-to-inject-f6e946f74c00).

## Reasoning
We heavily rely on the scan operation. We use an [event-sourcing](https://martinfowler.com/eaaDev/EventSourcing.html) pattern, which means that we are scanning over individual "events" and building up the corresponding state. We rely on the history of states and need to do this lazily (we stream events because they cannot fit in memory). Thus the scan operation is much more applicable than the inject operation.

We suspect that there are many applications that could leverage the scan operation. [This question](https://stackoverflow.com/questions/1475808/cumulative-array-sum-in-ruby) would be more easily answered by `#scan_left`. It is a natural fit for any application that needs to store the incrementally-computed values of an `#inject`, and a requirement for an application that needs to use `#inject` while maintaining laziness.

## Implementation
There is a Ruby implementation of this functionality [here](https://github.com/panorama-ed/scan_left/) and an implementation in C [here](https://github.com/ruby/ruby/pull/3078).

## Counterarguments
Introducing a new public method is committing to maintenance going forward and expands the size of the Ruby codebase -- it should not be done lightly. I think that providing the functionality here is worth the tradeoff, but I understand any hesitation to add yet more to Ruby!

---Files--------------------------------
scan_left_example.rb (2.93 KB)


-- 
https://bugs.ruby-lang.org/