Arrows - extended pipe operators (e.g. pipe to any argument position)

Provides OK-piping (similar to libraries like ok) and adds the ability to pipe into any argument position (similar to magritte) of the following function (or nested function). By inserting ... where you would like the value to be inserted, Arrows will override where it is placed.

This library was developed as part of the Bonfire project, which has some open bounties for any help improving performance: Bonfire Networks: An open-source framework for building federated digital spaces

So instead of only being limited to:

iex> 4 |> div(2)
2

You can also do:

iex> 4 |> div(2, ...)
0

Or even:

iex> 4 |> div(2, min(..., 1))
2

Links:

4 Likes

Ah, the three dots indicate where the object is put. Now I understand.

1 Like

That’s right! The documentation do with a rewrite… (have now edited the description to mention that)

1 Like

It is quite nice and readable :+1: However, it is also not so hard to replace with Kernel.then/2

could become

iex> 4 |> then(&div(2, &1))
0
4 |> then(&div(2, min(&1, 1)))
2
1 Like

Yeah this was created before then was introduced, so probably not needed if you’re used to that syntax.

I see that the main difference between this and Magritte is that it do not have my limitations. Actually I have intentionally added these limitations to avoid things like:

foo
|> bar(fn a ->
  a
  |> baz(2137, ...) # what value should be inserted there?
end)

Also, when you want to use ... twice within single pipe there is question whether there should be 2 separate parent pipes or single one. Aka whether this code:

:rand.uniform()
|> Kernel.==(..., ...)

Should always return true?

I probably should have describe reasoning better in Magritte documentation.

1 Like

Yeah that makes sense, though FWIW we’ve been happily using it without running into confusion on these topics so far.

In our case it pipes the same thing twice in that scenario:

    assert true = :rand.uniform()
    |> Kernel.==(..., ...)

Thanks for making this package! The ~> looks like a nice alternative to with. Any gotchas or downsides to be aware of? Looks like it could always be used in place of the |> or are there still scenarios where you’d want to use |>?

Also, would you mind documenting the other functions in more detail in the hex doc?

with serves a very different purpose. <- is a match operator much like = with the difference that it returns the value of the expression if there is no match instead of raising. with is about running its do block only when all of its clauses match. This package is just about being able to pipe into other positional arguments.

Oops, sorry I misread what ~> was. There is another library like this that uses ~>. My bad.

I’ve added some more docs and examples: Arrows — arrows v0.2.1

Regarding ~> I tend to use it mostly for short pipes where with may feel verbose, but for longer ones I prefer to be more explicit. In part because of this gotcha:

      # Note that the following would pass :error to `String.pad_leading` breaking our pipe: 
      # :error ~> Integer.to_string() |> String.pad_leading(2, "0")
      :error

      # Instead we want to do:
      iex> :error ~> Integer.to_string() ~> String.pad_leading(2, "0")
      :error
1 Like