[Proposal] Add Function.const/2

79 views
Skip to first unread message

Jesse Claven

unread,
Jan 29, 2020, 5:01:45 PM1/29/20
to elixir-lang-core
Hey everyone!

I propose adding `Function.const/2` as a function which for argument `x`, would always return `x`.

```
@spec const(any(), any()) :: any()
def const(_original_value, new_value), do: new_value
```

This is somewhat similar to the new `Function.identity/1` in that it may seem like we've easily lived without it, but it does afford some more concise/semantic ways to represent a common pattern.

```
fn _ -> something_else() end
```

This would become:

```
const(something_else())
```

```
Enum.map([0,1,2,3], &Function.const/2)
```

I'm new to Elixir, so I'm not sure of the weight that the name `const` would carry, so there could be a more acceptable name.

If accepted, I'd be happy to create a PR!

Equivalents in other languages:

- Haskell: https://hackage.haskell.org/package/base-4.12.0.0/docs/Data-Function.html#v:const
- PureScript: https://pursuit.purescript.org/packages/purescript-const/4.1.0/docs/Data.Const
- Elm: https://package.elm-lang.org/packages/elm/core/latest/Basics#always
- Scala: https://www.scala-lang.org/api/current/scala/Function$.html#const[T,U](x:T)(y:U):T

Thanks for your time.

Amos King - Binary Noggin

unread,
Jan 29, 2020, 5:43:51 PM1/29/20
to elixir-l...@googlegroups.com
Jesse,

I see where you are going, but Elixir doesn't have currying. For instance, the last example you gave will return an error.

You could make it work by changing it a bit.

`@spec const(a) :: fn(any() -> a) when a: var`

So, const would return an fn instead of taking two arguments. I don't know how useful that would be in the long run. I know it is common in function programming, but I don't see it being extremely helpful in Elixir.

I'm ready to be convinced.

Cheers,

Amos King
CEO
Binary Noggin
=======================================================
I welcome VSRE emails. Learn more at http://vsre.info/
=======================================================


--
You received this message because you are subscribed to the Google Groups "elixir-lang-core" group.
To unsubscribe from this group and stop receiving emails from it, send an email to elixir-lang-co...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/elixir-lang-core/576949a3-93e2-4117-b1f1-ab4621f10e88%40googlegroups.com.

Jesse Claven

unread,
Jan 30, 2020, 4:24:32 AM1/30/20
to elixir-lang-core
Hey Amos,

Ah that's right. Every now and then I'm reminded that Elixir doesn't support currying when I try to do something haha Your suggestion makes sense.

Would there be any downsides to having it, even if it isn't the most incredible function? Again to call on the `Function.identity/1`, and it's discussion, it's useful when it's useful for concise and semantic code.
To unsubscribe from this group and stop receiving emails from it, send an email to elixir-l...@googlegroups.com.

Wiebe-Marten Wijnja

unread,
Jan 30, 2020, 6:20:58 AM1/30/20
to elixir-l...@googlegroups.com

The reason `Function.identity/1` was added after it was requested many times previously, was that at some point everyone agreed that it would improve Elixir's documentation, because it is easier to search for than `&(&1)`.

The `const` pattern is much less wide-spread. In e.g. Haskell it sees some use in places where it is the single (or at least by far the simplest) way to make things typecheck in a pure functional environment.

In Elixir, I suspect that it would be used much less commonly. The fact that our functions contain statements that are executed from top to bottom and rebindable variable names means that we are even less likely to use it anywhere.

As such, I don't think `const` is a function that is important enough to include in the standard library.


Of course, I'm very much open to evidence of the contrary ^_^. Do you have any example code of where you'd see `const` being useful?


~Marten/Qqwy

To unsubscribe from this group and stop receiving emails from it, send an email to elixir-lang-co...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/elixir-lang-core/f8b6009d-e21a-434a-96e0-e2c6ae3c1a0e%40googlegroups.com.
signature.asc

Jesse Claven

unread,
Feb 2, 2020, 2:42:14 PM2/2/20
to elixir-lang-core
That all makes sense! I would say that in Elm it's used somewhat frequently. I don't have access to the previous code that I worked on (changed jobs), so unfortunately I'm unable to grep for `always` to find some good examples.

In the codebase at my new job, there's a couple of places where `fn _ -> something` is (my original example in this thread). It's basically for anywhere you'd want to ignore some value, and always return something else. I tried searching through GitHub for that code sample but the search functionality was a little subpar.

I understand about keeping the stdlib small, but for a relatively small function, and one that's considered "table stakes" in most FP languages, perhaps it would be a good fit?

Bruce Tate

unread,
Feb 3, 2020, 6:47:44 AM2/3/20
to elixir-l...@googlegroups.com
My counterpoint is this. Any time you can name a concept that makes it easier to see what's going on, it's important. 

Examples: 
* Create back padding of four blank table cells
* use with Stream.repeatedly and take, for example, to initialize a data structure for OTP. 

I do these two things with pretty good frequency because I build responsive layouts often needing tabular structure, but without HTML tables. 

Say you are laying out tables that are responsive but without HTML tables. You
d want to add padding to the end of uneven rows. To create the padding you'd do 

Stream.repeatedly( fn -> :padding end) |> Enum.take(4)  


where :padding is the constant padding you want. This pattern comes up with some regularity in my user interfaces. It doesn't hurt anything, and it would be a great addition to the function module. 

I like this proposal. This is exactly the kind of function you'd expect to see in the module. 

+1 from me. 

-bt

To unsubscribe from this group and stop receiving emails from it, send an email to elixir-lang-co...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/elixir-lang-core/294d4f77-6b91-4316-a152-0c209fad12db%40googlegroups.com.


--

Regards,
Bruce Tate
CEO

Michał Muskała

unread,
Feb 4, 2020, 5:03:39 AM2/4/20
to elixir-l...@googlegroups.com

I’d argue back that this particular pattern, where you want a list of fixed length with the same value, is much better served by `List.duplicate/2`.

 

I think in general, higher order combinator functions like identity, const, flip, and friends are usually used to facilitate the point-free style of programming in languages like Haskell. And in general point-free style usually does not lead to the most readable code.

 

Again, referring to the example provided, if I know anonymous functions, I know what’s going on. When using `Funcion.const`, I have to understand that concept as well. There’s one extra thing to learn.

 

Michał.

 

Regards,

Bruce Tate

CEO

 

Image removed by sender.

--

You received this message because you are subscribed to the Google Groups "elixir-lang-core" group.
To unsubscribe from this group and stop receiving emails from it, send an email to elixir-lang-co...@googlegroups.com.

Ben Wilson

unread,
Feb 4, 2020, 8:16:53 AM2/4/20
to elixir-lang-core
I agree with Michal. Additionally, I'm not clear how `const/1` could be used in Bruce's example at all.

To elaborate,  `fn -> foo() end` and `const(foo())` cannot be equivalent when `const/1` is merely a function. This becomes readily apparent when `foo()` is side effects or side causes. In the first case, `foo()` is never evaluated until the wrapping function is called, in the case of `const/1` however the function is evaluated and then its return value bound over by a function. Eg: `fn -> DateTime.utc_now() end` will always return the current time when evaluated, where as `const(DateTime.utc_now())` will evaluate the current time once and then always return that same time.

That might sound useful, except that we already can do that by simply binding the return value of `DateTime.utc_now()` to a variable and passing that variable around. I'm having difficulty coming up with a scenario where, instead of simply having the value, I have the value wrapped in an anonymous function that I need to call.

Consequently, I struggle to see where `const/1` can actually be used, or how it would work. In the example in the initial proposal, there is this:

```
Enum.map([0,1,2,3], &Function.const/2)
```

As a minor note, presumably that should be `const/1`, right? More importantly, what is the return value here? If it's `[0, 1, 2, 3]` then `const/1` is equivalent to `identity`. If it's

```
[fn -> 1 end, fn -> 2 end, fn -> 3 end, fn -> 4 end]
```

then a simple list of integers seems universally more useful.


 

--

 

Regards,

Bruce Tate

CEO

 

Image removed by sender.

--
You received this message because you are subscribed to the Google Groups "elixir-lang-core" group.

To unsubscribe from this group and stop receiving emails from it, send an email to elixir-l...@googlegroups.com.

Ben Wilson

unread,
Feb 4, 2020, 8:35:10 AM2/4/20
to elixir-lang-core
Addendum: I re-read the proposal because the const/1 vs const/2 thing confused me, and I'm seeing both in play there. The spec is arity 2, the example right after though is arity 1, and the Enum example is arity 2 but without a constant value. The Enum example perhaps makes the most sense, because you could do:

```
Enum.map([1,2,3,4], &const(&1, :foo))
```
which would return `[:foo, :foo, :foo, :foo]` effectively replacing the contents of the list with all `:foo`. Is that the idea?

Amos King

unread,
Feb 4, 2020, 8:58:12 AM2/4/20
to elixir-l...@googlegroups.com
Ben,

That is how const is used in Haskell. Although without currying I don’t see how it is useful. I’m waiting to see an example that drives it home. I agree with Bruce about naming concepts but I don’t see the concept as useful in Elixir.

Bruce, do you have a code sample using the idea?


Amos King
CEO
Binary Noggin

On Feb 4, 2020, at 07:35, Ben Wilson <benwil...@gmail.com> wrote:


To unsubscribe from this group and stop receiving emails from it, send an email to elixir-lang-co...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/elixir-lang-core/1a49eabe-6509-41c7-b425-21e207737472%40googlegroups.com.

Bruce Tate

unread,
Feb 4, 2020, 9:39:48 AM2/4/20
to elixir-l...@googlegroups.com
Thanks for the discussion, all. I enjoy being here. 

Summary: I will bow out of this conversation after this one last attempt. Feel free to skip it. 

I will say that I would use this function frequently in my teaching and in my day-to-day code if I had it. I think it describes a concept that is well known in math, and useful in any functional language. For example, resetting a counter in an agent or genserver with a similar API. Creating a stream of constants as padding/initialization data of arbitrary length (with take later). 

> we'll bloat the language. 

I don't think this is a good argument for foundational concepts. A library that provides names for significant functions in math and programming is important enough.

> I’d argue back that this particular pattern, where you want a list of fixed length with the same value, is much better served by `List.duplicate/2`.

Maybe. But List.duplicate/2 doesn't really solve the same problem. Stream.repeatedly with Function.constant gives me  padding of an arbitrary length, that I can then later instantiate based on the number of columns that I need. It's a great pattern. It's basically fixed vs variable. 

> if I know anonymous functions, I know what’s going on.

Variables, anonymous functions, and the like are not equivalent. A library function has weight and descriptive ability that writing code just doesn't. I can tell you that `fn(x) -> x end` and `Function.identity` are not remotely equivalent when it comes to scanning code. There's a cost to that much syntax. 

I'll say one more thing before moving on from this conversation. 

Naming concepts in code is important. The code Function.constant(c) is much more expressive and descriptive than fn _ -> c end. It gives us a common language across programming languages and theory plus math to express ideas. We in the Elixir community often undersell the idea, and we pay in the idioms that new coders must learn to be successful with our language.

So thanks to all for a great discussion. I do recognize there's no appetite for these ideas in Elixir, so I'll gracefully bow out. 

-bt



To view this discussion on the web visit https://groups.google.com/d/msgid/elixir-lang-core/05581EC8-0AF0-4301-8B45-4F0B78544B2D%40binarynoggin.com.


--

Regards,
Bruce Tate
CEO

Ben Wilson

unread,
Feb 4, 2020, 10:04:02 AM2/4/20
to elixir-lang-core
> The code Function.constant(c) is much more expressive and descriptive than fn _ -> c end.

To be clear, if this could work without macros, I'd be inclined to agree, although I recognize that there may be some who do not. My issue personally is that there simply isn't any function you can write that will behave this way, for all values of `c`. You could build `constant` as a macro, but I think that ends up killing the whole "name a concept" thing, because really the dominant concept at hand ends up being how macros work.

--
You received this message because you are subscribed to the Google Groups "elixir-lang-core" group.
To unsubscribe from this group and stop receiving emails from it, send an email to elixir-l...@googlegroups.com.

Austin Ziegler

unread,
Feb 4, 2020, 11:09:51 AM2/4/20
to elixir-l...@googlegroups.com
On Tue, Feb 4, 2020 at 10:04 AM Ben Wilson <benwil...@gmail.com> wrote:
> The code Function.constant(c) is much more expressive and descriptive than fn _ -> c end.

To be clear, if this could work without macros, I'd be inclined to agree, although I recognize that there may be some who do not. My issue personally is that there simply isn't any function you can write that will behave this way, for all values of `c`. You could build `constant` as a macro, but I think that ends up killing the whole "name a concept" thing, because really the dominant concept at hand ends up being how macros work.

I don’t really have a strong opinion on this proposal, but I do think that this could be done without macros and if naming a concept is important then it’s worth considering. I do see one problem with `Function.constant/1` in how it would be applied, because there would potentially need to be some magic involved (which brings it back to possible macros).

The simplest implementation would be:

def constant(const), do: fn -> const end

That will work well for `Stream.repeatedly/1` as the function parameter is `/0`, but not so well for `Enum.map/2` as the function parameter is `/1`. This could be addressed with multiple functions (constant1/1, etc.) or with an arity parameter (both examples shown below):

def constant(const), do: fn -> const end
def constant1(const), do: fn _ -> const end # OR
def constant(const, arity \\ 0)
def constant(const, 0), do: fn -> const end
def constant(const, 1), do: fn _ -> const end

I’m not sure it would be possible to know, at the call-site, whether you needed a `/0` or `/1` anonymous function through macros, and I’m not sure that it would be meaningful to have constant for _larger_ than `/{0,1}` functions. But I do know that I have occasionally wanted to write:

Enum.map([0, 1, 2], &:padding) # or any other non-function value

and have it act as the correct type of function.

-a
 
To unsubscribe from this group and stop receiving emails from it, send an email to elixir-lang-co...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/elixir-lang-core/89aa185d-368d-4c95-ac41-d5e91f343b7b%40googlegroups.com.


--
Austin Ziegler • halos...@gmail.com • aus...@halostatue.ca
http://www.halostatue.ca/ • http://twitter.com/halostatue

Jesse Claven

unread,
Feb 5, 2020, 4:16:01 PM2/5/20
to elixir-lang-core
Thanks everyone for the good discussion! Especially Bruce who was better able to explain what I was trying to explain. I don't have much to add aside from what he, and others have said.

> Naming concepts in code is important. The code Function.constant(c) is much more expressive and descriptive than fn _ -> c end. It gives us a common language across programming languages and theory plus math to express ideas. We in the Elixir community often undersell the idea, and we pay in the idioms that new coders must learn to be successful with our language.

As someone who has now been using Elixir professionally for just over a month, I agree. ftr I do really like Elixir. ;)

Amos King

unread,
Feb 5, 2020, 4:42:06 PM2/5/20
to elixir-l...@googlegroups.com
I’m voting for adding the concept. Bruce sold me with his arguments. 


Amos King
CEO
Binary Noggin

On Feb 5, 2020, at 15:16, Jesse Claven <jesse....@gmail.com> wrote:


To unsubscribe from this group and stop receiving emails from it, send an email to elixir-lang-co...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/elixir-lang-core/c7496796-b29c-4039-b95d-c76bb7d9c6a1%40googlegroups.com.
Reply all
Reply to author
Forward
0 new messages