Protocol implementations themselves seem intrinsically public to me, if for no other reason than that they are global. You cannot create a protocol implementation for a given protocol for a given type if one already exists. Every protocol implementation Elixir ships with has direct and public implications with respect to your code base because of this. If you rely on functions being enumerable, you literally have no choice but to use Elixir’s built in impl, it is impossible to write your own.
benwilson512
Author of Craft GraphQL APIs in Elixir with Absinthe
arikai
Well, sure, you are not wrong.
Let me just recap the whole issue for me here:
- There’s an undocumented code in the Elixir stdlib that exists simply because it was the only possible way to implement such “structures”
- There’s a better way to construct procedurally generated sequences, that is, Stream (e.g. Stream.resource)
That’s all.
0
benwilson512
Author of Craft GraphQL APIs in Elixir with Absinthe
Happy to debate whether it is a good idea to support functions or not but it isn’t undocumented:
Since enumerables can have different shapes (structs, anonymous functions, and so on), the functions in this module may return any of those shapes and that this may change at any time. For example, a function that today returns an anonymous function may return a struct in future releases.
That note has been there since Elixir 1.0
1
NobbZ
The current Stream struct, only works to transform some Enumerable into something else lazily, it can not represent a source, as well as it can’t represent a sink (well, nothing in the Stream module can create a sink, therefore thats not important).
And regardles how you represent lazily evaluated sequences, you’ll always need a function.
In my opinion it does not make any differences if a function is sent directly over the wire or wrapped in a struct. It will be called eventually…
0
arikai
@benwilson512 Huh, that is a small one. However, this line only documents that it is a bad decision to use function’s return value as anything but value of opaque type that somehow has impl for Enumerable. Nothing else. It doesn’t explicitly says that it’s ok to use function(2) as an Enumerable. Therefore, whoever used it is their own enemy.
@NobbZ Well, it does. IMO, if there exists Stream struct in a language, there shouldn’t be such a “hacky” (IMO) solutions as a function(2) for Enumerables, because it can easily lead to a rather strange errors, where somehow function(2) has been used as a first argument to Enumerable/Stream, and yet, it doesn’t tell that input itself is bad: it will tell that the function returned bad values, which is ambiguous to a programmer.
0
mat-hek
Membrane Core Team
Totally agree. I was digging in the Stream implementation when contributing to stream split and the problem you mention is one of the things that make it difficult to reason about.
0
josevalim
Creator of Elixir
I think there is a confusion on the arguments being made here.
-
The argument that using
function(2)can be the cause of confusion - because I can accidentally pass it around as a enumerable is a valid one -
The argument that using
function(2)is somewhat more dangerous is incorrect though. The issue that can be triggered withfunction(2)would also help if we wrapped it in aStreamor any other struct, because Streams are fundamentally about passing continuations of code around and changing it fromfunction(2)toStreamwon’t fix it.
So while the first argument 1 is valid and has some merit - it is not a major improvement to justify changing the current feature set. Maybe there could be other benefits in making them more structured - but if there are, I am not aware of them at the moment.
I don’t think it would make the implementation easier to digest as well. As ultimately the difference would be that a function would return Stream.stream(&foo(&1, &2)) instead of &foo(&1, &2). Streams are naturally complex, given the fact they provide zipping, filtering, and protection against dangling resources. Wrapping an anonymous function call in a remote call or not is not going to do much to counter that.
6
arikai
@mat-hek
That’s an interesting work you have there. Keep it up! I would definitely like to see this kind of functionality in the existing Stream!
@josevalim
It may be that I’ve made myself not totally clear. The second point wasn’t the argument all along, just as I’ve pointed out in the second paragraph of the OP.
However, I do agree that it is a rather minor issue to worry about in the current state of things.
It would be good though to put a removal of such an artifact on a roadmap/backlog if you think it might be for a good reason.
1