`Set` should not understand the `,` message

[jordanmontt]

The , message is typically used to concatenate ordered collections such as arrays or strings. Its expected behavior is to return a sequential, ordered concatenation of the receiver and the argument.

However, a Set does not preserve order by definition. Because of this, supporting the , message on Set introduces semantic inconsistency: the operation suggests ordered concatenation, while sets are unordered collections.

Proposed change

The Set class should not implement or respond to the , message. Users should call the union: method explicitly.

[estebanlm]

wait, wait, wait.
It is used to concatenate collections, in general. Nothing in the idiom explicit it should be an ordered collection.
I however, do not opose to implement , in a set as #union: (since a set is a set :P)

[guillep]

Well, in my understanding, which may be wrong, concatenate implies order.

Also, union: is already defined in Set.

Now, the thing about having this API in collection is that it can lead you to strange bugs, and difficult to answer questions.
Some examples:

What happens if I "concatenate" an array and a set?

• anArray, aSet => error
• aSet, anArray => we have a set forgetting the order of the array (because of set of course)
• dictionary, set => error
• set, dictionary => the set has the values of the dictionary (arguable, why not the keys?)
• array, dict => error
• dict, array => a dictionary, but it transformed the array elements in associations, why? It's rather arbitrary

set := #( 4 5 6) asSet.
array := #(1 2 3).
dict := { 7->8 . 9->10 } asDictionary .

set, array "a Set(4 1 5 2 6 3)".
array, set "Error"

set, dict "a Set(4 8 5 6 10)".
dict, set "Error".

array, dict "error".
dict, array. "a Dictionary(1->1 2->2 3->3 7->8 9->10 )"

From what I can see:

• the operation is sometimes commutative (on non-sequenceable collections), but not on sequenceable ones => this hints me that these are two different operations
• concatenating collections of different types (species?), the decision taken is at least arguable. Any of the cases above mixing types are just a potential bug that you may hit too late

[estebanlm]

uhm... maybe you're right.
maybe we should add arithmetic-like semantic for union: then (+)

[Ducasse]

Really interesting because I only use , for strings.
So to me , on other collections is just making sure that we will have more bugs

[DARSHAN-R-DARSHAN]

Hi, I am just asking, can I work on this? If yes, I will really be happy to work