to_set() function

[innovate-invent]

We need a mechanism to reduce an array to its unique elements.

The proposed function signature would be array to_set(array arg).

The first occurrence of each element will be kept in the order it is encountered. Elements are compared with each types equivalence logic.

example:
search([1,1,2,1,4], to_set(@)) -> [1,2,4]
search(["a", {"a":1}, {"a":1, "b":2}, [1],{"a":1}, 3, [2]], to_set(@)) -> [["a", {"a":1}, {"a":1, "b":2}, [1], 3, [2]]

I chose "to_set" to be consistent with the other "to_*" functions, but this could alternately be named "unique()" or "deduplicate()".

[springcomp]

That makes me think that we may want something like fold() in the functional programming world...

[innovate-invent]

I was thinking of that myself, but the merge(), flatten [] and map() operations can be used to cover most of its cases.

[springcomp]

Oh, interesting. That needs a new entry in the Frequently Used Paths then 😏.

Maybe this should be moved here or promoted to full blown site page...

[innovate-invent]

I am actually having a hard time imagining a use case for fold. Did you have any in mind?

[springcomp]

Together with simple arithmetic operators, I was thinking about the usual summation of values. For instance, to calculate the total cost of cloud resources out of a JSON report of billing usage.

Not sure if it's a real world scenario, that's more a thought experiment at this stage.

[innovate-invent]

Ah, I see, I wasn't thinking of folding primitives. In that case you are right, let me create another discussion for this.

[springcomp]

I think functions prefixed with to_ should be reserved to those that accept any type.

Your suggestion makes me think of a larger use case. Maybe this particular use case could be addressed with a sort() function.

array sort(array $subject [, bool $unique])

search([1,1,2,1,4], sort(@)) -> [1,1,1,2,4]
search([1,1,2,1,4], sort(@, `true`)) -> [1,2,4]

To address other use cases, such as sorting in reverse order, maybe the $unique argument should be a string accepting a combination of specifiers with values such as u and r, etc.

[innovate-invent]

I see what you mean about the to_ prefix. I am not sure if I like the idea that the items would have to be sorted to get a set of unique items, I might want to keep the order in some cases.

Adding a unique arg to the sort function doesn't seem fitting for the purpose of sort either.

The reverse() function should also be used rather than passing an argument to sort.

[innovate-invent]

I think we should keep the to_ prefix and expand the definition of this function to accept any. This would result in an implicit to_array() call if an array is not passed as the argument.

[springcomp]

So it would behave on a scalar value as if it were invoked with an array containing a single argument ?

[innovate-invent]

Exactly, it would behave exactly as to_array, but de-duplicate values.

[springcomp]

I’m not a strong proponent of side-effects like these. I would rather enforce the type constraint like avg() or map() do for instance.

[hell-racer]

I'd just call this distinct() :)
(I like unique() too)

[springcomp]

Python calls this set(). Should we use that instead ?

• search( set(@), [1, 2, 4, 3, 2, 1] ) -> [1, 2, 4, 3]

[jdevillard]

For me also , distinct or unique is more explicit than set !

[innovate-invent]

I think I am going to vote for 'distinct' if not 'to_set'

[springcomp]

Let's settle on distinct() then. 👍