parsing of ... with ranges

[mlubin]

This is potentially deceptive:

julia> f(x...) = dump(x)

julia> f(1:2)
(Range1{Int64},) (1:2,)

julia> f(1:2...)
(Range1{Int64},) (1:2,)

julia> f((1:2)...)
(Int64,Int64) (1,2)

Is there a reason why the ... is ignored in the parsing of 1:2...?

[carlobaldassi]

The ... is not ignored, but it's attached to 2:

julia> [2...]
1-element Int64 Array:
 2

julia> 1:2...
1:2

I suppose that the latter is because 1:2... gets translated to colon(1, 2...) and numbers are iterables which yield one value.

Se also:

julia> 1...:2
1:2

I actually don't see a reason why : shouldn't take precedence over ... when written as 1:2..., i.e. I don't see any reasonable usage of 1:(2...) (maybe if colon is overloaded on custom types? Still the usual case seems more frequent).

Maybe the precedence of these operators should be changed?

[JeffBezanson]

Yeah, this is unintentional. Seems better to parse it as (1:2)....