Avoiding operator-precedence confusion in 2 + 1:5

[adkabo]

Motivating case

julia> collect(10 .+ 1:5)
Int64[]

julia> collect(10 .+ (1:5))
5-element Array{Int64,1}:
 11
 12
 13
 14
 15

Confusion been reported before on Discourse.

Other languages

Someone in that thread said Julia follows Matlab here.

But : binds before + in R.

> 10 + 1:5
[1] 11 12 13 14 15

: is after + in Python

>>> range(50)[10 + 1:5]
range(11, 5)

but there isn't much opportunity for confusion in Python because 1:5 doesn't form an expression by itself:

>>> range(50)[10 + (1:5)]
  File "<stdin>", line 1
    range(50)[10 + (1:5)]
                     ^
SyntaxError: invalid syntax

and even if it did, it would probably build a slice object, which doesn't have arithmetic special methods so it would give a TypeError anyway.

What's the problem

Julia parses 10 .+ 1:5 in a way that's surprising to some users.

What can be done

Switch the precedence

Julia could switch the precedence of those two operators. This would cause silent behavior change; I don't think that's a good option, even for 2.0.

Get rid of x:y and x:y:z

and just use range(). Getting rid of that syntax would solve the problem, but the short syntax is convenient.

Require parentheses in ambiguous cases

In cases that are potentially ambiguous like 2 + 1:5, require parentheses to disambiguate.

Disallow whitespace inconsistent with precedence

In the 2018 Discourse thread, @simonbyrne suggested

I like the Fortress idea where whitespace which is inconsistent with precedence is a syntax error: this would have caught the original example here.

and @StefanKarpinski said

Yes, I really like that too. It would be breaking though so it would have to be opt-in or wait until 2.0.

It's a similar solution to the one taken in #19089 for 1.+2.

---

I like the latter two options- whitespace, parentheses.