ternary operator ambiguous behaviour

[berceanu]

Expressions like
1 > 0 ? a=2 : a=3
fail with
syntax: invalid assignment location
This can be solved by
1 > 0 ? a=2 : (a=3)
but I think the expression is unambiguous enough to work without parantheses.

[quinnj]

Why not do a = 1 > 0 ? 2 : 3? <-- That's actually the only allowed use of the ternary in Java.

[berceanu]

what about 1 > 0 ? a=2 : b=3 then?

[mbauman]

Sure, but this seems inconsistent with the other parsing behaviors of ?:. My mental model is that it should try to find the matching else : greedily before allowing : to be used within a range.

1 > 0 ? a=2 : 3 # works fine
1 > 0 ? 2 : a=3 # fails (2:a is an invalid assignment location)

That said, this seems unambiguous to me only because #6823 hasn't yet been decided upon — there still must be a colon somewhere in the statement. Were that to change this becomes more tricky.

Edit: oooh, I see. It's not 2:a that's the invalid location; it's the whole ternary expression! That does make a little more sense to me now.

[joehuchette]

The problem is that = has low precedence, so this gets parsed as

julia> a = :(1 > 0 ? a=2 : a=3)
:(if 1 > 0
            a = 2
        else
            a
        end = 3)

[JeffBezanson]

? is parsing the then argument as nonterminal eq*, and the else argument as nonterminal cond (the precedence of ?). They should probably both be parsed as eq*, fixing this issue.