docs: Clarify associativity of operators.

[elenakrittik]

Clarifies operator associativity as per this comment on rocket.chat.

[AThousandShips]

As I suggested in chat I'd put the note at the top of the list, after "the following is", to make sure it's visible, and then remove the note from the ** entry as it's redundant, but keep the note in part you've edited here

[elenakrittik]

As I suggested in chat I'd put the note at the top of the list, after "the following is", to make sure it's visible, and then remove the note from the ** entry as it's redundant, but keep the note in part you've edited here

Sorry, didn't read your message carefully enough!

[AThousandShips]

LGTM, unsure if it should be in the note or not but I think it makes sense at the top and clear, might put the details below and just the mention here

[dalexeev]

Suggested change

	The following is the list of supported operators and their precedence. All operators are `left-associative <https://en.wikipedia.org/wiki/Operator_associativity>`_,
	The following is the list of supported operators and their precedence. All binary operators are `left-associative <https://en.wikipedia.org/wiki/Operator_associativity>`_,

Binary operators use a table (get_rule()), which does not provide the ability to specify associativity. Unary and ternary operators call parse_precedence() in place, so we need to check their associativity. It looks like the ternary operator is right-associative:

func test():
    1 if 2 else 3 if 4 else 5
    1 if 2 else (3 if 4 else 5)
    (1 if 2 else 3) if 4 else 5

Class <unnamed> :
|   Function test(  ) :
|   |   (1) IF (2) ELSE ((3) IF (4) ELSE (5))
|   |   (1) IF (2) ELSE ((3) IF (4) ELSE (5))
|   |   ((1) IF (2) ELSE (3)) IF (4) ELSE (5)

[elenakrittik]

Found the sources:

• Ternary - https://github.com/godotengine/godot/blob/master/modules/gdscript/gdscript_parser.cpp#L2738-L2761
• Unary - https://github.com/godotengine/godot/blob/master/modules/gdscript/gdscript_parser.cpp#L2571-L2615

Was able to confirm your observation about ternary operator, but i got really confused when i tried to determine associativity of unary operators. First of all, there is no code (or at least i think so) explicitly handling associativity, only precedence. But even then, it doesn't look like precedence is considered for unary operators because both -1 + 1 (lower precedence than unary -) and -1 ** 1 (higher precedence than -) are parsed the same ((-1) + 1 and (-1) ** 1 accordingly), so i'm not sure what exactly precedence does in this case. And in general, i struggle to understand what "associativity" even means in case of unary operators, because parsing them the way it's done right now is the only way that makes sense to me (-1 + 1 == -(1 + 1)? this would be unreasonable; neither it's possible to chain unary ops like binary ops). I'm absolutely lost on the theoretical part here 🤷

But anyway, back from crying into others' jackets. I tested all unary operators with two cases (<unary op> <expr> <binary op with higher precedence> <expr> and the same with lower precedence binary op) and in all cases it results in (<unary op> <expr>) <binary op> (<expr>), so from what i know this means all unary ops are left-associative as well?

[dalexeev]

TLDR: I think we can just write that all binary operators in GDScript are left-associative, the ternary operator is right-associative, and leave out the unary operators.

Indeed, in the case of unary operations, the definition of associativity is not obvious. I see two approaches: 1. the binding strength between operators and operands, 2. the order of code execution.

Different operators have different binding strengths. If there are operators with equal precedence to the left and right of the operand, then the order is determined using associativity. The following options are possible:

Left	Right
(... infix operand) infix ...	... infix (operand infix ...)
(... infix operand) postfix	... infix (operand postfix)
(prefix operand) infix ...	prefix (operand infix ...)
(prefix operand) postfix	prefix (operand postfix)

Some similar operators are grouped, have the same precedence and associativity: + and -, * and /. But unary operators are too different, so have different precedence. GDScript does not have prefix/postfix increment/decrement operators (++ and --), they would be a good example to illustrate. So we simply don't have a case where associativity is important for different unary operators.

And the same unary operator cannot appear on both the left and right (unlike infix binary operators): prefix operators are always located on the left, and postfix operators on the right. 1 + 2 + 3 is valid, but ~x~ is invalid.

If we consider associativity not as the binding strength, but as the order of code execution, then there are several nuances:

1. Prefix operators can be considered right-associative (--x is -(-x)), and postfix operators left-associative (a[x][y] is (a[x])[y]).
2. The ternary operator and the AND and OR operators do not always evaluate all of their operands (short-circuit). However, this does not affect the order.
3. In some languages, the order of expression execution is not always guaranteed. The compiler can make optimizations such as move common subexpressions.

Finally, in some cases, operators can be non-associative, such as chain comparisons in Python.

[dalexeev]

I think it's better not to change the table layout, just add a note before the table.

I think we can just write that all binary operators in GDScript are left-associative, the ternary operator is right-associative, and leave out the unary operators.

[elenakrittik]

I think it's better not to change the table layout, just add a note before the table.

I think we can just write that all binary operators in GDScript are left-associative, the ternary operator is right-associative, and leave out the unary operators.

Hopefully good now! I keep forgetting things all around

[mhilbrunner]

Merged. Thank you! 🎉

[mhilbrunner]

Cherry-picked to 4.2 in #9648.