ternary operator: "left" associativity?

[traq]

Anyone feel like explaining this to me?

I understand the concept of associativity with regards to mathmatical and assignment operators, but this is going right over my head (and making a whistling noise as it passes).
I understand what is happening (the outcome); I don't understand how it is happening (the process):

<?php
echo (FALSE ? "a" : FALSE ? "b" : "c")."\n";
echo (FALSE ? "a" : TRUE  ? "b" : "c")."\n";
echo (TRUE  ? "a" : FALSE ? "b" : "c")."\n";
echo (TRUE  ? "a" : TRUE  ? "b" : "c")."\n";

/** Expected **|** Actual **
    c              c
    b              b
    a              b
    a              b
*/

"Right" associativity (e.g., an assignment statement) evaluates like this:

 $a =  $b = 42;
($a = ($b = 42))

So... what does the "left" associativity of a ternary statement look like?

[Weedpacket]

($a ? $b : $c) ? $d : $e 

Use $a to decide whether to use $b or $c when deciding whether to use $d or $e.

Dumb, but that's how it was made.

[traq]

Ahh... thanks!

I'm pretty sure I've run into this before, too - but I didn't know what was going on. Good to know I am not the crazy one (at least, not for this reason )

[NogDog]

I always use parentheses when I combine ternary expressions like that, so when I come back to them later I don't have to remember how they work again.

Well, actually, I no longer use ternary expressions except for very simple cases, as I'm more into readability now than I am into terseness.

[traq]

I try to use it only for if/else variable assignments.
Since discovering this issue, I've resolved to use them only in simple cases (no more "nested" ternaries).

[traq]

Just a flash. Take these (the constructs that don't work)

echo (TRUE  ? "a" : FALSE ? "b" : "c")."\n"; 
echo (TRUE  ? "a" : TRUE  ? "b" : "c")."\n"; 

and indent them (as you normally would, if they were actual if{} else{} blocks). You'd get:

//  (exaggerated indentation)
echo (TRUE  ? 
            "a" : 
            FALSE ? 
                    "b" : 
                    "c"); 
echo (TRUE  ? 
            "a" : 
             TRUE  ? 
                    "b" :  
                    "c"); 

see how it kinda has a "step-down" look to it?

Invert the conditions (i.e., !condition) and switch the order of their (if true) / (if false) expressions. You'll end up indenting out, then indenting back in, like this:

echo ( !TRUE ?
            !FALSE ?
                        "b":
                        "c":
            "a"
);

echo ( !TRUE ?
            !TRUE ?
                        "b":
                        "c":
            "a"
); 

equivalent conditions, but it works as expected.

Now, I'm not advocating using deeply nested ternaries everywhere. It's just a thorn in my paw when I can't get something to work that should work, dammit. So, just remember to always put new conditionals first, in the "if true" expression, and never in the "if false".

echo (!TRUE  ? !FALSE ? "b" : "c" : "a")."\n"; 
echo (!TRUE  ? !TRUE  ? "b" : "c" : "a")."\n"; 

**********

BTW Merry Christmas, everyone : )

[Weedpacket]

Or parenthesise to override the associativity with a higher-precedence operator (namely the parentheses).

[traq]

yeah. both options are kinda kludgy, unfortunately. I don't think it's reasonable/practical to have to put conditions in a specific (unexplained!) order, or manually nest everything in parenthesis, to make it work.

[Weedpacket]

I know what you mean (hey, I've already said left-associativity was dumb!). For example, it would otherwise be quite tolerable for me to have

$foo = condition1 ? value1 :
       condition2 ? value2 :
       condition3 ? value3 :
       condition4 ? value4 : value0; 

with additional spacing to ensure the ? and : actually do line up.

Reversing the conditionals or putting the parentheses in to fix the associativity give

$foo = condition1 ? value1 :
      (condition2 ? value2 :
      (condition3 ? value3 :
      (condition4 ? value4 : value0)));

$foo = !condition1 ?
       !condition2 ?
       !condition3 ?
       !condition4 ? value0 :
                     value4 : 
                     value3 : 
                     value2 : 
                     value1; 

(I'm not advocating deeply-nested conditionals everywhere either, but the above is completely lucid compared to something like c1 ? c2 : c3 ? c4 ? c5 ? v3 : v4 : (c6 ? v5 : v6) : (c7 ? v7 : v4).)

One thing I'll observe at this point - because I see a lot of code where the author has apparently gone to a lot of trouble to avoid it - is that if(condition ? test1 : test2) is legal and doesn't need to be written as if((condition && test1) || (!condition && test2)).

[traq]

... if(condition ? test1 : test2) is legal and doesn't need to be written as if((condition && test1) || (!condition && test2)).

wow - hadn't thought of that.