What is better in a foreach loop… using the & symb

问题:

Consider the following PHP Code:

//Method 1
$array = array(1,2,3,4,5);
foreach($array as $i=>$number){
  $number++;
  $array[$i] = $number;
}
print_r($array);


//Method 2
$array = array(1,2,3,4,5);
foreach($array as &$number){
  $number++;
}
print_r($array);

Both methods accomplish the same task, one by assigning a reference and another by re-assigning based on key. I want to use good programming techniques in my work and I wonder which method is the better programming practice? Or is this one of those it doesn't really matter things?

回答1:

Since the highest scoring answer states that the second method is better in every way, I feel compelled to post an answer here. True, looping by reference is more performant, but it isn't without risks/pitfalls.
Bottom line, as always: "Which is better X or Y", the only real answers you can get are:

• It depends on what you're after/what you're doing
• Oh, both are OK, if you know what you're doing
• X is good for Such, Y is better for So
• Don't forget about Z, and even then ...("which is better X, Y or Z" is the same question, so the same answers apply: it depends, both are ok if...)

Be that as it may, as Orangepill showed, the reference-approach offers better performance. In this case, the tradeoff one of performance vs code that is less error-prone, easier to read/maintan. In general, it's considered better to go for safer, more reliable, and more maintainable code:

'Debugging is twice as hard as writing the code in the first place. Therefore, if you write the code as cleverly as possible, you are, by definition, not smart enough to debug it.' — Brian Kernighan

I guess that means the first method has to be considered best practice. But that doesn't mean the second approach should be avoided at all time, so what follows here are the downsides, pitfalls and quirks that you'll have to take into account when using a reference in a foreach loop:

Scope:
For a start, PHP isn't truly block-scoped like C(++), C#, Java, Perl or (with a bit of luck) ECMAScript6... That means that the $value variable will not be unset once the loop has finished. When looping by reference, this means a reference to the last value of whatever object/array you were iterating is floating around. The phrase "an accident waiting to happen" should spring to mind.
Consider what happens to $value, and subsequently $array, in the following code:

$array = range(1,10);
foreach($array as &$value)
{
    $value++;
}
echo json_encode($array);
$value++;
echo json_encode($array);
$value = 'Some random value';
echo json_encode($array);

The output of this snippet will be:

[2,3,4,5,6,7,8,9,10,11]
[2,3,4,5,6,7,8,9,10,12]
[2,3,4,5,6,7,8,9,10,"Some random value"]

In other words, by reusing the $value variable (which references the last element in the array), you're actually manipulating the array itself. This makes for error-prone code, and difficult debugging. As opposed to:

$array = range(1,10);
$array[] = 'foobar';
foreach($array as $k => $v)
{
    $array[$k]++;//increments foobar, to foobas!
    if ($array[$k] === ($v +1))//$v + 1 yields 1 if $v === 'foobar'
    {//so 'foobas' === 1 => false
        $array[$k] = $v;//restore initial value: foobar
    }
}

Maintainability/idiot-proofness:
Of course, you might say that the dangling reference is an easy fix, and you'd be right:

foreach($array as &$value)
{
    $value++;
}
unset($value);

But after you've written your first 100 loops with references, do you honestly believe you won't have forgotten to unset a single reference? Of course not! It's so uncommon to unset variables that have been used in a loop (we assume the GC will take care of it for us), so most of the time, you don't bother. When references are involved, this is a source of frustration, mysterious bug-reports, or traveling values, where you're using complex nested loops, possibly with multiple references... The horror, the horror.
Besides, as time passes, who's to say that the next person working on your code won't foget about unset? Who knows, he might not even know about references, or see your numerous unset calls and deem them redundant, a sign of your being paranoid, and delete them all together. Comments alone won't help you: they need to be read, and everyone working with your code should be thoroughly briefed, perhaps have them read a full article on the subject. The examples listed in the linked article are bad, but I've seen worse, still:

foreach($nestedArr as &$array)
{
    if (count($array)%2 === 0)
    {
        foreach($array as &$value)
        {//pointless, but you get the idea...
            $value = array($value, 'Part of even-length array');
        }
        //$value now references the last index of $array
    }
    else
    {
        $value = array_pop($array);//assigns new value to var that might be a reference!
        $value = is_numeric($value) ? $value/2 : null;
        array_push($array, $value);//congrats, X-references ==> traveling value!
    }
}

This is a simple example of a traveling value problem. I did not make this up, BTW, I've come across code that boils down to this... honestly. Quite apart from spotting the bug, and understanding the code (which has been made more difficult by the references), it's still quite obvious in this example, mainly because it's a mere 15 lines long, even using the spacious Allman coding style... Now imagine this basic construct being used in code that actually does something even slightly more complex, and meaningful. Good luck debugging that.

side-effects:
It's often said that functions shouldn't have side-effects, because side-effects are (rightfully) considered to be code-smell. Though foreach is a language construct, and not a function, in your example, the same mindset should apply. When using too many references, you're being too clever for your own good, and might find yourself having to step through a loop, just to know what is being referenced by what variable, and when.
The first method hasn't got this problem: you have the key, so you know where you are in the array. What's more, with the first method, you can perform any number of operations on the value, without changing the original value in the array (no side-effects):

function recursiveFunc($n, $max = 10)
{
    if (--$max)
    {
        return $n === 1 ? 10-$max : recursiveFunc($n%2 ? ($n*3)+1 : $n/2, $max);
    }
    return null;
}
$array = range(10,20);
foreach($array as $k => $v)
{
    $v = recursiveFunc($v);//reassigning $v here
    if ($v !== null)
    {
        $array[$k] = $v;//only now, will the actual array change
    }
}
echo json_encode($array);

This generates the output:

[7,11,12,13,14,15,5,17,18,19,8]

As you can see, the first, seventh and tenth elements have been altered, the others haven't. If we were to rewrite this code using a loop by reference, the loop looks a lot smaller, but the output will be different (we have a side-effect):

$array = range(10,20);
foreach($array as &$v)
{
    $v = recursiveFunc($v);//Changes the original array...
    //granted, if your version permits it, you'd probably do:
    $v = recursiveFunc($v) ?: $v;
}
echo json_encode($array);
//[7,null,null,null,null,null,5,null,null,null,8]

To counter this, we'll either have to create a temporary variable, or call the function tiwce, or add a key, and recalculate the initial value of $v, but that's just plain stupid (that's adding complexity to fix what shouldn't be broken):

foreach($array as &$v)
{
    $temp = recursiveFunc($v);//creating copy here, anyway
    $v = $temp ? $temp : $v;//assignment doesn't require the lookup, though
}
//or:
foreach($array as &$v)
{
    $v = recursiveFunc($v) ? recursiveFunc($v) : $v;//2 calls === twice the overhead!
}
//or
$base = reset($array);//get the base value
foreach($array as $k => &$v)
{//silly combine both methods to fix what needn't be a problem to begin with
    $v = recursiveFunc($v);
    if ($v === 0)
    {
        $v = $base + $k;
    }
}

Anyway, adding branches, temp variables and what have you, rather defeats the point. For one, it introduces extra overhead which will eat away at the performance benefits references gave you in the first place.
If you have to add logic to a loop, to fix something that shouldn't need fixing, you should step back, and think about what tools you're using. 9/10 times, you chose the wrong tool for the job.

The last thing that, to me at least, is a compelling argument for the first method is simple: readability. The reference-operator (&) is easily overlooked if you're doing some quick fixes, or try to add functionality. You could be creating bugs in the code that was working just fine. What's more: because it was working fine, you might not test the existing functionality as thoroughly because there were no known issues.
Discovering a bug that went into production, because of your overlooking an operator might sound silly, but you wouldn't be the first to have encountered this.

Note:
Passing by reference at call-time has been removed since 5.4. Be weary of features/functionality that is subject to changes. a standard iteration of an array hasn't changed in years. I guess it's what you could call "proven technology". It does what it says on the tin, and is the safer way of doing things. So what if it's slower? If speed is an issue, you can optimize your code, and introduce references to your loops then.
When writing new code, go for the easy-to-read, most failsafe option. Optimization can (and indeed should) wait until everything's tried and tested.

And as always: premature optimization is the root of all evil. And Choose the right tool for the job, not because it's new and shiny.

回答2:

As far as performance is concerned Method 2 is better, especially if you either have a large array and/or are using string keys.

While both methods use the same amount of memory the first method requires the array to be searched, even though this search is done by a index the lookup has some overhead.

Given this test script:

$array = range(1, 1000000);

$start = microtime(true);
foreach($array as $k => $v){
    $array[$k] = $v+1;
}
echo "Method 1: ".((microtime(true)-$start));

echo "\n";

$start = microtime(true);
foreach($array as $k => &$v){
    $v+=1;
}
echo "Method 2: ".((microtime(true)-$start));

The average output is

Method 1: 0.72429609298706
Method 2: 0.22671484947205

If I scale back the test to only run ten times instead of 1 million I get results like

Method 1: 3.504753112793E-5
Method 2: 1.2874603271484E-5

With string keys the performance difference is more pronounced. So running.

$array = array();
for($x = 0; $x<1000000; $x++){
    $array["num".$x] = $x+1;
}

$start = microtime(true);
foreach($array as $k => $v){
    $array[$k] = $v+1;
}
echo "Method 1: ".((microtime(true)-$start));

echo "\n";

$start = microtime(true);
foreach($array as $k => &$v){
    $v+=1;
}
echo "Method 2: ".((microtime(true)-$start));

Yields performance like

Method 1: 0.90371179580688
Method 2: 0.2799870967865

This is because searching by string key has more overhead then the array index.

It is also worth noting that as suggested in Elias Van Ootegem's Answer to properly clean up after yourself you should unset the reference after the loop has completed. I.e. unset($v); And the performance gains should be measured against the loss in readability.

回答3:

There are some minor performance differences, but they aren't going to have any significant effect.

I would choose the first option for two reasons:

1. It's more readable. This is a bit of a personal preference, but at first glance, it's not immediately obvious to me that $number++ is updating the array. By explicitly using something like $array[$i]++, it's much clearer, and less likely to cause confusion when you come back to this code in a year.

2. It doesn't leave you with a dangling reference to the last item in the array. Consider this code:

   $array = array(1,2,3,4,5);
   foreach($array as &$number){
       $number++;
   }
   
   // ... some time later in an unrelated section of code
   $number = intval("100");
   
   // now unexpectedly, $array[4] == 100 instead of 6
   

回答4:

I guess that depends. Do you care more about code readability/maintainability or minimizing memory usage. The second method would use slightly less memory, but I would honestly prefere the first usage, as assigned by reference in foreach definition does not seem to be commonplace practice in PHP.

Personally if I wanted to modify an array in place like this I would go with a third option:

array_walk($array, function(&$value) {
    $value++;
});  

回答5:

The first method will be insignificantly slower, because each time it will go through the loop, it will assign a new value to the $number variable. The second method uses the variable directly so it doesn't need to assign a new value for each loop.

But, as I said, the difference is not significant, the main thing to consider is readability.

In my opinion, the first method makes more sense when you don't need to modify the value in the loop, the $number variable would only be read.

The second method makes more sense when you need to modify the $number variable often, as you don't need to repeat the key each time you want to modify it, and it is more readable.

回答6:

Have you considered array_map? It is designed to change values inside arrays.

$array = array(1,2,3,4,5);
$new = array_map(function($number){
  return $number++ ;
}, $array) ;
var_dump($new) ;

回答7:

I'd choose #2, but it's a personal preference.

I disagree with the other answers, using references to array items in foreach loops is quite common, but it depends on the framework you're using. As always, try to follow existing coding conventions in your project or framework.

I also disagree with the other answers that suggest array_map or array_walk. These introduce the overhead of a function call for each array element. For small arrays, this won't be significant, but for large arrays, this will add a significant overhead for such a simple function. However, they are appropriate if you're performing more significant calculations or actions - you'll need to decide which to use depending on the scenario, perhaps by benchmarking.

回答8:

Most of the answers interpreted your question to be about performance.

This is not what you asked. What you asked is:

I wonder which method is the better programming practice?

As you said, both do the same thing. Both work. In the end, better is often a matter of opinion.

Or is this one of those it doesn't really matter things?

I wouldn't go so far as to say it doesn't matter. As you can see there can be performance considerations for Method 1 and reference gotchas for Method 2.

I can say what matters more is readability and consistency. While there are dozens of ways to increment array elements in PHP, some look like line noise or code golf.

Ensuring your code is readable to future developers and you consistently apply your method of solving problems is a far better macro programming practice than whatever micro differences exist in this foreach code.