PHP SPL: Why is RecursiveIteratorIterator 100x slower than recursive search

[Marcus]

This is the old school recursive search for a node=20 within the grouped nodes:

    function search($array, $key, $value)
     {
       $results = array();
    
       search_r($array, $key, $value, $results);
    
       return $results;
     }
    
     function search_r($array, $key, $value, &$results)
     {
       if (!is_array($array))
         return;
    
       if (isset($array[$key]) && $array[$key] == $value)
         $results[] = $array;
    
       foreach ($array as $subarray)
         search_r($subarray, $key, $value, $results);
     }

     $outputArray = search($groupedNodes, 'node_id', 20);

PHP SPL is much cleaner, but also more than 100x slower:

     $arrIt = new RecursiveIteratorIterator(new RecursiveArrayIterator($groupedNodes));
    
     foreach ($arrIt as $key =>$sub) {
       $subArray = $arrIt->getSubIterator();
       if (isset($subArray['node_id'])) {
         if ($subArray['node_id'] === 20) {
           $outputArray[] = iterator_to_array($subArray);
         }
       }
     }

Do you have any idea why its so much slower, and if SPL is in general slower than using custom made PHP 4 code?

 

[Jeremy]

$key => $value is slower in general than without it or directly assigning a $key as in your original code.

 

[Marcus]

When removed $key = , its around the same speed. More than 100 times slower.

I prefer using PHP SPL for this problem, how come it is that much slower?

 

[Jeremy]

Without looking or running the code its hard to tell. You are however doing 2 separate class instantiations (which, depending on complexity could be difficult) and you convert to an array, which again, depending on the complexity of that function can slow it down.

 

[Marcus]

To compare it easier, the bare php function takes 1 time point.

These class instantiation takes 70% (0,7 time points) of the time the old school solution needs for its whole seach:

$arrIt = new RecursiveIteratorIterator(new RecursiveArrayIterator($groupedNodes));

However this is what slows down everything (99 time points):

    foreach ($arrIt as $sub) {
       $subArray = $arrIt->getSubIterator();
       if (isset($subArray['node_id'])) {
         if ($subArray['node_id'] === 20) {
           $outputArray[] = iterator_to_array($subArray);
         }
       }
     }

Do you know if PHP SPL is in general slower than using bare php for solving these array problems?

 

[Jeremy]

I've never used the functions you are using, so no, I don't know from experience. However, you do an iterator to array conversion (or it seems like it) on every iteration, that may be a slow function (causing the new code to be slower).

 

[Marcus]

First line takes a bit less time than the whole old school search function. The second line is slowing everything 100x down just browsing through the array. I thought PHPs implementation of build in function would be more efficient.

$arrIt = new RecursiveIteratorIterator(new RecursiveArrayIterator($groupedNodes));
foreach ($arrIt as $sub) {}

 

[Jeremy]

I subscribe by the theory that just because its new, doesn't mean its better. Any reason your older code doesn't work?

 

[Marcus]

I want to use the more modern functions.

I guess I know now why the newer function is so slow. It walks through the array bit by bit. And that means, that every single item is handled like a sub array (node_id, title, ...) and it tries to walk through it. I want to just check against node_id and then do a ->next(). But that would take me to 'title', and not the next node_id array. There is no such thing as a "get out of this current tree" where after doing that a ->next would bring me to the next node.

 

[Marcus]

Guess PHP SPL is not doing some wonderful assembler speed accelleration. Maybe it is just a library and casual php functions are executed therein. If that is true, it's better to code my solution in regular PHP.