[5.4] Low-hanging optimization in Eloquent model

[fernandobandeira]

Re-submit of #17656

Changed just getAttributeFromArray so we keep the current behavior...

function foo() {
    $data = ['key' => null];
    if (array_key_exists('key', $data)) {
        // return $data['key'], which is null
    }
}

function bar() {
    $data = ['key' => null];
    if (isset($data['key'])) {
        // not reached
    }
    // return implicit null
}

We are saving around 5% with this change...

props @silwerclaw
also ping @vlakoff

[browner12]

In the case of an attribute being null, do we need an explicit return null at the end, or is that implied?

[fernandobandeira]

Thats implied

[GrahamCampbell]

In the case of an attribute being null, do we need an explicit return null at the end, or is that implied?

Yeh, that's not needed. The introduction of the void return type into the language is just a confusion, since returning nothing is treated as null.

[browner12]

thanks guys. looks like the behavior is identical then.

i can't speak to the speed savings, though.

[fernandobandeira]

Here's a gist that vlakoff did to benchmark it:

https://gist.github.com/vlakoff/2569a9de8b33b05a203f6993ec39e69b

You can modify to apply just the change on getAttributeFromArray

[taylorotwell]

Can you provide before and after benchmarks? I typically do not accept "performance" PRs without them.

[fernandobandeira]

Running the benchmark on PHP7.0

Code that was executed:

User.php

protected function getAttributeFromArray($key)
    {
        global $BENCHMARK_1;
        if (! empty($BENCHMARK_1)) {
            if (isset($this->attributes[$key])) {
                return $this->attributes[$key];
            }
        } else {
            if (array_key_exists($key, $this->attributes)) {
                return $this->attributes[$key];
            }
        }
    }

benchmark.php
        global $BENCHMARK_1, $BENCHMARK_2;
        $nb = 1000;
        $user = \App\User::first();        
        $t1 = microtime(true);
        $BENCHMARK_1 = false;
        $BENCHMARK_2 = false;
        for ($i = $nb; $i--; ) {
            $user->id;
        }
        $t2 = microtime(true);
        $BENCHMARK_1 = true;
        $BENCHMARK_2 = true;
        for ($i = $nb; $i--; ) {
            $user->id;
        }
        $t3 = microtime(true);
        echo $t2 - $t1;
        echo PHP_SAPI == 'cli' ? "\n" : '<br>';
        echo $t3 - $t2;
        echo PHP_SAPI == 'cli' ? "\n\n" : '<br><br>';
        echo $t2 - $t1 == 0 ? 'N/A' : ($t3 - $t2) * 100 / ($t2 - $t1);
        exit;

Results:
0.033835172653198
0.032655954360962

96.514815206285 Around 3.5%.