Awesome
Reflection benchmark
This is a benchmark of all methods allowing to access private properties (read and write).
We all know that micro-optimization is most of the time useless. However, if you write libraries or frameworks making extensive use of reflection techniques, you probably wonder how to spare the impact of using reflection to your users. This comparison is here to help by maintaining an up-to-date and reproducible summary.
Methods
Reflection
API: read & write
$reflectionProperty = new ReflectionProperty($object, $property);
$reflectionProperty->setAccessible(true);
$value = $reflectionProperty->getValue($object);
$reflectionProperty->setValue($object, 'new value');
$array = (array) $object;
$protectedKey = "\0*\0" . $this->propertyName;
$privateKey = "\0" . get_class($this->object) . "\0" . $this->propertyName;
if (array_key_exists($protectedKey, $array)) {
$value = $array[$protectedKey];
} elseif (array_key_exists($privateKey, $array)) {
$value = $array[$privateKey];
}
- Bind a closure to the scope of an object: read & write, but on PHP >= 5.4
$get = Closure::bind(function($object, $prop) { return $object->{$prop}; }, null, $object);
$set = Closure::bind(function($object, $prop, $value) { $object->{$prop} = $value; }, null, $object);
$value = $get($object, $property);
$set($object, $property, 'new value');
Running the benchmarks
Install the dependencies using Composer:
$ composer install
Run the benchmarks:
$ php -n vendor/bin/athletic -p benchmarks -b vendor/autoload.php
Results
Those results were obtained on a development OS X machine using PHP 7.1.10. Feel free to run the benchmark on your server and update these results if you get different ones. Be aware that "slightly faster" means that it's probably not significant, whereas "faster" or "way faster" means that results are expected to be reproducible for the same PHP version.
- Read a single property on one object:
The closure is slightly faster.
- Read a single property on many objects:
Reflection is way faster, the closure is almost as fast.
- Reading all the properties of an object:
The array cast is way faster, with the closure not far behind.
- Writing a single property on one object:
The closure is slightly faster.
- Writing a single property on many objects:
Reflection is slightly faster.
Summary
With PHP 7, if you were to use only one method for all operations then use the closure. It is the solution that provides the best performances accross all operations.
Raw results
Benchmark\ReadAllPropertiesOnOneObjectEvent
Method Name Iterations Average Time Ops/second
---------- ------------ -------------- -------------
reflection: [10,000 ] [0.0000030814409] [324,523.50188]
arrayCast : [10,000 ] [0.0000008020163] [1,246,857.51657]
closure : [10,000 ] [0.0000010994911] [909,511.66623]
Benchmark\ReadSinglePropertyOnManyObjectsEvent
Method Name Iterations Average Time Ops/second
---------- ------------ -------------- -------------
reflection: [100,000 ] [0.0000001455283] [6,871,514.93306]
arrayCast : [100,000 ] [0.0000002697277] [3,707,442.63338]
closure : [100,000 ] [0.0000001647019] [6,071,573.94943]
Benchmark\ReadSinglePropertyOnOneObjectEvent
Method Name Iterations Average Time Ops/second
---------- ------------ -------------- -------------
reflection: [100,000 ] [0.0000004917526] [2,033,542.78179]
arrayCast : [100,000 ] [0.0000004168963] [2,398,677.79182]
closure : [100,000 ] [0.0000004045892] [2,471,642.98721]
Benchmark\WriteSinglePropertyOnManyObjectsEvent
Method Name Iterations Average Time Ops/second
---------- ------------ -------------- -------------
reflection: [50,000 ] [0.0000001499557] [6,668,633.93538]
closure : [50,000 ] [0.0000001540565] [6,491,122.94169]
Benchmark\WriteSinglePropertyOnOneObjectEvent
Method Name Iterations Average Time Ops/second
---------- ------------ -------------- -------------
reflection: [50,000 ] [0.0000005018568] [1,992,600.26414]
closure : [50,000 ] [0.0000004431438] [2,256,603.61117]
Contribute
Please provide fixes and improvements through pull requests.
Don't forget to update the results in the README.
Credits
This is a compilation of the methods Marco Pivetta exposed on his blog.
Read also this article.