BENCHMARK.md

Benchmark

How I measured the times

I did each benchmark test ten times and calculated the average time. Benchmarks were done on the command line on a Fedora 20 system with 6 GiB RAM and a i5-33170 CPU processor @ 1.70 GHz with 4 cores. I used the following script:

TypeHinting::initialize();

function foobar_boolean( boolean $value ) {} # or any other data type

$i = 0;
$start_time = microtime( true );

while ( $i < 1000 ) {
  ++$i;
  
  # exception routine is needed because elsewise
  # the script would stop when the first exception occurs
  try {
    foobar_boolean( true ); # or any other function and data type
  } catch ( Exception $e ) {}
}

$end_time = microtime( true );
$time     = $end_time - $start_time;

echo 'time used: ', $time;

Benchmark results

expected type	given type	time for 1000 calls
boolean	boolean	6150 µsec
boolean	integer	17546 µsec
integer	integer	6324 µsec
integer	boolean	17783 µsec
double	double	6852 µsec
double	boolean	18228 µsec
string	string	6707 µsec
string	boolean	20813 µsec
resource	resource	6460 µsec
resource	boolean	18907 µsec
number	integer	6975 µsec
number	double	7211 µsec
number	boolean	18823 µsec
scalar	boolean	6703 µsec
scalar	integer	6909 µsec
scalar	double	7054 µsec
scalar	string	7019 µsec
scalar	resource	18611 µsec
mixed	boolean	6541 µsec
numeric	integer	13490 µsec
numeric	double	13713 µsec
numeric	string	14046 µsec
numeric	boolean	24074 µsec

Conclusion

• there is no differenc between data types as long (in the same case)
• boolean, integer, double, string, resource and mixed need almost the same time
• number and scalar need a little longer
• numeric needs nearly twice the time
• calls need up to three times if you pass the wrong data type

You should bear in mind that in the most cases you will stop if there is an error. Also you will not put an exception handler around every function call. So the times might be some lower than my measured times.