PHP Pass By Reference Increases Memory Usage

Question

When calling functions and passing in a large string I normally pass by reference. The intent is to keep PHP from passing by value a copy of the string, thus reducing memory usage. However, I have noticed the opposite under certain circumstances.

In the below example, peak memory usage doubles when passing the string by reference into the function, at the is_string() call.

Code:

$buf = '';
report_memory(__LINE__);

$buf = file_get_contents('./20MB.pdf');
report_memory(__LINE__);

example($buf);
report_memory(__LINE__);

function example(& $str) {
    report_memory(__LINE__);
    is_string($str);
    report_memory(__LINE__);
}

function report_memory($line=0) {
    echo 'Line: ' . str_pad($line,3) . ' ';
    echo 'Mem: '  . str_pad(intval(memory_get_usage()/1024     ) . 'K',8) . ' ';
    echo 'Peak: ' . str_pad(intval(memory_get_peak_usage()/1024) . 'K',8) . ' ';
    echo "\n";
}

Output:

Line: 2   Mem: 625K     Peak: 631K
Line: 5   Mem: 21058K   Peak: 21076K
Line: 11  Mem: 21058K   Peak: 21076K
Line: 13  Mem: 21058K   Peak: 41494K
Line: 8   Mem: 21058K   Peak: 41494K

Changing the function example() to remove the reference:

function example($str) {
    report_memory(__LINE__);
    is_string($str);
    report_memory(__LINE__);
}

Output:

Line: 2   Mem: 625K     Peak: 631K
Line: 5   Mem: 21058K   Peak: 21076K
Line: 11  Mem: 21058K   Peak: 21076K
Line: 13  Mem: 21058K   Peak: 21076K
Line: 8   Mem: 21058K   Peak: 21076K

This is what I would expect. So then why the memory doubling with is_string() when the variable is a reference?

The environment is CentOS 6.6 running a stock PHP 5.3.3.

Answer 1

Zend Engine uses a lazy-copy mechanism. Until the variable is modified it does not create a copy of the variable. PHP create a set of structures that it uses to maintain the reference that why it is actually “slower” to pass a variable by reference.