PHP: file_exists vs stream_resolve_include_path - What Performs Better?

Question

It seems as of late there has been a fair amount of wondering on by php developers on whether it is better to use file_exists() or stream_resolve_include_path() when doing checks of whether or not a file exists (be it for including them, caching systems, etc).

It got me wondering if anybody out there has done any benchmark testing on which of these is a better option, for both page load time, server performance and memory usage.

I could not find anything here at SO that addressed this issue so figured it would be time for us to do so.

Answer 1

I have done a little benchmark, but before results, let's see how these functions work. You can read the PHP source code here. There is a french version of this answer, written earlier in the week, good timing ;).

I will talk about is_file() too, as it is defined into the same core function in the source. By core function, I say the C source, not accessible from PHP language into your scripts.

Of what I understand, file_exists() and is_file() are children of the core function php_stat(). This is the highly simplified pseudo-code of the process:

function php_stat($file)
{
    'file_exists'
        ↳ virtual_file_ex($file)
            ↳ virtual_access($file)
                'Windows'
                    ↳ tsrm_win32_access($file)
                        ↳ return access($file)
                'Other systems'
                    ↳ return access($file)
    'is_file'
        ↳ return $file.st_mode == S_IFREG
}

And the pseudo-code of the stream_resolve_include_path() process:

function stream_resolve_include_path($file)
{
    zend_resolve_path($file)
        ↳ php_resolve_path_for_zend($file)
            ↳ php_resolve_path($file)
                ↳ tsrm_realpath($file)
                    ↳ return estrdup($file)
}

From here, without numeric result of a benchmark, you can see how one function is expensive in resource.

---

The code for the benchmark:

function bench_file($file) {
    $res = array();
    $max = 1000000;

    // is_file()
    $res[] = microtime(1);
    for ( $i = 0; $i < $max; ++$i ) {
        if ( is_file($file) ) {
            //
        }
    }
    $res[] = microtime(1);

    clearstatcache();

    // file_exists()
    $res[] = microtime(1);
    for ( $i = 0; $i < $max; ++$i ) {
        if ( file_exists($file) ) {
            //
        }
    }
    $res[] = microtime(1);

    clearstatcache();

    // stream_resolve_include_path()
    $res[] = microtime(1);
    for ( $i = 0; $i < $max; ++$i ) {
        if ( stream_resolve_include_path($file) !== false ) {
            //
        }
    }
    $res[] = microtime(1);

    printf(
        'is_file = %f, file_exists = %f, stream_resolve_include_path = %f',
        $res[1] - $res[0], $res[3] - $res[2], $res[5] - $res[4]
    );

}

Let's test with an existante file (1) and a inexistant one (2):

1 : is_file = 0.218582, file_exists = 0.742195, stream_resolve_include_path = 1.626521
2 : is_file = 0.458983, file_exists = 0.644638, stream_resolve_include_path = 5.623289

Results speak for themselves ;)

---

Benchmark v2 - just an easier way to add new functions to test.

function micro($func, $file) {
    $max = 1000000;
    $start = microtime(1);
    for ( $i = 0; $i < $max; ++$i ) {
        if ( $func($file) ) {
            //
        }
    }
    $end = microtime(1);
    clearstatcache();
    return $end - $start;
}

function bench_file($file) {
    $res = array(
        'is_file' => micro('is_file', $file),
        'file_exists' => micro('file_exists', $file),
        'stream_resolve_include_path' => micro('stream_resolve_include_path', $file)
    );
    $ret = '';
    foreach ( $res as $key => $value ) {
        $ret .= sprintf('%s = %f, ', $key, $value);
    }
    return trim($ret, ', ');
}

echo '<pre>', bench_file('file-ok'), "\n", bench_file('file-ko'), '</pre>';

Results:

is_file = 0.295752, file_exists = 0.852082, stream_resolve_include_path = 1.759607
is_file = 0.527770, file_exists = 0.724793, stream_resolve_include_path = 5.916151

There is a little cost to call $funct(), this explains the slightly higher numbers.