About this gist

This gist aims to prove or disprove the assertion made here[1] and linked on
Ajaxian[2] that dynamic script requests, or more specifically JSONP requests,
leak memory in all modern browsers.

The specific setup is a 200ms interval that requests an uncached script pointed at
the simple php file.  This file in response sends the following string:
update()

This is not technically a JSONP response, since the would-be callback is executed
without arguments.  However, memory footprint results can be gauged on generic
script content rather than focused on the JSONP pattern.

With all versions of this setup in the gist history, there IS a memory leak.

Log:
version 1: use a simple xbrowser getScript in the setup described
version 2: move the script node clearing code out of the getScript function to test
for closure related leakage.  Add the "About this gist" file to track progress.
version 3: formatting updates to "About this gist" file.
version 4: testing setting defer to true (no change)
version 5: wrapping node creation in try/finally to null node references (no change)
version 6: calling script.clearAttributes() if available (IE) before node removal.  Interestingly, this did seem to flatline the mem footprint until about 500 requests, at which point memory climbed rapidly.
version 7: per Andrea Giammarchi's destroy suggestion[3][4], added while loop to remove
all childNodes before removing from the DOM (no change)
version 8: moved the clearAttributes and childNode removal after the script node
removal.  Again, seemed flat until ~400 requests, then shot up.
version 9: replaced clearAttributes call with manual iteration of script props,
assigning each to null inside a try/catch because IE throws errors.  Memory
consumption shot up faster than any prior configuration.
version 10: replaced property assignment to null with deleting properties.  This seriously limited the leak (*validates the author's assertion*).
Browser  BEFORE      AFTER
IE6      (forthcoming, but it was an obvious decrease in Drip)
Saf4     +6.5M/500   +2M/500
FF3.5    +0/500      (no leak?)
other browsers forthcoming
version 11: Changed to Process Explorer[5] from Drip (via @tobie) to avoid false
positives.  Switched to a new FF profile with no plugins to avoid false positive from
add-ons (I'm looking at you FireBug).  Returned to onload = null approach, removed
defer = false, and did lengthier trials.
Results:
Browser  RUN    START     END      REQs    NOTES
IE6      1/1    16M/10M   26M/20M  ~18600  (Physical Mem/Virtual Mem)
FF3.5    1/2    66.33M    66.99M   500     memory returned to normal after a few secs
         2/2    54.39M    58.91M   22400   memory did not return after 5mins
Saf4     1/1    66.16M    ~104M    8400    yikes!  Opening Web Inspector brought app
                                           to a crawl and mem jumped to 384M. 
                                           Resources tab showed every request.
IE8      1/1    39M/33M   55M/48M  19939   IE8 mode (not IE7 compat mode)

version 12: Reapplied delete-all-properties patch for retest.
Results:
Browser  RUN    START     END      REQs    NOTES
IE6      1/1    16M/10M   21M/14M  8046    Patch didn't work? I cleared cache and
                                           viewed source to make sure the script
                                           state was correct.  Test cut short because
                                           it's past bed time.
FF3.5    1/1    51.38M    53.86M   17664   The clear winner here.
Saf4     1/1    29.54     149.36M  17619   The clear loser.
IE8      1/1    18M/16M   39M/35M  17224


1. http://neil.fraser.name/news/2009/07/27/
2. http://ajaxian.com/archives/dynamic-script-generation-and-memory-leaks
3. http://webreflection.blogspot.com/2009/04/drip-under-control-via-another-ie.html
4. http://ajaxian.com/archives/dynamic-script-generation-and-memory-leaks#comment-274727
5. http://technet.microsoft.com/en-us/sysinternals/bb896653.aspx

server side component to mem leak test

<?php

header("Cache-control: no-cache");
header("Expires: -1");
header("Content-type: text/javascript");
echo "update()"

?>

Testing mem leak in dynamic script requests

<!doctype html>
<html>
<head>
    <title>Test Page</title>
</head>
<body>

<p id="inc">0</p>
<p id="remaining">0</p>

<button id="go">Start</button>

<script type="text/javascript">
var timer = null,
    update_int = null,
    count = 0;

// Executed by the requested scripts
function update() { ++count; }

function clearScript() {
    var hasReadyState = ('readyState' in this);

    if (!hasReadyState || /loaded|complete/.test(this.readyState)) {
        this.parentNode.removeChild(this);
        for (var k in this) {
            try {
                delete this[k];
            }
            catch (e) {}
        }
    }
}

function getScript(U){
    var h = document.documentElement.firstChild,
        s = document.createElement('script');

    try {
        s[('readyState' in s) ? 'onreadystatechange' : 'onload'] = clearScript;

        s.src = U;

        h.insertBefore(s,h.firstChild);
    }
    finally {
        s = h = null;
    }
}

document.getElementById('go').onclick = function () {
    if (timer) {
        clearInterval(timer);
        timer = null;

        update();
        clearInterval(update_int);
        update_int = null;

        this.innerHTML = "Start";
    } else {
        timer = setInterval(function () {
            getScript('script_leak.php?x='+(new Date().getTime()));
        },200);

        update_int = setInterval(function () {
            document.getElementById('inc').innerHTML = count;
            document.getElementById('remaining').innerHTML = document.getElementsByTagName('script').length;
        },1000);

        this.innerHTML = "Stop";
    }
};
</script>
</body>
</html>