Global variables are not accessible on load/unload in Firefox

You need to provide a URL where this effect can be observed and steps to reproduce the error.

Although i cannot provide a reproducable demo for this (not yet) i just want to confirm this "behaviour".

We use the latest Prototype (1.5.0_rc0) / Scriptaculous (1.6.1) Combination.

It does not seem easy to reproduce, but it occurs quite often (without Scriptaculous too). I guess the bind function is used somewhere inside Prototype itself before $A is defined.

It seems to me that it often occurs in combination with running AJAX Requests. If a request is not finished yet and you leave or reload the page the message appears.

I've setup a small "barebone" case at http://helpdesk.aps-web.de/undefined_a/ which is only a page which is periodically (every second) requesting the date and time from the server.

The server will need >1s for the ajax answer (the script sleeps 1 second before replying to increase the chance of seeing the error message).

Simply reload the page after a few ajax requests are "in the queue" and you have a VERY GOOD chance to see the message "$A is not defined" in the JS-Debug console of Firefox (Have not seen this error in IE or any other browser so it might be a bug in firefox not finishing AJAX requests timely..?)

I can confirm this behaviour too, and I too think it is related to Ajax requests. The interesting thing is that the errors happens when the page freshly loads, e.g. after I pressed CTRL-F5 for cache bypassed reload. Usually you expect the clean page to not habe errors. Maybe it's FF-centric bug? Cannot test IE with our application here.

I can reproduce this from the test case (provided by sp@aps-web.de) after the first refresh (FF2). This seems like a browser bug, although I'm not sure.

My guess is that Firefox continues to make requests while the window unloads in a refresh. Very strange indeed

We VnRuby use Prototype 1.5.0_rc0 and Scriptaculous 1.6.5 at http://vnruby.org (you can go there to reproduce the error).

When one clicks on one of the 4 links to go to another page, Firebug 0.4.1 (on Firefox 2.0 on Windows XP) sometimes reports: $A is not defined, prototype.js (line 71).

I agree with mislav. My guess is that Firefox continues to process the JavaScript of old page while loading the new page.

Definitely a Firefox bug. I think this is it: 361271.

A workaround (not to show errors) may be possible, but I can't seem to think of one now. Maybe storing the reference to $A somewhere else than on the window object?

*sigh*

I don't think fixing this is within our range. The bug doesn't break apps, though.

Please forgive the interjection of a total newbie (I'm not even sure it's appropriate to comment when a ticket has been closed!)

After reading this, I realized I was provoking my own version of this error by trying an Ajax fetch of a signal file on unload (to log the fact that the user left the page).

On thinking about it, I shouldn't be surprised that this fails -- I mean, who's gonna be around to catch the callback response??!!

Replying to jfkelley:

Please forgive the interjection of a total newbie (I'm not even sure it's appropriate to comment when a ticket has been closed!)

Comments are always welcome.

After reading this, I realized I was provoking my own version of this error by trying an Ajax fetch of a signal file on unload (to log the fact that the user left the page). On thinking about it, I shouldn't be surprised that this fails -- I mean, who's gonna be around to catch the callback response??!!

Maybe you can try doing it synchronous, but I wouldn't recommend it since it will lag down browsing.

Doing stuff (even crazy hacks) on unload is possible, though. You'll have to investigate.

Hello, I've found a solution to all your problems, just go to my site and grab the code: http://www.skybyte.net/articles/prototype.js/

(In [7986]) prototype: Prevent a Firefox bug from throwing errors on page load/unload. Closes #5393, #9421.

This bug still occurs in Prototype 1.6.0.1 and Mac Firefox 2.0.0.11. I get it with 100% reproducibility if there's an Ajax.Request call in a window.unload handler.

For example, load this page, watch the Javascript console, and browse to another page with prototype.js on it:

<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN"
   "http://www.w3.org/TR/html4/loose.dtd">
<html>
<title>Test</title>
<body>
  
<script src="javascripts/prototype.js" type="text/javascript"></script>
<script type="text/javascript">
  function unloader()
  {
    new Ajax.Request('/example/page',
    {
       method: 'get',
       parameters: { gone : 1 }
    });
  }

  Event.observe(window, 'load', function() {
    Event.observe(window, 'unload', unloader);
  });
</script>

  <p>Test page</p>
</body>
</html>

The skybyte patch does fix the problem - reproducing here in case that site goes away.

In prototype.js, change the bind definition to:

Function.prototype.bind = function() {
  var __method = this, args = $A(arguments), object = args.shift();
  return function() {
    if (typeof $A === 'function')
      return __method.apply(object, args.concat($A(arguments)));
  }
}

I did some pretty extensive research on the problem.

It appears that just before the whole page is loaded AND just before unload Firebug randomly(?) loses the scope of global functions / Objects. In other words: it doesn't know that "$A" or "bindAsEventListener" or whichever function belongs to the window object.

It sometimes also loses methods / props within objects. For example 'Event.observers'. Event is defined but Event.observers is not.

Even it's own console.log doesn't work from time to time.

You can easily test it:

Function.prototype.bind = function() {
  var __method = this, args = window.$A(arguments), object = args.shift();
  return function() {
    return __method.apply(object, args.concat(window.$A(arguments)));
  }
}

should be working like a charm.

The above workaround is just a patch. Problems will occur in different places like document.defaultView.getComputedStyle, unloadCache, anywhere where a global reference is used without 'window.'.

It is almost impossible and not worthy patching the entire framework. And as you figured it is not a prototype only problem, any script (specially with mouseovers on links) suffers from it.

I consider it a firebug problem, a little flaw in this lovely tool, too bad.

Some of us still believe that this is a Firefox problem related to bad unloading. It doesn't affect the apps, though. Annoying as it may be, I don't think that anyone found it harmful.

Firefox bug tracker could be a source of more info.

I was able to reproduce the error on a machine without Firebug, so it doesn't appear to be anything to do with that tool.

I'm speculating here, but is this occuring because the wrong page is handling the Ajax response (ie, the new one rather than the old)? If so, can you abort the response in general once this condition is recognized, as a single place to patch the problem rather than having to patch all over the place?

I checked on a FF without firebug installed too, and yes, I can reproduce it, which makes the problem a lot worse of course.

In my script there aren't any Ajax calls. It's just a simple hover / unhover with Scriptaculous Effect.Highlight. It lo

I will post a link to the simplified version of the script soon.

MarkD: with which version of Prototype is this an issue ?

It is Prototype version 1.5.1.1 although I really doubt if the version matters.

The problem still occurs in Prototype 1.6.

MarkD: yes it does, the hover in/out problem should be corrected in Proto 1.6 afaik.

If you make an Ajax-call in window.onunload (like I'm doing, because I want to save the settings of the user when leaving the site), you should really make a synchronous call, even it blocks the browser for the time of the request. Otherwise there will be errors on the next page.

Okay, strange. I'm getting this error in combination with ModalBox, but only if I pass it a certain string to display.

I can't figure this out for the life of me, but it's breaking my app (modalbox just doesn't appear).

firefox 2.0 with firebug (occurs with it disabled as well)

Also occurs on firefox 3 beta 4

And IE7

I've resolved it.. looks to have been related to attempting to display an unordered list. Weirdest thing I've ever seen...

Does this still happen with 1.6.0.2?

may be related to #9211

This problem is still reproducable with Firefox 2.0.0.13 and Prototype 1.6.0.2 when using Ajax.Request on a slow connection and making multiple requests at the same time.

It appears that my issue was unrelated to this. If possible, please remove my comments.

I apologize for the misleading information.