IE problems with handleRequest

[bcc]

Sometimes, IE6 does not send the parameters when POSTing the chat page. It does not send any parameters (or values), such as handleRequest. As a result, the xml response contains the whole html of the chat page, as if it were a regular request.

I don't know how to reproduce the problem, but if I leave IE running for a while, it happens quite often.


Can anyone else confirm it?

[bcc]

I had it confirmed by several people. At some point, IE doesn't make a proper request and everything gets screwed up.

[phpfreechat]

It's a bad news for IE6 users ... if it's not reproducable I don't know how to handle this problem.
Moreover, did you encounter the problem on the official demo page ? Maybe it's related to the server configuration ?

[bcc]

Hmm, good idea. I'll let it run for a while on the demo page.

[bcc]

I'm running a test now, and not seeing any problems.
Could it be due to slow connections?
For example, if the previous ajax request haven't been completed yet, and it's already time to make the next call. Is there a mechanism to synchronize them?

[phpfreechat]

For example, if the previous ajax request haven't been completed yet, and it's already time to make the next call. Is there a mechanism to synchronize them?

Yes, maybe it should be possible to block next http requests since the last one is not finished.
I will look at this and try to implement something.

[bcc]

I could not reproduce the problem with the demo on this site.
But now, I changed register_globals to off, and I'm no longer having problems with IE.

Could that have any effect or am I just imagining things?

[bcc]

Also, IE doesn't handle caching properly. So that all images that are loaded in style background or by dynamically adding elements to the dom get fetched multiple times.

As a result, user.gif and newmsg.gif are reqeusted from the server many times on the first page load.
It's not the PFC's problem, but I just thought I would mention it.

It's especially bad when you have 150 messages displayed in a buffer when the new visitor logs in.

[bcc]

In pfcclient.js, on line 731
// adjust the refresh_delay if the connection was lost
if (res == false) { this.refresh_delay = this.refresh_delay * 2; }

Does that keep increasing the interval every time a request is not returned?

Wouldn't that eventually make refresh rate something huge if there are some intermittent connection problems?

If that the case, is it possible to add a max_refresh_delay parameter to limit it?

[phpfreechat]

Humm maybe this could be fixed with a better scr/client/proxy.php.tpl implementation... I have a very little knowledge in HTTP headers but it should possible to indicate that the file should not be downloaded again as long as a date is not expired.
I think, an optimisation is possible in this proxy.php file.

[bcc]

Would something like this work for synchronization?

Code: Select all

Index: src/client/pfcclient.js
===================================================================
--- src/client/pfcclient.js     (revision 2049)
+++ src/client/pfcclient.js     (working copy)
@@ -1,6 +1,7 @@
 var is_ie = navigator.appName.match("Explorer");
 var is_khtml = navigator.appName.match("Konqueror") || navigator.appVersion.match("KHTML");
 var is_ff = navigator.appName.match("Netscape");
+var is_update_request_pending=false;

 /**
  * This class is the client part of phpFreeChat
@@ -716,7 +717,19 @@
     var req = cmd+" "+this.clientid+" "+(recipientid==''?'0':recipientid)+(param?" "+param : "");
     if (pfc_debug)
       if (cmd != "/update") alert(req);
-    return eval('pfc_handleRequest(req);');
+       var retVal=null;
+       if (cmd == "/update"){
+               if(is_update_request_pending){
+                       retVal = true;
+               } else {
+                       is_update_request_pending = true;
+                       retVal = eval('pfc_handleRequest(req);');
+                       is_update_request_pending = false;
+               }
+       } else {
+       retVal = eval('pfc_handleRequest(req);');
+       }
+    return retVal;
   },

   /**

Is "/update" only called in one place?

It doesn't help when the client tries to send some other command at the same time as update is happening, but I guess it's better than nothing.

Also, with many channels, that might create some problems.

But I do think that IE problems I'm noticing are due to multiple requests being sent at the same time and IE not handling them properly (especially over the same TCP connection).

I increased the refresh rate to 7 seconds and my server's keep-alive is 5 seconds. And that alone made a lot of users stop complaining. And a few actually e-mailed me to tell me that things are much better.

[phpfreechat]

I've allready done some work on the /update synchronization, it's commited in the subversion.

However I think your code is useless because this line :
retVal = eval('pfc_handleRequest(req);');
will return before the request is sent (because it's asynchronous requests)

Maybe I'm wrong but I doubt...

[phpfreechat]

here is the patch I've written to avoid /update command to overlap :
http://svn.sourceforge.net/viewvc/phpfr ... athrev=833

[bcc]

However I think your code is useless because this line :

Oh yeah, I didn't actually look at the eval itself. I just assumed it's a normal function call.

[bcc]

With you patch, what happens if resposne is not "ok" after the canupdatenexttime is set to false? What happens on the next interval and a call for update?

It seems like it would stop updating for that point on. No?