Description: ------------ I use keep-alive and when the server finally closes the connection, the connection is not being re-created and everything goes error. It looks like when the server finally closes the connection like that: Response n-1: Keep-Alive: timeout=15, max=1 Response n: Connection: close HTTP_Request doesn’t recreate the connection. If the server sends Connection: close at the first response the problem does show up Test script: --------------- while (1) { $req = & new HTTP_Request("http://server/test.php"); $req->addHeader("Connection", "Keep-Alive"); $req->sendRequest(); } it will bug when the keep-alive is reached

Hey Romain, can you dig up a bit of the http spec or some articles on what user agents should be doing when a server dishonours the Keep-Alive requests? I don't think the connection should be re-initiated, raising an error / exception is probably a better behaviour - then you can decide if you want to reconnect or not external to http_request.

Also, what's the output when this script fails?

No need for additional feedback, I asked Romain to open this report after confirming bug's existence. Quoting my email: > I had a brief look through the code, it seems that the connection *is* closed properly but the connection object is not purged from the "static" array and then everything breaks on the next request when it is reused.

Cannot reproduce the bug with the following script: ---------- require_once 'HTTP/Request.php'; require_once 'HTTP/Request/Listener.php'; echo '<pre>'; while (true) { $req = new HTTP_Request('http://localhost/images/0.gif'); $req->addHeader('connection', 'Keep-Alive'); $req->setMethod(HTTP_REQUEST_METHOD_HEAD); $req->attach(new HTTP_Request_Listener()); $req->sendRequest(); } echo '</pre>'; ---------- The Listener properly reports disconnect and connect events after the server closes the connection due to timeout or keepalive limit. I suspect there may be some other stuff going on, but I need a self-contained test script to check.

No feedback received, please reopen the bug if you come up with a proper script to reproduce the problem.