Bugs fixed in V0.2-2

There have been a few more bugs since the release of the last bug fix version V0.2-1 and which this release attempts to fix. Here is the list of them:

• If redirecting a request containing a query string, then the old query string was added to the new location even if the new location contained a new query string.
• In a proxy redirect (status code 305) the wrong resource was used for the new request.
• A few problems with the chunked encoding were solved. There was a bug were multiple responses on the same connection that used this encoding would be decoded wrong. Furthermore, the parser for the chunk length has been loosend to accept non-conforming chunk headers.
• Most of the timeout thread problems have been worked around. suspend() is not used anymore to avoid Netscape problems and stop() problems, and the thread is set to daemon mode to avoid hanging at exit time.
• The JDK 1.0.2 (and possibly other VMs) has a bug where it doesn't release resources referenced by a thread if the thread is stop()'d instead of being allowed to fall off the end of run(). This meant that the Timeout thread would keep a large part of the resources from being gc'd, thereby creating the equivalent of a memory leak. Therefore the thread is not stop()'d anymore, but instead falls of the end of it's run().
• The parser for responses has been loosend up. It now deals with erroneous white space preceding the response, it correctly handles non existent reason phrases, and it also handles responses from buggy NCSA servers. It also now throws a ProtocolException if the response version is not HTTP/0.9 or HTTP/1.* .
• HTTPResponse.getHeaderAsXXX() would not work if they were the first header requested (i.e. no other getHeader() had been called previously). The side effect of this fix is that these two methods may now (correctly) also throw an IOException.
• A couple deadlock problems in the response stream handling have been fixed. One could only turn up if you had done a setRawMode(true) for the connection, and the other turned up under certain conditions when an IOException occured.
• Sending a request URI "*" would not work correctly when using a proxy.
• Some (buggy) servers return a 500 status on a HTTP/1.1 request. This situation is now handled by retrying the request as HTTP/1.0 .
• Some (buggy) servers don't like a port specification in the Host: header if the port is the default port (80). The port part is now only sent if a non-default port is used.
• The max parameter in the Keep-Alive header is now honored. This is necessary as Apache would ReSeT a connection (instead of just closing it) if too many requests were sent, resulting in unnecessary IOExceptions.
• The signature of the Codecs.mpFormDataEncode() method has been changed slightly: the last parameter is now an array of NVPairs instead of an NVPair. This was a bug in that NVPair ought to be immutable, but this method violated that.

Thanx to Doron Shpasser and Kuo Chiang for finding and reporting a number of the bugs.