[SQU] 304response does not override cached headers.

I use Squid 2.3 Stable 3 on RetHat Linux 6.2.
Now, I am planning to design and implement another Http auth scheme, which needs sending auth-info
from server to client. I decide to use Pragma: header to transfer auth-info to client
(because Authentication-Info: header is not supported by Squid yet).

Squid supports "public", "max-age" and "must-revalidate" cache-control directives,
so I use these directives to cache authorized response. I am very happy.
But, when I (= server) send to 304 response to squid with Pragma: header(including auth-info),
Squid sends old (cached) Pragma: header to client! My authentication sheme does not work!

I refer to RFC of HTTP/1.1.
RFC 2616 (Latest HTTP/1.1 RFC) says:

13.5.3 Combining Headers

   When a cache makes a validating request to a server, and the server
   provides a 304 (Not Modified) response or a 206 (Partial Content)
   response, the cache then constructs a response to send to the
   requesting client.

   If the status code is 304 (Not Modified), the cache uses the entity-
   body stored in the cache entry as the entity-body of this outgoing
   response.

   The end-to-end headers stored in the cache entry are used for the
   constructed response, except that

      - any end-to-end headers provided in the 304 or 206 response MUST
        replace the corresponding headers from the cache entry.

I think that Squid does not obey this rule.

I browse Squid src, then I add the following code to "clientBuildReplyHeader()" in client_side.c:

   /* Append X-Cache */
       .....
    if(is_hit) {
        if(httpHeaderHas(&(http->entry->mem_obj->reply->header), HDR_PRAGMA))
        {
        String s;
        s = httpHeaderGetList(&(http->entry->mem_obj->reply->header),
                                HDR_PRAGMA);
        httpHeaderDelById(hdr, HDR_PRAGMA);
          if(strLen(s)) httpHeaderPutStr(hdr, HDR_PRAGMA, strBuf(s));
        }
    }
       .....

This modification is not perfect, but suitable for my purpose.
My modification is legal to HTTP/1.1 protocol?
How about this problem?

----
Hirohiko Nakano  nakano@hp-info.med.osaka-u.ac.jp
--
To unsubscribe, see http://www.squid-cache.org/mailing-lists.html