Jim Jagielski wrote: > Yes, Martin noted this as well. because of this, and the > Server header fixes, I'd like to see 1.3.25 in relatively short > order once we find out why. From what I can see, we explicitely > *remove* Transfer-Encoding, so I've no idea how it's getting > back in there... yet...
I have been investigating this bug - there are two parts to it. The first problem is that hop-by-hop headers are not being stripped from a backend response, as required by RFC2616. As a result of this, if the backend server sent a chunked response, this gets passed to the frontend, it triggers ap_set_keepalive() to turn on chunking by mistake which results in chunking the chunked data. Removing hop-by-hop headers partially fixes this problem - we no longer get chunked chunked data, we get chunked data instead. This then reveals the second part of the bug. It was assumed that buff.c would handle dechunking of an incoming response - this turns out not to be the case. The required fix is to modify ap_proxy_send_fb() so that it understands how to read a chunked response, and how to dechunk that response correctly. This will get rid of the second "chunked" layer. Busy working on a fix, after I have had some sleep :) Regards, Graham -- ----------------------------------------- [EMAIL PROTECTED] "There's a moon over Bourbon Street tonight..."
smime.p7s
Description: S/MIME Cryptographic Signature