- From: David W. Morris <dwm@xpasc.com>
- Date: Wed, 11 Jun 1997 10:54:27 -0700 (PDT)
- To: http-wg@cuckoo.hpl.hp.com
- Cc: http-wg@cuckoo.hpl.hp.com
On Wed, 11 Jun 1997, Scott Lawrence wrote: > > >>>>> "JF" == John Franks <john@math.nwu.edu> writes: > > JF> Would it be acceptable to say that the server can check to see if > JF> it has already received PUT or POST data from the client and if it > JF> has the server MAY choose not to send "100 CONTINUE"? This would > JF> at least permit the server not to send "100 CONTINUE" when the > JF> POST data arrives in the same packet as some of the headers. > > This has pretty serious implementation problems; it may be that the > POST has no body, and any data pending in TCP is actually another > request. AHa ... a use for that extra CRLF following a POST. Just more justification for my proposal to make the client declare if its waiting which I think is cleaner. But in any case how is this a problem for the client. It just receives the actual response and no 100 CONTINUE since it has nothing else to send. > Most of the objections raised here to all this have been to the > requirement that the 100 response be sent, not that the client wait > for it before sending the body - note that sending the 100 response > has been a requirement for some time. The client wait issue was largely dealt with by Roys clarification that clients can just ignore the issue and forge ahead. I object to the extraneous transmissions of 100 CONTINUE. I also object to the implication that 100 CONTINUE somehow improves the data transfer integrity when in fact the the real issue is that clients SHOULD wait for the response to a non-idempontent transaction. Trying to somehow mitigate a problem with pipelined non-itempotent requests is what got all of us looking closely at the rationale for 100 CONTINUE and it's implications. Dave Morris
Received on Wednesday, 11 June 1997 10:59:43 UTC