POST-WITH-NO-SIDE-EFFECTS method

Jeffrey Mogul writes:
 > 
 > This suggestion (a new kind of POST method) has a certain elegance
 > to it.  What worries me most is that it is probably incompatible
 > with existing proxies (or do they forward methods that they don't
 > know about?)

According to Roy in previous discussions about this, proxies are not
allowed to forward methods they don't know about.  I was told by
others (Ari?) that this is "for security reasons" but I don't believe
those reasons.  (I would like to discuss this as a general issue in
connection with extensibility, but not yet)

  I would like to operate under the assumption that
 > we will never entirely get rid of HTTP/1.0 proxies, since as far
 > as I can tell, the computer industry has never entirely gotten
 > rid of anything.
 > 
Yes, well put.

 > I suppose we might be able to figure out techniques that decide
 > whether the entire request chain is HTTP/1.1, and allow the
 > browser to send POST instead of POST-W-N-S-E in that case.  But
 > this makes me nervous.
 > 
Don't you mean that the browser would send POST if the chain 
contains any HTTP/1.0 servers?

	This should work with older proxies,
 > since they presumably do not ever use their caches to respond
 > to POST methods.  (Or am I wrong?)
 > 

I think you're right.  Things as they stand now would be horribly broken 
if POSTs weren't guaranteed to get through to the origin server.

 > However, I suspect that server implementors will hate this, because
 > it would require them to either provide a mechanism for server
 > administrators to control POST-cachability on a per-URI basis,

Since POSTs are typically handled by scripts, and the scripts generate
some of the headers, this is not that unnatural.  What would be nice
is for proxies and servers that handle POST-W-N-S-E requests to be able to
handle them without invoking the scripts at all unless the appropriate
response was not already cached and fresh.

 > or to simply turn off POST-caching for servers that have any
 > URIs that need POST-W-N-S-E.

??? Not sure what you mean.
It's servers that *support* POST-W-S-N-E that could take advantage
of POST-caching (by which I assume you mean "the ability of caches to
respond to non-side-effecting POSTs without forwarding the request to
the origin server").

--Shel

Received on Friday, 5 January 1996 20:20:39 UTC