Idempotent sequences of methods
Miles Sabin (msabin@cromwellmedia.co.uk)
Fri, 21 Jan 2000 16:47:15 -0000
The discussion in the 'On piplining' thread reminded me of
something I found puzzling in 9.1.
I have a strong suspicion that although 9.1.2 suggests that
an idempotent sequence of methods might contain non-safe
idempotent methods (PUT, DELETE), this can't actually be the
case.
Here's the argument,
* A sequence which contains a PUT/DELETE <uri1> after a
GET/HEAD/TRACE/OPTIONS <uri2> is idempotent only if
uri1 is independent of uri2.
* There is nothing in the definition of an HTTP URI which
allows us to infer that two URIs are independent.
Suppose, for the sake of argument, that a URI path maps
in a natural way onto part of the servers filesystem tree.
In this case, it's clear that we can't infer that,
http://www.foo.com/a/b/c
is independent of,
http://www.foo.com/a
on a purely textual basis because if we were to DELETE the
latter we would have also deleted the former.
Again, with the filesystem mapping assumption, we can't even
assume that,
http://www.foo.com/a/b/c
is independent of,
http://www.foo.com/d/e/f
because they might both reference the same underlying file
in the servers filesystem (eg. via a symlink, or via an
HTTP server path mapping).
So, we can't use textual difference of URIs to infer
independence. Given that we haven't anything else to go on at
the user-agent end I conclude that an idempotent sequence of
methods can't contain any non-safe methods.
QED ;-)
Comments?
Cheers,
Miles
--
Miles Sabin Cromwell Media
Internet Systems Architect 5/6 Glenthorne Mews
+44 (0)20 8817 4030 London, W6 0LJ, England
msabin@cromwellmedia.com http://www.cromwellmedia.com/