Re: [core] Patch idempotent?

[Carsten Bormann <[e-mail]>]

Hi Peter,

I also believe that, if we are going down the idempotent alley, we'll
need to distinguish PATCH (not necessarily idempotent, like HTTP PATCH)
from iPATCH (idempotent, something new).  Whether, in the end, we want
to add both to CoAP: I don't have an opinion on that yet.

A few more data points:

RFC 6902 "add" is not necessarily idempotent.

See
http://tools.ietf.org/html/rfc6902#appendix-A.2
for an illustration.

(Doing the patch again leads to
   { "foo": [ "bar", "qux", "qux", "baz" ] })

Similar with "remove":

http://tools.ietf.org/html/rfc6902#appendix-A.4

In contrast, RFC 7396 operations always appear to be idempotent, but
then, there are no array operations in RFC 7396 at all.  Positional
array operations are harder to get idempotent because positions change
because of the operation.  Getting idempotency for array operations
would require some content searching (i.e., if "the item" is already
there, don't add it again; if it is already gone, don't remove another
one in its place -- this requires a way to express what is meant by "the
item").

Grüße, Carsten