Re: [Json] The names within an object SHOULD be unique.

[Nico Williams <[e-mail]>]

On Thu, Jun 6, 2013 at 5:37 PM, John Cowan <[e-mail]> wrote:
> Nico Williams scripsit:
>
>> Thinking of encoders that do not have full knowledge of an object as
>> it is encoded, I'm uncomfortable with saying "MUST NOT send duplicate
>> keys", since determining whether a given key has already been sent
>> might require large amounts of state.  But note that this concern also
>> applies to parsers!
>
> The amount of state can be reduced probabilistically by storing hashes
> of the values rather than the values themselves.

Or a refinement: Bloom filters.  But I don't think that's a good idea.
 For one the parser would still have to grow the filter (with
adaptable Bloom filters) as it gets too big.  For another, we have the
risk of false positives.  Finally, such a parser could at best produce
an error on [suspected] duplicate key (excuse me, name) usage -- it
could not prevent the *first* value from being used, and if the
consumer has already taken some action based on the first value then
we suddenly have a conflict vis-a-vis parsers that accept only the
last value.  I'd rather accept that streaming parsers can do nothing
about duplicate names.

Nico
--