I do not feel there is a gain from adjusting the syntax, in the role of
JSON Pointer as a stand alone specification. The addition of such a change
adds an implied checking of a list vs a collection, as part of pointer
resolution.
If you are using the pointer as a means to get a value, it makes little
difference in the fact that the value either exists at the end of that path
or it doesn't. (Get me the value at /a/b/c/2)
If you are using the pointer as a method of identity, again it makes little
difference as only one point can presented by that pointer. (You are at
/a/b/c/2)
For me, these are the primary uses of the JSON Pointer specification.
Rarely is the underlying structure, in practice, of consequence.
Something like JSON-Schema is much better for testing the shapes of JSON
data.
However, I can see the point in such a change when paired with JSON-Patch.
But I still feel that even this is more of an attempt to overload the role
of the pointer, to make it carry additional meaning that should be
expressed in either the test or the operation keywords of that particular
specification.
My vote is to leave the specification syntax as is, and keep it simple to
produce and consume, even at the cost of not being able to infer the
structure of any given step in the pointer without the actual json object
being traversed.
--
Matthew P. C. Morley
On Tue, Dec 18, 2012 at 12:24 AM, James M Snell <jasnell(_at_)gmail(_dot_)com>
wrote:
What I'm not clear on, however, is what significant benefit making this
kind of change would provide. Yes, the syntax can be more verbose and exact
but it's not clear that the change is worth the cost.
Robert mentioned that the addition of the json-predicates test doubled the
size of the json-patch file, and yes, I get that.. but if we're talking
about sending a two-object json-patch document to modify a JSON document
many times larger than that, it's still worth the extra handful of bytes it
takes. Sure, if we had a complex patch document consisting of large amounts
of json-patch and json-predicate objects being applied to an arbitrarily
structured document, I could understand the concern, but in that case one
would have to question the wisdom of doing a partial update in the first
place. JSON-Patch documents are best served small and simple, and the
inclusion of a json-predicate or two per instance is not going to blow the
budget on bits on the wire.
So the key question becomes: what added benefit does a more verbose
pointer syntax provide us? And is there a concrete need for that benefit
that can be demonstrated?
- James
On Mon, Dec 17, 2012 at 9:12 PM, Manger, James H <
James(_dot_)H(_dot_)Manger(_at_)team(_dot_)telstra(_dot_)com> wrote:
If we were starting from scratch and defining JSON Pointer again I would
argue for distinguishing array indices and object names in the syntax. For
instance, prefix an object name with "/" and an array index with ":".
json-pointer = *segment
segment = "/" name / ":" index
name = *( unescaped / escaped )
unescaped = %x00-2E / %x30-39 / %x3B-7D / %x7F-10FFFF
escaped = "~" ( "0" / "1" / "2" )
index = "0" / %x31-39 *(%x30-39) / "-"
1. It makes parsing marginally harder: you cannot just split on "/" and
unescape each segment.
2. It doesn't make much difference for selecting a value from some JSON,
or for finding a spot to insert a new value.
3. It would allow you to automatically create object *or array* ancestors
when setting a new value (eg adding 23 at /a:0/b:0 to {} could give
{"a":[{"b":[23]}]}).
4. It might encourage better validation of pointers, but that is probably
wishful thinking.
But JSON Pointer drafts have used the /{name|index} format for a year.
There are a bunch of implementations. The difference is minor in most
circumstances. So while I would be happy to change, I am also comfortable
staying with the current pointer syntax.
There's no good reason for it to be that way, is there?
I don't think so.
--
James Manger
_______________________________________________
apps-discuss mailing list
apps-discuss(_at_)ietf(_dot_)org
https://www.ietf.org/mailman/listinfo/apps-discuss