Re: [apps-discuss] json-pointer #5 - semantics [was: Feedback on draft-ietf-appsawg-json-pointer-00]

[Martin Thomson <[e-mail]>]

On 31 May 2012 09:42, Murray S. Kucherawy <[e-mail]> wrote:
> I think the right approach is to say explicitly in pointer that it's left to the application using pointer to decide whether a match failure is an error or should be handled in some other way.  There are some cases where one might abort totally, or might create the path being referenced.  Patch could, for example, abort on a "change" instruction that references a nonexistent token, while "add" referencing a nonexistent token could create the intermediate objects needed to contain the new token.

I disagree with what appears to be an emerging consensus, for one
reason.  There is an inherent ambiguity in pointers that requires a
concrete document to resolve.  Numerical indices can be used in two
contexts:

  /1 => "foo"
  { "1": "foo" }
  [ "foo" ]

Without a concrete document, I can't decide what the pointer "points" at.

That said, there is no stopping someone from going off-spec.  But I'd
prefer to keep it just that: off-spec.