Re: [apps-discuss] JSON Patch

["Manger, James H" <[e-mail]>]

>> <[e-mail]>  wrote:
>>> On escaping: how about replacing every '/' in an object member's name with the Unicode REPLACEMENT CHARACTER U+FFFD when creating a JSON pointer.

>On 2011/11/24 11:08, Martin Thomson wrote:
>> Interesting that you choose U+FFFD in the same way that backslash was
>> chosen as an escape character in the first place.  I'm not a big fan
>> of that approach.

On 2011/11/24, Martin J. Dürst wrote:
> Yes, please don't. The semantics of U+FFFD is mostly a character that 
> wasn't successfully converted from some other encoding. Overloading that 
> with "escaping a slash" is a bad idea.

JSON pointer theoretically needs a proper escaping mechanism since it reserves 1 char as a delimiter. However, choosing 1 char not to support avoids a fair amount of inevitable confusion and complexity. I chose U+FFFD REPLACEMENT CHARACTER. If you don’t like that choice, how about U+001F INFOMATION SEPARATOR ONE.

Not supporting pointers for names with, say, a U+FFFD char is not ideal but hinders almost no practical uses.
Escaping '/' with '\/' could be worse. It seems certain to cause significant confusion. You can no longer simply split a pointer on '/' characters, but need to use, say, a regular expression to parse a pointer. A / can be legitimately escaped in a JSON string as "\/", and apparently often is [http://stackoverflow.com/questions/1580647/json-why-are-forward-slashes-escaped] so developers will read and write 0, 1, 2, or 3 backslashes (but only 1 %2F), when they are dealing with a name with a '/', or a '\'. Will "\x" be undefined when x isn’t \ or /?
 
Object member name:
Logically:  Bjørn/Carsten and bar
In JSON  : "Bjørn/Carsten" and "bar"
      Or : "Bjørn\/Carsten" and "bar"

JSON pointer:
Logically:  Bjørn\/Carsten/bar
In JSON  : "Bjørn\\/Carsten/bar"
      Or : "Bjørn\\\/Carsten\/bar"
In IRI   :  Bjørn%2F/Carsten/bar


--
James Manger