On Mon, Dec 17, 2012 at 2:01 AM, Robert Sayre <sayrer@xxxxxxxxx> wrote:
[snip]>The first case that can come up is editing documents that have been
> You might argue that someone who (for example) adds to "/foo/1" in the mistaken belief that it's an array, when in fact it's an object, will get surprising results. That's true, but if we were to solve this problem, that person would still need to understand the underlying semantics of "foo" to do anything useful to it -- and I'm not hearing anyone complain about that (I hope).
refactored. It's quite common to refactor JSON documents as below when
a little extra metadata is needed for a collection:
{ "foo":[1,2,3] }
then becomes
{
"foo":
"bar": "baz",
"members": [1,2,3]
}
now, an operation on /foo/1 is quite a different thing in the second
document. It's not too hard to see how a client that's been out of
contact with the server for a while could send an addition to /foo
intended for the array that's been moved down to "members". Now, there
are ways to catch this if you're willing to bloat the patch document
with other checks, or by insisting on a matching ETag. But one of the
nice things about patch formats is that you can try for more
optimistic concurrent editing. This ambiguity in the patch format
makes it needlessly more costly to detect unsound edits.
Appreciate the concrete use case... while it has come up a few times, this is the first actual case that's been presented. If I understand the issue correctly, you're basically talking about a client making a blind partial update to an arbitrarily structured JSON document whose structure may have changed since the last time the client edited it... In such cases, I would argue that you get whatever you end up with.
Everyone's idea of "bloat" is different. Adding etag checking is a simple, effective best practice for detecting concurrent document changes and is likely a really good idea in a scenario where we're making partial modifications to arbitrary documents. Adding in support for json-predicates adds a bit more (not a lot really but opinions vary.. in the ruby implementation it's only approximately 140 more lines of code, mostly formatting) but can make it rather more reliable... For instance,
[
{"op": "type", "path": "/foo/1", "value": "array"},
{"op": "add", "path": "/foo/1", "value": 1}
]
Again, the point is, I don't see this as a problem in practice. Implementers that make blind edits to arbitrary docs can expect surprising results sometimes. That's just the way it is. There are mechanisms defined (test, json-predicates and conditional requests) that can make the edit more reliable.
Another source of bugs from _javascript_ side of things will be
"array-like" objects. These are really quite common (jQuery lists,
arguments object, etc), and authors very often don't realize they
won't be serialized as an array. To be sure, this is a bug on the
author's side, but some checks in the patch format would really help.
They'll probably become even more common as ES6 is standardizing
iterators http://wiki.ecmascript.org/doku.php?id=harmony:iterators.
Here's an example in a JS REPL using the arguments object (but keep in
mind there are many other objects out there that behave this way).
> function foo(list) { console.log(JSON.stringify(list)) }
> function bar(a,b,c) { foo(arguments) }
> bar("a", "b", "c");
{"0":"a","1":"b","2":"c"}
Definitely understand this case, but json-patch is defined in terms of the basic json model and not _javascript_. In other words... again.. the developer using the patch needs to be aware of the structure of the target resource. Json-predicate provides a good approach to dealing with that by allowing a developer to test the structure before attempting to edit.
- James
>Yes, really.
> Put another way -- do you really think
I don't see that requirement in the draft, and I certainly wouldn't
> that people PATCHing something as if it's an array (when in fact it's an object) is a significant, real-world problem, given that the patch author already has to understand the semantics of the document they're patching?
bet on it being true in all cases. Wasn't one of the motivating
use-cases for this patch format CouchDB? That software operates on
arbitrary, user-defined JSON documents.
- Rob
> I don't, and the WG didn't either.
>
> Regards,
>
>
> On 17/12/2012, at 3:36 PM, Robert Sayre <sayrer@xxxxxxxxx> wrote:
>
>> The cost of fixing it seems low, either by changing the path syntax of
>> JSON pointer or changing the names of operations applied to arrays.
>> Array-like objects are common enough in _javascript_ to make this a
>> worry. The other suggestions either assume a particular policy for
>> concurrent edits or require more machinery (test operation etc).
>> Wouldn't it be simpler to make the patch format more precise?
>>
>> - Rob
>>
>> On Sun, Dec 16, 2012 at 4:33 PM, Matthew Morley <matt@xxxxxxxx> wrote:
>>> I am usually lurking and struggling to keep up with these posts. But, I
>>> concur with James, this really is a non-issue in practice.
>>>
>>> The JSON Pointer expresses a path down a JSON object to a specific context.
>>> The Patch expresses a change within or to that context.
>>> Everything about the both standards is about that end context.
>>>
>>> If you want to confirm the type of the context before applying a patch, this
>>> should probably be part of a test operation. I'm not sure if this is
>>> possible at this point (?), but that is where the logic should exist.
>>>
>>>
>>>
>>> On Sun, Dec 16, 2012 at 12:22 AM, James M Snell <jasnell@xxxxxxxxx> wrote:
>>>>
>>>>
>>>>
>>>>
>>>> On Sat, Dec 15, 2012 at 8:36 PM, Robert Sayre <sayrer@xxxxxxxxx> wrote:
>>>>>
>>>>> On Fri, Dec 14, 2012 at 9:17 AM, Markus Lanthaler
>>>>> <markus.lanthaler@xxxxxxx> wrote:
>>>>>>
>>>>>> Hmm.. I think that’s quite problematic. Especially considering how JSON
>>>>>> Pointer is used in JSON Patch.
>>>>>
>>>>> I agree--I provided the same feedback privately. It seems
>>>>> straightforwardly unsound.
>>>>>
>>>>
>>>> In practice it doesn't seem to be much of an issue.
>>>>
>>>> Specifically, if I GET an existing document and get an etag with the JSON,
>>>> then make some changes and send a PATCH with If-Match, the fact that any
>>>> given pointer could point to an array or object member doesn't really matter
>>>> much.
>>>>
>>>> For example:
>>>>
>>>>> GET /the/doc HTTP/1.1
>>>>
>>>> < HTTP/1.1 200 OK
>>>> ETag: "my-document-tag"
>>>> Content-Type: application/json
>>>>
>>>> {"1":"foo"}
>>>>
>>>>> PATCH /the/doc HTTP/1.1
>>>> If-Match: "my-document-etag"
>>>> Content-Type: application/json-patch
>>>>
>>>> [{"op":"add","path":"/2","value":"bar"}]
>>>>
>>>> Generally speaking, someone should not be using PATCH to perform a partial
>>>> modification if they don't already have some knowledge in advance what they
>>>> are modifying. The only time the apparent ambiguity becomes an issue is when
>>>> a client is blindly sending a patch to an unknown endpoint... in which case,
>>>> you get whatever you end up with.
>>>>
>>>> - James
>>>>
>>>>
>>>>>
>>>>> - Rob
>>>>>
>>>>>
>>>>>>
>>>>>> --
>>>>>>
>>>>>> Markus Lanthaler
>>>>>>
>>>>>> @markuslanthaler
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> From: James M Snell [mailto:jasnell@xxxxxxxxx]
>>>>>> Sent: Friday, December 14, 2012 5:41 PM
>>>>>> To: Markus Lanthaler
>>>>>> Cc: IETF Discussion; IETF Apps Discuss
>>>>>> Subject: Re: [apps-discuss] Last Call:
>>>>>> <draft-ietf-appsawg-json-pointer-07.txt> (JSON Pointer) to Proposed Standard
>>>>>>
>>>>>>
>>>>>>
>>>>>> JSON Pointer does not distinguish between objects and arrays. That is
>>>>>> not determined until the pointer is applied to an actual object instance...
>>>>>> the pointer "/1" is valid against {"1":"a"} or ["a","b"]
>>>>>>
>>>>>>
>>>>>>
>>>>>> On Fri, Dec 14, 2012 at 2:51 AM, Markus Lanthaler
>>>>>> <markus.lanthaler@xxxxxxx> wrote:
>>>>>>
>>>>>> I've asked that before but didn't get an answer. So let me ask again
>>>>>> (even
>>>>>> though I'm quite sure it has already been asked by somebody else).
>>>>>>
>>>>>> How does JSON Pointer distinguish between objects and arrays? E.g.
>>>>>> consider
>>>>>> the following JSON document:
>>>>>>
>>>>>> {
>>>>>> "foo": "bar",
>>>>>> "1": "baz"
>>>>>> }
>>>>>>
>>>>>> As I read the draft, the JSON Pointer "/1" would evaluate to "baz" even
>>>>>> though that's probably not what the author intended. Is there a way to
>>>>>> avoid
>>>>>> that?
>>>>>>
>>>>>>
>>>>>> Thanks,
>>>>>> Markus
>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>> Markus Lanthaler
>>>>>> @markuslanthaler
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>> -----Original Message-----
>>>>>>> From: apps-discuss-bounces@xxxxxxxx [mailto:apps-discuss-
>>>>>>> bounces@xxxxxxxx] On Behalf Of The IESG
>>>>>>> Sent: Tuesday, December 11, 2012 4:01 PM
>>>>>>> To: IETF-Announce
>>>>>>> Cc: apps-discuss@xxxxxxxx
>>>>>>> Subject: [apps-discuss] Last Call: <draft-ietf-appsawg-json-pointer-
>>>>>>> 07.txt> (JSON Pointer) to Proposed Standard
>>>>>>>
>>>>>>>
>>>>>>> The IESG has received a request from the Applications Area Working
>>>>>>> Group
>>>>>>> WG (appsawg) to consider the following document:
>>>>>>> - 'JSON Pointer'
>>>>>>> <draft-ietf-appsawg-json-pointer-07.txt> as Proposed Standard
>>>>>>>
>>>>>>> The IESG plans to make a decision in the next few weeks, and solicits
>>>>>>> final comments on this action. Please send substantive comments to
>>>>>>> the
>>>>>>> ietf@xxxxxxxx mailing lists by 2012-12-25. Exceptionally, comments
>>>>>>> may
>>>>>>> be
>>>>>>> sent to iesg@xxxxxxxx instead. In either case, please retain the
>>>>>>> beginning of the Subject line to allow automated sorting.
>>>>>>>
>>>>>>> Abstract
>>>>>>>
>>>>>>>
>>>>>>> JSON Pointer defines a string syntax for identifying a specific
>>>>>>> value
>>>>>>> within a JSON document.
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> The file can be obtained via
>>>>>>> http://datatracker.ietf.org/doc/draft-ietf-appsawg-json-pointer/
>>>>>>>
>>>>>>> IESG discussion can be tracked via
>>>>>>>
>>>>>>> http://datatracker.ietf.org/doc/draft-ietf-appsawg-json-pointer/ballot/
>>>>>>>
>>>>>>>
>>>>>>> No IPR declarations have been submitted directly on this I-D.
>>>>>>>
>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> apps-discuss mailing list
>>>>>>> apps-discuss@xxxxxxxx
>>>>>>> https://www.ietf.org/mailman/listinfo/apps-discuss
>>>>>>
>>>>>> _______________________________________________
>>>>>> apps-discuss mailing list
>>>>>> apps-discuss@xxxxxxxx
>>>>>> https://www.ietf.org/mailman/listinfo/apps-discuss
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> _______________________________________________
>>>>>> apps-discuss mailing list
>>>>>> apps-discuss@xxxxxxxx
>>>>>> https://www.ietf.org/mailman/listinfo/apps-discuss
>>>>>>
>>>>
>>>>
>>>>
>>>> _______________________________________________
>>>> apps-discuss mailing list
>>>> apps-discuss@xxxxxxxx
>>>> https://www.ietf.org/mailman/listinfo/apps-discuss
>>>>
>>>
>>>
>>>
>>> --
>>> Matthew P. C. Morley
>> _______________________________________________
>> apps-discuss mailing list
>> apps-discuss@xxxxxxxx
>> https://www.ietf.org/mailman/listinfo/apps-discuss
>
> --
> Mark Nottingham http://www.mnot.net/
>
>
>
_______________________________________________
apps-discuss mailing list
apps-discuss@xxxxxxxx
https://www.ietf.org/mailman/listinfo/apps-discuss