Re: [apps-discuss] draft-pbryan-json-patch-04 - comments

[Mike Acar <[e-mail]>]

On 03/13/2012 08:18 AM, Paul C. Bryan wrote:
> On Mon, 2012-03-12 at 13:53 -0700, Mike Acar wrote:
>
>> I'm unsure about test as well. It seems strange to me to express "change
>> this document if it looks like this" in a patch; if the application
>> creating the patch doesn't know how the document looks, how can it
>> create a patch?
>
> I'm surprised that it seems so strange, as this is very much like how
> text-based diff/patches work.

But in diff, the "test" is not optional. The diff (AFAIK anyway) always 
expresses "replace this with that". Patch doesn't.

So I guess I should say it's the optionality that makes it a bit weird 
to me - mixing tested and not-tested operations. When would that make sense?

I'm not yet convinced it's a problem, but I wonder about the desired 
semantics and if this is sufficient to get them. For example, I can see 
that test combined with Patch's atomicity could be useful, but it seems 
like it can't protect you from race conditions. Maybe it's sufficient 
for some use cases, though.

> It allows assumptions about the state of the resource being modified
> to be a precondition to successfully modifying the resource. With
> HTTP, it's ideal if precondition headers be used, but this is not
> always feasible, nor is HTTP the only avenue for using JSON Patch.

Mhm.

>> Paul, is there an archived discussion of the "test" operation,
>> particularly regarding use cases?
>
> This was originally discussed on the JSON Patch Google Group discussion.

I've read some of the archives, but not all, so I'll have another look 
through them for this discussion.

-- 
Mike Acar -                                 - macar at cloudmark dot com