Re: [Json] JSON Schema Language

[Anders Rundgren <[e-mail]>]

Since I have what appears to be the diametrically opposed opinion,
I'd better leave it there [*].  Flame-wars are kinda boring.

That a stricter interpretation (=subset) could break parsers
is beyond my understanding.  Probably it is because you seem
to plot with a different order of execution.

Or maybe the problem is rather than you want to apply schemas to
existing systems, [almost] irrespective how broken these may be?

In my book you start by validating data (strict) and only if that
works out you dispatch it to your good ol' JSON tools and applications
to digest.  For verifying serializations you would do the opposite.

Anyway, I wish you good luck with your project!

cheers,
Anders

*] I won't respond....

On 2019-05-08 18:07, Nico Williams wrote:
> On Wed, May 08, 2019 at 06:40:14AM +0200, Anders Rundgren wrote:
>> I believe I have stated my opinion but here it is again:
>>
>> The idea is not changing JSON in any way but to add *strictness* to
>> its interpretation.
> 
> That's not an innocent change.  It breaks interoperability.
> 
>> This has ZERO impact on JSON parsers but could in some cases affect
> 
> Not true: it makes them reject more inputs, which makes them less
> interoperable.
> 
>> JSON serialization.
> 
> Exactly, so not OK.
> 
>> There are no problems defining a set of numeric types having a range
>> and syntax that conforms to what is the norm for most programming
>> languages since all of that fits nicely into JSON.
> 
> This is insufficient justification for making JSON parsing more strict.
> 
>> However, assuming that any JSON parser also could be used for
>> *implementing* JSL (including verifying arbitrary input data) will set
>> the bar way too low.
>>
>> Regarding the integer discussion I believe Open API/Swagger which has
>> quite an active and big user community already flags 3.0 as an invalid
>> integer.  A bug report on Open API (or on Jackson) would most likely
>> return "works as intended".
> 
> That's an interop bug.  They should fix it.
> 
> Nico
>