I think that the result of the discussion is as follows:
1. support of string values for big numbers by default
2. support the override of properties by default
3. fail fast when non-unique property (after override and rename) is found
4. define default algorithm for override (based on java mechanism) and
check uniqueness of properties (not affecting performance)
5. Do not enable I-JSON by default due to the following recommendations
- support only root objects/arrays by default
- serialize binary data to base64 by default
6. There will be static final constant in Jsonb to define I-JSON
compatibility mode
WDYT? Could we agree on that or is there some point(s) we should discuss
further?
Thanks,
MartinV
On Mon, Apr 20, 2015 at 11:24 PM, Romain MB <rmannibucau_at_tomitribe.com>
wrote:
> 2015-04-20 23:11 GMT+02:00 Eugen Cepoi <cepoi.eugen_at_gmail.com>:
> >
> >
> > 2015-04-20 22:22 GMT+02:00 Romain MB <rmannibucau_at_tomitribe.com>:
> >>
> >> 2015-04-20 20:35 GMT+02:00 Eugen Cepoi <cepoi.eugen_at_gmail.com>:
> >> >
> >> >
> >> > 2015-04-20 19:40 GMT+02:00 Romain MB <rmannibucau_at_tomitribe.com>:
> >> >>
> >> >> I understand, that's where we don't see it the exact same way. You
> >> >> merge it by name where I would merge it by java model with standard
> >> >> order. So basically I'd do a reduce (java) then validation (json
> side
> >> >> if we can call it this way).
> >> >
> >> >
> >> > Ah ok, so you would only merge by the name of the field or method.
> >> > This is viable but would force to make a distinction between resolved
> >> > name
> >> > using java bean convention and other naming strategies (annotation,
> >> > custom
> >> > user strategy etc). For Genson I prefered to have only one name
> >> > resolution
> >> > system and use it always the same way, there are less "rules" involved
> >> > making things easier to understand.
> >> > But it can have some advantages to make the distinction...
> >> >
> >>
> >> yes basically all is resolved on java side then we can check
> >> @JsonProperty (supposed it is the right name) to check the name to
> >> use, if not present use the one deduced from java. Merge is less
> >> aggressive and allows few validations.
> >
> >
> > So by java side resolution you mean we use the same strategy as java?
> > This means that fields/methods from class A overrides what matches in its
> > super class.
> > In short:
> > 1) reduce accessors/mutators by name and definition order in class
> hierarchy
> > 2) apply any "renaming" strategy on the result of 1)
> > 3) reduce the result of 2) using the same strategy as in 1)
> >
> > If yes then I think we can simplify things by directly applying the
> renaming
> > and then doing a single "reduce" that overrides or throws an exception.
> >
>
> well while we have this reduce phase it is fine. Surely an
> implementation detail then.
>
> > Allowing to override has some advantages, for example you can extend a
> class
> > of which you don't have control and redefine how some property would be
> > ser/de - a quick solution vs implementing a full custom ser/de for that
> > enclosing type.
> >
> > Would be interesting to know what others think.
> >
>
> +1
>
> >>
> >> >>
> >> >> Doesnt impact performances.
> >> >
> >> >
> >> > True, doing it on the resolved java model (with renamed or not
> >> > properties)
> >> > doesn't change the perfs.
> >> >
> >> >>
> >> >>
> >> >> Now the real question is: do we gain anything compared to what you
> >> >> propose ie ensuring we don't have conflicts. I still think it can
> make
> >> >> sense to validate there isn't a "model conflict". Implicit merging
> can
> >> >> be hard to debug from my experience.
> >> >
> >> >
> >> > So you would throw an exception for example if there are two
> properties
> >> > with
> >> > the same name in a class and its parent?
> >> > If yes then we should make sure of the "rules" applied as there might
> be
> >> > ambiguous situations. How do you think we should distinguish valid
> from
> >> > invalid conflicts for renamed properties?
> >> >
> >>
> >> We can have rules like "an overriden method is not a conflict if a
> >> field is not hiding parent one" etc but we can also do like in bean
> >> validation and define inheritance API. A bit more verbose but clearer
> >> IMO (+ does it really happen often in real life?). Something like
> >> @OverrideJsonAttribute("parentName").
> >>
> >
> > Having a specific API for overriding looks overkill, I'd prefer just to
> > agree on a good default behaviour.
> >
>
> Let's try - agree on your statement. I guess releasing a 1.0 without
> such API would be good and we can still add it in 1.1 if needed/asked.
>
> >>
> >> >>
> >> >> That said if we ensure the
> >> >> framework logs it properly (level = warn + logger name =
> >> >> javax.json.something) then it would work for me as well.
> >> >
> >> >
> >> > I would prefer to just throw an exception or consider it as valid
> rather
> >> > than log a warn.
> >> >
> >>
> >> Works, just want it to be visible and not too implicit.
> >>
> >> >>
> >> >>
> >> >>
> >> >> Romain Manni-Bucau
> >> >> @rmannibucau
> >> >> http://www.tomitribe.com
> >> >> http://rmannibucau.wordpress.com
> >> >> https://github.com/rmannibucau
> >> >>
> >> >>
> >> >> 2015-04-20 19:34 GMT+02:00 Eugen Cepoi <cepoi.eugen_at_gmail.com>:
> >> >> >
> >> >> >
> >> >> > 2015-04-20 19:28 GMT+02:00 Romain MB <rmannibucau_at_tomitribe.com>:
> >> >> >>
> >> >> >> well not supporting the validation for custom serialization code
> >> >> >> doesnt sound like a big issue for me but it doesn't prevent to
> >> >> >> validate the model. I didn't speak of the merging but for
> instance 2
> >> >> >> different methods/fields decorated with @JsonProperty(name="foo").
> >> >> >
> >> >> >
> >> >> > Yes I was including renamed properties in that and by merging I was
> >> >> > meaning
> >> >> > take N properties (field/method/ctr) and reduce it to one (how you
> >> >> > achieve
> >> >> > it is up to the impl).
> >> >> > To me renamed or original name is the same, as I first resolve the
> >> >> > names
> >> >> > and
> >> >> > then do all the selection/merging stuff.
> >> >> > Amongst others, this allows to override in subclasses things from
> >> >> > parent
> >> >> > classes - by using the same name.
> >> >> >
> >> >> >>
> >> >> >>
> >> >> >> For me just doing it already matches the intention behind I-JSON.
> >> >> >>
> >> >> >> Romain Manni-Bucau
> >> >> >> @rmannibucau
> >> >> >> http://www.tomitribe.com
> >> >> >> http://rmannibucau.wordpress.com
> >> >> >> https://github.com/rmannibucau
> >> >> >>
> >> >> >>
> >> >> >> 2015-04-20 19:24 GMT+02:00 Eugen Cepoi <cepoi.eugen_at_gmail.com>:
> >> >> >> > I think doing it on the model is something else as basically you
> >> >> >> > already
> >> >> >> > merge detected properties, so you end up with only one.
> >> >> >> > At least that's what I do in Genson, I give preference to the
> >> >> >> > detected
> >> >> >> > properties (methods/fields/whatever else) by hierarchy (from
> most
> >> >> >> > concrete
> >> >> >> > to the "highest" parent). And the compiler provides already some
> >> >> >> > safeguards.
> >> >> >> >
> >> >> >> > Still all that doesn't solve the problem of colliding properties
> >> >> >> > being
> >> >> >> > written by the user (via some custom ser. code).
> >> >> >> > But well... we can't always prevent people from doing stupid
> >> >> >> > things
> >> >> >> > :)
> >> >> >> >
> >> >> >> > 2015-04-20 19:11 GMT+02:00 Romain MB <rmannibucau_at_tomitribe.com
> >:
> >> >> >> >>
> >> >> >> >> 2015-04-20 19:05 GMT+02:00 Eugen Cepoi <cepoi.eugen_at_gmail.com
> >:
> >> >> >> >> >
> >> >> >> >> > 2015-04-19 19:44 GMT+02:00 Romain MB
> >> >> >> >> > <rmannibucau_at_tomitribe.com>:
> >> >> >> >> >>
> >> >> >> >> >> Hi Martin
> >> >> >> >> >>
> >> >> >> >> >> for me it makes sense to support it by default with a flag
> >> >> >> >> >> like
> >> >> >> >> >> the
> >> >> >> >> >> one you proposed to switch it off (a bit off topic but do we
> >> >> >> >> >> have
> >> >> >> >> >> constants for these flags, something like
> >> >> >> >> >> Jsonb.I_JSON_COMPLIANCE?).
> >> >> >> >> >>
> >> >> >> >> >> UTF8 by default is quite mandatory IMO, must-ignore as well
> >> >> >> >> >> (already
> >> >> >> >> >> discussed IIRC). ISO8601 is the standard for dates of most
> of
> >> >> >> >> >> frameworks, big numbers need to be string whatever we do so
> >> >> >> >> >> finally
> >> >> >> >> >> open points from my window are:
> >> >> >> >> >>
> >> >> >> >> >> - byte data as base64: on johnzon we discussed it and I
> didn't
> >> >> >> >> >> want
> >> >> >> >> >> it
> >> >> >> >> >> cause it was far from Java - that said it is not a super
> >> >> >> >> >> common
> >> >> >> >> >> type
> >> >> >> >> >> so not sure it is that important
> >> >> >> >> >> - top level construct: don't recall the spec but I think
> >> >> >> >> >> primitives
> >> >> >> >> >> (string, numbers) can be supported as well. Jackson does it
> at
> >> >> >> >> >> least
> >> >> >> >> >> so +1 to not support this fail fast behavior
> >> >> >> >> >
> >> >> >> >> >
> >> >> >> >> > I don't think there is much value in being too restrictive,
> +1
> >> >> >> >> > for
> >> >> >> >> > it.
> >> >> >> >> > I don't remember any discussion about accepting to deser.
> >> >> >> >> > numbers
> >> >> >> >> > as
> >> >> >> >> > strings, strings as numbers etc when possible, but it would
> be
> >> >> >> >> > nice.
> >> >> >> >> >
> >> >> >> >> >>
> >> >> >> >> >> - no members with duplicate name: the only point I don't
> know.
> >> >> >> >> >
> >> >> >> >> >
> >> >> >> >> > This one is painful as it has an impact on performances (esp.
> >> >> >> >> > if
> >> >> >> >> > we
> >> >> >> >> > want
> >> >> >> >> > to
> >> >> >> >> > prevent user custom code to generate such output)... at the
> >> >> >> >> > moment
> >> >> >> >> > I
> >> >> >> >> > don't
> >> >> >> >> > support it in Genson for this reason.
> >> >> >> >> >
> >> >> >> >>
> >> >> >> >> Can't we do it on the model (ie once for the whole application
> >> >> >> >> lifecycle)? I guess here we are lucky compared to JSON-P.
> >> >> >> >>
> >> >> >> >> >>
> >> >> >> >> >> Intellij Idea makes this kind of JON invalid. This is
> >> >> >> >> >> something
> >> >> >> >> >> we
> >> >> >> >> >> would need to be able to switch off but I think it makes
> sense
> >> >> >> >> >> to
> >> >> >> >> >> have
> >> >> >> >> >> it by default to detect wrong models.
> >> >> >> >> >>
> >> >> >> >> >>
> >> >> >> >> >> Romain Manni-Bucau
> >> >> >> >> >> @rmannibucau
> >> >> >> >> >> http://www.tomitribe.com
> >> >> >> >> >> http://rmannibucau.wordpress.com
> >> >> >> >> >> https://github.com/rmannibucau
> >> >> >> >> >>
> >> >> >> >> >>
> >> >> >> >> >> 2015-04-18 19:53 GMT+02:00 Martin Vojtek <voytoo_at_gmail.com
> >:
> >> >> >> >> >> > Hi Experts,
> >> >> >> >> >> >
> >> >> >> >> >> >
> >> >> >> >> >> > JSON-B should provide support or some level of compliance
> >> >> >> >> >> > with
> >> >> >> >> >> > I-JSON
> >> >> >> >> >> > specification.
> >> >> >> >> >> >
> >> >> >> >> >> > https://tools.ietf.org/html/rfc7493
> >> >> >> >> >> >
> >> >> >> >> >> > There are several things to discuss.
> >> >> >> >> >> >
> >> >> >> >> >> > Should JSON-B support I-JSON by default? My proposal is to
> >> >> >> >> >> > not
> >> >> >> >> >> > support
> >> >> >> >> >> > all
> >> >> >> >> >> > the recommendations of I-JSON by default.
> >> >> >> >> >> >
> >> >> >> >> >> > If we agree on that, what specific parts of I-JSON should
> >> >> >> >> >> > JSON-B
> >> >> >> >> >> > provide
> >> >> >> >> >> > by
> >> >> >> >> >> > default?
> >> >> >> >> >> >
> >> >> >> >> >> >
> >> >> >> >> >> > Proposal of parts (of I-JSON) to support by JSON-B by
> >> >> >> >> >> > default:
> >> >> >> >> >> >
> >> >> >> >> >> > Should be supported by default:
> >> >> >> >> >> > - Encoding and Characters (UTF-8 by default)
> >> >> >> >> >> > - Object constraints (no members with duplicate name)
> >> >> >> >> >> > - MUST-IGNORE policy - partial mapping
> >> >> >> >> >> > - Time and Date Handling - serialize accoring to I-JSON
> >> >> >> >> >> >
> >> >> >> >> >> > Should not be supported by default:
> >> >> >> >> >> > - Numbers - serialize big number(s) into string (and deser
> >> >> >> >> >> > given
> >> >> >> >> >> > strings
> >> >> >> >> >> > into number)
> >> >> >> >> >> > - Top-Level Constructs - fail fast when ser/deser
> something
> >> >> >> >> >> > different
> >> >> >> >> >> > than
> >> >> >> >> >> > object/array as top-level JSON
> >> >> >> >> >> > - binary data encoded as a string in base64url
> >> >> >> >> >> >
> >> >> >> >> >> >
> >> >> >> >> >> > To support I-JSON fully in some case, I propose to provide
> >> >> >> >> >> > I-JSON
> >> >> >> >> >> > compatibility mode available via config property.
> >> >> >> >> >> >
> >> >> >> >> >> > JsonbConfig config = new
> >> >> >> >> >> > JsonbConfig().setProperty("jsonb.i-json.compliance",
> true);
> >> >> >> >> >> > Jsonb jsonb = JsonbBuilder.create(config);
> >> >> >> >> >> >
> >> >> >> >> >> > Looking forward to your feedback.
> >> >> >> >> >> >
> >> >> >> >> >> > MartinV
> >> >> >> >> >
> >> >> >> >> >
> >> >> >> >
> >> >> >> >
> >> >> >
> >> >> >
> >> >
> >> >
> >
> >
>