Re: microprofile openapi @asf?

[Romain Manni-Bucau]

Le dim. 24 juin 2018 21:59, Andriy Redko  a écrit :

> Hi Romain,
>
> Just went through the issues and comment threads. I am not really involved
> in MP (sadly)
> but the YAML+JSON discussion makes sense to me, at least from the platform
> perspective. JSON
> should be a must, YAML is optional (although it is very popular in OpenAPI
> community). My personal
> position regarding the builder vs annotations is a matter of choice /
> preference. There are
> centainly pros and cons of both, valid arguments are listed. I don't think
> either of them is
> perfect for everyone, supporting both options sounds like a good
> trade-off, let devs pick whatever
> fits better to the particular project / context.
>

It assumes the ee5 pattern and forgets the cdi/event ones. I agree it is
not yet mainstream but it is a convergence opportunity. In particular when
you see all the reference workarounds annotation require and an
event/programmatic solution doesnt. It is a huge gain in practise if you
have endpoints using the same models.

Kind of theory vs practise feedback probably.


> The issue related to model serialization takes unexpected turn towards
> https://github.com/OpenAPITools
> project ... I don't know the full details but afaik these guys are forking
> Swagger projects (swagger-codegen notably)
> and rebranding under OpenApiTools umbrella. I am working on the PR
> https://github.com/swagger-api/swagger-codegen-generators/pull/101 to
> replace code generation of old Swagger / OpenAPI 2.x with OpenAPI 3.x
> (since Apache CXF
> supports that out of the box). If things work out here as expected, I
> would be happy to help to introduce MP part
> (server stubs or/and client) as well.
>

Yep. My concern here is that using a custom serializer leads to limit the
spec usage to the spec endpoint. It is likely 20% only of the main usages
so spec will likely be replaced by something else anyway if it stays as
such :(.


> Thanks.
>
> Best Regards,
> Andriy Redko
>
> RMB> Hi guys,
>
> RMB> opened several issues about the spec and a few of them are serious
> concerns
> RMB> for me (others are easier):
>
> RMB> 1. https://github.com/eclipse/microprofile-open-api/issues/231
> RMB> 2. https://github.com/eclipse/microprofile-open-api/issues/230
> RMB> 3. https://github.com/eclipse/microprofile-open-api/issues/228
>
> RMB> Seems like there was no time to think about an API so swagger was just
> RMB> copied (and adapted to openapi) which leads to something quite
> inconsistent
> RMB> for end users and also inconsistent with the platform.
> RMB> It doesn't prevent us to implement it but would be great if some of
> you can
> RMB> check out issues and potentially vote for them. There is no Strong API
> RMB> stability requirement at microprofile so there is stilla  hope the
> API is
> RMB> made simpler and usable by end users.
>
> RMB> In short (if you don't want to open the links) the issues are:
>
> RMB> 1. YAML is mandatory but there is nothing standard to modelize it so
> it is
> RMB> an internal of the implementation and the format is not user friendly
> until
> RMB> you use something outside the spec
> RMB> 2. The model is using OpenAPI object graph but it is not integrated
> with
> RMB> JSON-B so it is not (de)serializable correctly for end user. It also
> breaks
> RMB> the JAXRS serialization since each single object of the graph will
> need a
> RMB> custom message reader/writer to work (but the spec doesnt spec about
> that
> RMB> so payloads will not be the expected ones, in particular if you send
> back
> RMB> from a client which got OpenAPI instance some subgraph!)
> RMB> 3. There are 2 API in the spec: a builder one and an annotation
> driven one.
> RMB> The builder is sufficient and associated with a startup event allows
> to
> RMB> avoid the annotations need which just duplicates the builder 1-1 with
> very
> RMB> few semantic differences for ref and map management.
>
> RMB> In one sentence it means that the API could be easier, less ambiguous
> for
> RMB> end users, the integration with the platform more consistent and that
> it is
> RMB> a very simple investment and work. It just needs to be made portable
> RMB> accross vendor.
>
> RMB> Romain Manni-Bucau
> RMB> @rmannibucau  |  Blog
> RMB>  | Old Blog
> RMB>  | Github <
> https://github.com/rmannibucau> |
> RMB> LinkedIn  | Book
> RMB> <
> https://www.packtpub.com/application-development/java-ee-8-high-performance
> >
>
>
> RMB> Le jeu. 21 juin 2018 à 16:20, Raymond Auge 
> a
> RMB> écrit :
>
> >> Great!
>
>
> >> On Thu, Jun 21, 2018 at 10:12 AM, Romain Manni-Bucau <
> >> rmannibu...@gmail.com> wrote:
>
> >>> @Raymond: the diff between CDI and OSGi will be where the OpenAPI
> >>> instance will be created mainly so very doable (aries can even import
> >>> G-openapi for that). Only diff which can be quite intrusive is that @

Re: microprofile openapi @asf?

[Andriy Redko]

Hi Romain,

Just went through the issues and comment threads. I am not really involved in 
MP (sadly)
but the YAML+JSON discussion makes sense to me, at least from the platform 
perspective. JSON 
should be a must, YAML is optional (although it is very popular in OpenAPI 
community). My personal 
position regarding the builder vs annotations is a matter of choice / 
preference. There are 
centainly pros and cons of both, valid arguments are listed. I don't think 
either of them is 
perfect for everyone, supporting both options sounds like a good trade-off, let 
devs pick whatever 
fits better to the particular project / context.

The issue related to model serialization takes unexpected turn towards 
https://github.com/OpenAPITools
project ... I don't know the full details but afaik these guys are forking 
Swagger projects (swagger-codegen notably)
and rebranding under OpenApiTools umbrella. I am working on the PR 
https://github.com/swagger-api/swagger-codegen-generators/pull/101 to replace 
code generation of old Swagger / OpenAPI 2.x with OpenAPI 3.x (since Apache CXF
supports that out of the box). If things work out here as expected, I would be 
happy to help to introduce MP part
(server stubs or/and client) as well.

Thanks.

Best Regards,
Andriy Redko

RMB> Hi guys,

RMB> opened several issues about the spec and a few of them are serious concerns
RMB> for me (others are easier):

RMB> 1. https://github.com/eclipse/microprofile-open-api/issues/231
RMB> 2. https://github.com/eclipse/microprofile-open-api/issues/230
RMB> 3. https://github.com/eclipse/microprofile-open-api/issues/228

RMB> Seems like there was no time to think about an API so swagger was just
RMB> copied (and adapted to openapi) which leads to something quite inconsistent
RMB> for end users and also inconsistent with the platform.
RMB> It doesn't prevent us to implement it but would be great if some of you can
RMB> check out issues and potentially vote for them. There is no Strong API
RMB> stability requirement at microprofile so there is stilla  hope the API is
RMB> made simpler and usable by end users.

RMB> In short (if you don't want to open the links) the issues are:

RMB> 1. YAML is mandatory but there is nothing standard to modelize it so it is
RMB> an internal of the implementation and the format is not user friendly until
RMB> you use something outside the spec
RMB> 2. The model is using OpenAPI object graph but it is not integrated with
RMB> JSON-B so it is not (de)serializable correctly for end user. It also breaks
RMB> the JAXRS serialization since each single object of the graph will need a
RMB> custom message reader/writer to work (but the spec doesnt spec about that
RMB> so payloads will not be the expected ones, in particular if you send back
RMB> from a client which got OpenAPI instance some subgraph!)
RMB> 3. There are 2 API in the spec: a builder one and an annotation driven one.
RMB> The builder is sufficient and associated with a startup event allows to
RMB> avoid the annotations need which just duplicates the builder 1-1 with very
RMB> few semantic differences for ref and map management.

RMB> In one sentence it means that the API could be easier, less ambiguous for
RMB> end users, the integration with the platform more consistent and that it is
RMB> a very simple investment and work. It just needs to be made portable
RMB> accross vendor.

RMB> Romain Manni-Bucau
RMB> @rmannibucau  |  Blog
RMB>  | Old Blog
RMB>  | Github 
 |
RMB> LinkedIn  | Book
RMB> 



RMB> Le jeu. 21 juin 2018 à 16:20, Raymond Auge  a
RMB> écrit :

>> Great!


>> On Thu, Jun 21, 2018 at 10:12 AM, Romain Manni-Bucau <
>> rmannibu...@gmail.com> wrote:

>>> @Raymond: the diff between CDI and OSGi will be where the OpenAPI
>>> instance will be created mainly so very doable (aries can even import
>>> G-openapi for that). Only diff which can be quite intrusive is that @G we
>>> don't use plain reflection to enable CDI meta model to be mutated during
>>> startup and therefore let the user configure most of the model instead of
>>> hardcoding it, but it is not that hard to abstract so I'm very confident to
>>> keep it abstracted and to support OSGi once we support the spec with CDI
>>> (and why not supporting CDI in aries ;)).


>> Regarding supporting CDI in Aries ;) it should look pretty much like any
>> normal CDI extension with a tiny amount of extra OSGi metadata and what I
>> hope are very reasonable restrictions on how extensions provide beans, if
>> any.

>> Sincerely,
>> - Ray



>>> Romain Manni-Bucau
>>> @rmannibucau  |  Blog
>>>  | Old Blog
>>>  | Github
>>> 

Re: microprofile openapi @asf?

[Romain Manni-Bucau]

Hi guys,

opened several issues about the spec and a few of them are serious concerns
for me (others are easier):

1. https://github.com/eclipse/microprofile-open-api/issues/231
2. https://github.com/eclipse/microprofile-open-api/issues/230
3. https://github.com/eclipse/microprofile-open-api/issues/228

Seems like there was no time to think about an API so swagger was just
copied (and adapted to openapi) which leads to something quite inconsistent
for end users and also inconsistent with the platform.
It doesn't prevent us to implement it but would be great if some of you can
check out issues and potentially vote for them. There is no Strong API
stability requirement at microprofile so there is stilla  hope the API is
made simpler and usable by end users.

In short (if you don't want to open the links) the issues are:

1. YAML is mandatory but there is nothing standard to modelize it so it is
an internal of the implementation and the format is not user friendly until
you use something outside the spec
2. The model is using OpenAPI object graph but it is not integrated with
JSON-B so it is not (de)serializable correctly for end user. It also breaks
the JAXRS serialization since each single object of the graph will need a
custom message reader/writer to work (but the spec doesnt spec about that
so payloads will not be the expected ones, in particular if you send back
from a client which got OpenAPI instance some subgraph!)
3. There are 2 API in the spec: a builder one and an annotation driven one.
The builder is sufficient and associated with a startup event allows to
avoid the annotations need which just duplicates the builder 1-1 with very
few semantic differences for ref and map management.

In one sentence it means that the API could be easier, less ambiguous for
end users, the integration with the platform more consistent and that it is
a very simple investment and work. It just needs to be made portable
accross vendor.

Romain Manni-Bucau
@rmannibucau  |  Blog
 | Old Blog
 | Github  |
LinkedIn  | Book



Le jeu. 21 juin 2018 à 16:20, Raymond Auge  a
écrit :

> Great!
>
>
> On Thu, Jun 21, 2018 at 10:12 AM, Romain Manni-Bucau <
> rmannibu...@gmail.com> wrote:
>
>> @Raymond: the diff between CDI and OSGi will be where the OpenAPI
>> instance will be created mainly so very doable (aries can even import
>> G-openapi for that). Only diff which can be quite intrusive is that @G we
>> don't use plain reflection to enable CDI meta model to be mutated during
>> startup and therefore let the user configure most of the model instead of
>> hardcoding it, but it is not that hard to abstract so I'm very confident to
>> keep it abstracted and to support OSGi once we support the spec with CDI
>> (and why not supporting CDI in aries ;)).
>>
>
> Regarding supporting CDI in Aries ;) it should look pretty much like any
> normal CDI extension with a tiny amount of extra OSGi metadata and what I
> hope are very reasonable restrictions on how extensions provide beans, if
> any.
>
> Sincerely,
> - Ray
>
>
>>
>> Romain Manni-Bucau
>> @rmannibucau  |  Blog
>>  | Old Blog
>>  | Github
>>  | LinkedIn
>>  | Book
>> 
>>
>>
>> Le jeu. 21 juin 2018 à 15:21, Raymond Auge  a
>> écrit :
>>
>>> It would be _nice_ if we could figure out a way for this to be usable by
>>> Apache Aries JAXRS Whiteboard [1] which is an implementation of OSGi JAXRS
>>> Whiteboard [2].
>>>
>>> It would seem that a small SPI on the part of Geronimo's mp-openapi
>>> might be enough (so as not to pressure this up onto the mp spec).
>>>
>>> [1] https://github.com/apache/aries-jax-rs-whiteboard
>>> [2] https://osgi.org/specification/osgi.cmpn/7.0.0/service.jaxrs.html
>>>
>>>
>>> On Thu, Jun 21, 2018 at 9:06 AM, Mark Struberg <
>>> strub...@yahoo.de.invalid> wrote:
>>>
 I think it fits well to geronimo.
 The question is rather if CXF is fine with relying on CDI for openapi?
 But since MicroProfile _requires_ CDI I think there is safe to assume
 so.

 LieGrue,
 strub

 > Am 21.06.2018 um 09:59 schrieb Romain Manni-Bucau <
 rmannibu...@gmail.com>:
 >
 > Hello guys,
 >
 > we created a repo for that and to be able to share what we do:
 > https://gitbox.apache.org/repos/asf?p=geronimo-openapi.git
 >
 > I pushed a basic starting structure of the code. The big TODO is the
 > conversion from the model (annotations) to OpenAPI instance (which
 should
 > be somewhere here
 >
 https://gitbox.apac

Mp health?

[Romain Manni-Bucau]

Hi guys,

I will continue openapi spec impl very soon but thanks to a recent thread i
realized health still needs to be impl.

This one is pretty easy compared to last ones we did so I wonder if anyone
wants to do it or if you know a padawan who would be interested in playing
with this kind of work before I move it forward in a few weeks.

Just put a word here. I am happy to help with technical parts if needed
(setup, discuss+repo creation etc).