Hi Kevin,

1.       Ok then we will leave the interface in the API. But I will address 
this at the OData committee since there should be no information difference 
between the two documents.

2.       Yes please think about it and let me know if you have another idea. In 
the meantime I think that your arguments about modularity weigh more than 
having just one internal class because it looks nice from the outside. So 
please go ahead with the approach you have started with.

3.       Thanks ☺

If you have any questions about design decisions we took with Olingo let me 
know and we will sort it out.

Best regards,
Christian

From: Kevin Ratnasekera [mailto:djkevincr1...@gmail.com]
Sent: Montag, 13. Juli 2015 21:45
To: Amend, Christian; Christian Amend
Cc: dev@olingo.apache.org
Subject: Re: JSON Metadata Document client side deserializer

Hi Christian,
Please find my answers to your questions,

1.       Why is the interface “EdmJsonSchema” visible in the API? For a client 
it should not matter if he deserializes a Json or XML Metadata Document.

Yes this is true, but if you compare two metadata documents JSON and XML, json 
metadata document does not have dataservices tag and version compared to its 
xml document. I thought it was inappropriate to extend the Edmx interface for 
the with has getDataservises () getVersion() abstract methods since, those are 
no longer valid here. EdmJsonSchema and Edmx interfaces refers two counter 
parts of metadata document two types, and the implementations of these two 
interfaces are ClientJsonSchemaCsdl and ClientCsdlEdmx. If you look at the 
classes this is part of a deserialization of two metadata documents, I have too 
implemented the API XMLMetadata interface for the JsonMetadata document by 
ClientCsdlJSONMetadata, final out come of metadata is just same for the two 
cases.

2. You have used the existing Architecture where every type of Edm object 
results in a separate class which always calls the parser.getCodec().readTree() 
method. If you are using the JacksonParser library and parse the tree manually 
instead of using annotations why not follow the approach of writing a single 
parser class which reads the whole Json Metadata document in one go. Please 
have a look at the JsonDerserializer.entity() method in the server core 
package. This is what I would have thought about when you said you would like 
to write a custom deserializer. If you follow the approach on the server more 
closely you don´t have to replicate so many classes. Maybe this would make it 
easier on client side as well.

Sorry If I made you misunderstand, I think I have mentioned about writing 
custom deserializer per metadata document element, so there will several 
deserializer classes in this mail thread. Actually I wanted to follow the token 
based approach (not parsing entire tree once and build the model) similar to 
what current client-core currently has for the XML metadata document, This 
architecture follows writing multiple deserializer classes activated when parse 
through the tree, But as I did mentioned to you because of the dependency 
having between the definitions and schemas with the newly introduced 
definitions filed json metadata document, this field get deserialized first 
before the schema field. That is why I had to look for other approach which you 
mentioned. Yes, I agree with you completely writing a single class will do the 
thing. This approach basically each deserializer handles part of the document. 
I have looked several the deserializer implementations of Olingo library.This 
the same for the Json Service Document deserializer for client-core follows. 
But I wanted to follow the architecture what exist for the xml document has so 
my implementation will look similar to that. Other than that I see several 
advantages of this approach such as increase modularity of code, that will lead 
to better cohesive, understandable, reusable code.
Anyway with the issues I faced, I will have to rethink about the way things get 
deserialized inside jackson library, because as I see, there can be several way 
of doing the same thing. May be there are other workarounds too for the issues 
I faces. I will look more into this too as I am not very familiar with the 
jackson.

3.       The logic to split a type name is already implemented in the 
FullQualifiedName class. Could you ruse that logic in your code or do yoou have 
to react to specific parts of the type name?
Yes, completely agree with you, I will do the changes.
I am really thankful for you taking time to review my code :)
Regards
Kevin

On Mon, Jul 13, 2015 at 11:14 PM, Kevin Ratnasekera 
<djkevincr1...@gmail.com<mailto:djkevincr1...@gmail.com>> wrote:
Hi Christian,
Please find my answers to yours questions,

1.       Why is the interface “EdmJsonSchema” visible in the API? For a client 
it should not matter if he deserializes a Json or XML Metadata Document.

Yes this is true, but if you compare two metadata documents JSON and XML, json 
metadata document does not have dataservices tag and version compared to its 
xml document. I though it was inappropriate to extend the
Edmx interface for the with has getDataservises () getVersion() abstract 
methods since, those are no longer valid here.






On Mon, Jul 13, 2015 at 9:20 PM, Amend, Christian 
<christian.am...@sap.com<mailto:christian.am...@sap.com>> wrote:
Hi Kevin,

I had a look at your code and it seems good although I have some questions:


1.       Why is the interface “EdmJsonSchema” visible in the API? For a client 
it should not matter if he deserializes a Json or XML Metadata Document.

2.       You have used the existing Architecture where every type of Edm object 
results in a separate class which always calls the parser.getCodec().readTree() 
method. If you are using the JacksonParser library and parse the tree manually 
instead of using annotations why not follow the approach of writing a single 
parser class which reads the whole Json Metadata document in one go. Please 
have a look at the JsonDerserializer.entity() method in the server core 
package. This is what I would have thought about when you said you would like 
to write a custom deserializer. If you follow the approach on the server more 
closely you don´t have to replicate so many classes. Maybe this would make it 
easier on client side as well.

3.       The logic to split a type name is already implemented in the 
FullQualifiedName class. Could you ruse that logic in your code or do yoou have 
to react to specific parts of the type name?

Let me know what you think about these points ☺

Aside from that the code you wrote to put the json definition parts into their 
respective bean classes looks good to me. Using the CSDL… classes is the right 
approach to store the data. This way the EDM implementation can be reused.

Best Regards,
Christian

From: Kevin Ratnasekera 
[mailto:djkevincr1...@gmail.com<mailto:djkevincr1...@gmail.com>]
Sent: Sonntag, 12. Juli 2015 20:15
To: Amend, Christian; Christian Amend
Cc: dev@olingo.apache.org<mailto:dev@olingo.apache.org>

Subject: Re: JSON Metadata Document client side deserializer

Hi Christian/Olingo Community,
As I have mentioned in my previous email, I have come with a for first 
prototype for the client side Json metadata document deserialization using 
custom json deserializers. With my latest commit[1] i have pushed the changes 
for prototype. First prototype has the following deserializers , definitions - 
enum type (done-done) , references - includes and include annotations 
(done-done), definitions - entity type ( partially done ) , and
schemas - entity set ( partially done ). This changes commit is somewhat lager 
but I have added in the purpose of getting the big picture. They may look 
similar when compared to the current deserializer implementation in client-core 
but there are some changes you will notice. I will more document these with 
google doc for the project, cases where I had some issues how  I overcame those 
issues using Jackson library. I have added test cases for support the prototype 
too. Please have look at it and provide me some feedback so that I can proceed 
with implementation of the rest.
Regards
Kevin
[1]https://github.com/djkevincr/olingo-odata4/commit/ebecb573fc7917533afe8d3f90fa2e6522f671fd

On Fri, Jul 10, 2015 at 2:30 PM, Amend, Christian 
<christian.am...@sap.com<mailto:christian.am...@sap.com>> wrote:
Hi Kevin,

Thanks. I applied the changes to the Apache jsonMetadata branch and all tests 
are now working for me. I also packed you commits in diff files and attached 
them to the JIRA issue as a future reference. Keep up the good work ☺

Best Regards,
Christian

From: Kevin Ratnasekera 
[mailto:djkevincr1...@gmail.com<mailto:djkevincr1...@gmail.com>]
Sent: Donnerstag, 9. Juli 2015 21:20
To: dev@olingo.apache.org<mailto:dev@olingo.apache.org>; Amend, Christian
Subject: Re: JSON Metadata Document client side deserializer

Hi Christian,
With my latest commit, I have added a fix for test case failures due to setting 
default pretty printer to Jackson json generator. If you still have issues 
running test cases for metadata document json serializer,  Please let me know. 
I will come with prototype very soon too.
[1]https://github.com/djkevincr/olingo-odata4/commit/e20f35268638d44785c967e17ec647a65460102b
Regards
Kevin

On Tue, Jul 7, 2015 at 1:28 AM, Kevin Ratnasekera 
<djkevincr1...@gmail.com<mailto:djkevincr1...@gmail.com>> wrote:
Hi Christian,
I agree with your concerns, I will come with a prototype as soon as possible to 
get your feedback proceed forward.
Regards
Kevin

On Mon, Jul 6, 2015 at 2:02 PM, Amend, Christian 
<christian.am...@sap.com<mailto:christian.am...@sap.com>> wrote:
Hi Kevin,

your idea of using a custom deserializer instead of pure annotations sounds 
very good to me. This would make it easier to first deserialize the existing 
schemas and then move to the type definitions afterwards. This way you already 
have the schema CSDL classes ready and can attach the types as you deserialize 
them. I am looking forward to a first prototype ☺

About the general architecture: I would suggest that the deserializer should 
not validate the document it is parsing for semantics. This means that if a 
server sends a metadata document that contains an entity set with a type that 
is not present we do not throw an exception during parsing. This is to save 
performance for bigger metadata documents and also because a server should 
always send a valid metadata document.

WDYT?

Best Regards,
Christian

From: Kevin Ratnasekera 
[mailto:djkevincr1...@gmail.com<mailto:djkevincr1...@gmail.com>]
Sent: Samstag, 4. Juli 2015 21:49
To: dev@olingo.apache.org<mailto:dev@olingo.apache.org>
Cc: Amend, Christian; 
amend.christ...@gmail.com<mailto:amend.christ...@gmail.com>
Subject: JSON Metadata Document client side deserializer

Hi Olingo Community,
Regarding the Json metadata document GSoC project currently I have progressed 
upto the metadata document serialization at the server side. [1] Now I have 
moved implementation of client side metadata document de-serialization.
When I have look at the current Service Document Json deserializer and metadata 
document xml deserialzer in client core lib, the approach which is used to 
deserialize either Json and xml payloads of metadata/service document is 
writing custom Jackson deserializer classes. For the Json service document a 
single custom deserializer  ( JSONServiceDocumentDeserializer ) is being used, 
and for the xml metadata document case set of deserializer classes for parts of 
metadata document (EntityContainerDeserialize, ReferenceDeserializer) and they 
are registered on jackson mapper object using annotations first and then called 
recursively when we go through the each element of metadata document when 
parsing the relevant payload.
I am thinking of adapting the similar approach to Json metadata document 
deserialization, basically to write Jackson custom deserializer classes for 
json metadata document elements similar to xml case ( but there are some 
changes documents). I see some advantages of this basically this approach is 
complaint with the current approach, custom deserializer classes are more 
robust and adaptable complex custom cases when compared to approaches just 
simply use Jackson annotations to map json fields to java object fields.
Are there any other approaches available to achieve this? And What should be 
the best approach to use?
Regards
Kevin
[1]https://github.com/djkevincr/olingo-odata4/tree/jsonMetadata





Reply via email to