JSON libs compare and benchmark

Supported libs

• /UI2/CL_JSON
• XCO library
• ABAPify JSON ( ZCL_JSON )
• AJSON which is used in abapGit
• Standard identity transformation

Feature comparison

Feature	UI2	XCO	Abapify	Identity	aJSON
Camel case	✅	✅	✅	❌	❔
Custom case	❔	✅	✅	❌	❔
Custom transformation	❌	✅	❌	❌	❔
Suppress initial components	✅	❌	✅	✅	❔
Support data refs (render)	✅	❌	✅	⚠️	❔
Support polymorphic tables	✅	❌	✅	⚠️	❔
Root array	✅	✅	✅	❌	❔

Camel case

Camel case is usually the most common case in JSON world. Removing separators like - or _ helps to reduce the size of payload and devs use a special camelCase notation. Let's see how we can reach this with our libs:

• UI2 has pretty_name flag which enforces camel case automatically

result = /ui2/cl_json=>serialize( EXPORTING data = data pretty_name = abap_true ).

• XCO lib uses quite flexible mechanism of custom transformations including predefined such as xco_cp_json=>transformation->underscore_to_camel_case

result = xco_cp_json=>data->from_abap( data )->apply( it_transformations = VALUE #(
        ( xco_cp_json=>transformation->underscore_to_camel_case )
    )  )->to_string( ).

• ABAPify ZCL_JSON uses camel case by default

json = zcl_json=>render( data ).

• Identity transformation supports only asJSON format which means all names come as UPPER_SNAKE_CASE notation

Custom case

• UI2 ( to be checked )
• XCO as it's shown above apply function accepts callbacks which you can use for any rules
• Abapify JSON supports also case parameter which implements ZIF_ABAP_CASE interfaces introduced in case library

json = zcl_json=>render( data = data case = new lcl_your_own_case( ) ).
" or use predefined like
json = zcl_json=>render( data = data case = zcl_abap_case=>kebab( ) ).

Suppress initial components

initial_components - is a standard option for identity transformation.

You can use it like

 call transformation id
      source data = data
      options
        initial_components = 'suppress'
      result xml data(lv_xml).

Abapify JSON is using identity transformation internally and therefore propagates this option and supports all of available modes:

• include
• suppress_boxed
• suppress

In Abapify JSON you can provide it explicitly if you want

json = zcl_json=>render( data = data initial_components = 'include' ).

In UI2 this can be achieved by using compress = true and in XCO library i was not able to find such methods. Please contribute if you found.

Support data refs

Another interesting feature of identity transformation is data_refs By default they are ignored. You can include data refs into your JSON but they will be output in a very weird manner:

{"DATA":{"SAMPLE":{"%type":"cls:ZCL_JSON_BENCHMARK.TY_MAIN","%val":{"ID":1,"NAME":"Benchmark Data",

So for this example:

 types:
            BEGIN OF root_ts,
              name TYPE REF TO data,
            END OF root_ts.
        data(ref_to_sample) = VALUE root_ts( name = ref #( sample-name ) ).

UI2 and Abapify libs seem to handle refs correctly, but XCO just dumps..

Ref to data
/ui2/cl_json
{"name":"Benchmark Data"}  
zcl_json
{"name":"Benchmark Data"}  
identity
{"DATA":{"NAME":{"%type":"xsd:string","%val":"Benchmark Data"}}}  

Both libs also support table of ref to data which is also nice.

Performance evaluation

Rendering and parsing (binary format)

Setup:

• Same data object
• 100 iterations

As you can see identity is almost free, XCO is disappointingly slow. UI2 seems working 2-3 times faster than Abapify currently.

to be continued....