Inline declarations are a new way of declaring variables and field symbols at operand positions.

Data Declarations

In ABAP you have many operand positions, where the value of the operand is changed by the statement. The most typical of these “write positions” is the left hand side lhs of an assignment.

lhs = rhs.

But of course there are more. The data objects you can use at these write positions are either writable formal parameters of the procedure you are working in or variables declared with DATA in front of the statement.

In many cases the variables filled by a statement are helper variables that you only need close to the statement. For each of  these helper variables you had to write a data declaration with the DATA statement and of course it was your task to give the variable an adequate type.

Well, the operand type of most write positions is statically fixed and well known to the compiler. And this is why ABAP can offer inline data declarations with Release 7.40. The ingredients are so called declaration positions (write positions with fully known operand type)  and the new declaration operator DATA(…).

Let’s look at some examples.

Declaration of a lhs-variable for a simple assignment

Before 7.40

DATA text TYPE string.
text = `…`.

With 7.40

DATA(text) = `…`.

Declaration of table work areas

Before 7.40

DATA wa like LINE OF itab.
LOOP AT itab INTO wa.  
  ...
ENDLOOP.

With 7.40

LOOP AT itab INTO DATA(wa).  
  ...
ENDLOOP
.

Declaration of a helper variable

Before 7.40

DATA cnt TYPE i.
FIND … IN … MATCH COUNT cnt.

With 7.40

FIND … IN … MATCH COUNT DATA(cnt).

Declaration of a result

Before 7.40

DATA xml TYPE xstring.
CALL TRANSFORMATION … RESULT XML xml.

With 7.40

CALL TRANSFORMATION … RESULT XML DATA(xml).

Declaration of actual parameters

Before 7.40

DATA a1 TYPE …

DATA a2 TYPE …

oref->meth( IMPORTING p1 = a1

            IMPORTING p2 = a2

            … )

With 7.40

oref->meth( IMPORTING p1 = DATA(a1)

            IMPORTING p2 = DATA(a2)

            … ).

Declaration of reference variables for factory methods

Before 7.40

DATA ixml           TYPE REF TO if_ixml.
DATA stream_factory TYPE REF TO if_ixml_stream_factory.
DATA document       TYPE REF TO if_ixml_document.

ixml           = cl_ixml=>create( ).
stream_factory = ixml->create_stream_factory( ).
document       = ixml->create_document( ).

With 7.40

DATA(ixml)           = cl_ixml=>create( ).
DATA(stream_factory) = ixml->create_stream_factory( ).
DATA(document)       = ixml->create_document( ).

This example is my favorite. When working with class libraries as the iXML-Library you don’t have to care about the data type of the reference variables too much any more. You simply create them inline and use them. As you will see in the 7.40 version of the ABAP Example Library, this feature has facilitated my writings of example programs considerably.

 

Field Symbols

For field symbols there is the new declaration operator FIELD-SYMBOL(…) that you can use at exactly three declaration positions.

ASSIGN … TO FIELD-SYMBOL(<fs>).

LOOP AT itab ASSIGNING FIELD-SYMBOL(<line>).

ENDLOOP.

READ TABLE itab ASSIGNING FIELD-SYMBOL(<line>) …

I guess it is clear to you what happens here.

Outlook

In my upcoming blogs I will make use of inline declarations when introducing other new features. Be prepared for code like this:

TYPES t_itab TYPE TABLE OF i WITH EMPTY KEY.

DATA(itab) = VALUE t_itab( ( 1 ) ( 2 ) ( 3 ) ).

Yes, this is ABAP 7.40 …

To report this post you need to login first.

137 Comments

You must be Logged on to comment or reply to a post.

  1. Rüdiger Plantiko

    Great news, Horst! Looking forward to 7.40 being shipped.

    What I find even more annoying in present-day ABAP than the necessity of type declarations for lvalues, is the necessity of declaring auxiliary variables and to move data to them only for being able to pass these data to a method (or function module).

    One example: Suppose I have a field in a STRING variable iv_name. Just because some other developer didn’t use the general type ‘CSEQUENCE’  for his API import parameter, I have to add a data declaration for it and to move the STRING content to it before calling his API:

    data: lv_bname type BNAME.

    lv_bname = iv_name.  ” (… supposing iv_name is a STRING field )

    lo_api->method( exporting iv_bname = lv_bname ). ” Doesn’t accept the iv_name, bc. it’s a STRING.

    Will there be a simplification for this kind of call too?

    Actually, the local variable lv_bname is needed only for the call. The compiler knows itself what type is expected (at least for non-generic types, for which we have this kind of problem). So the compiler could generate a data object on the fly and perform an implicit MOVE. Syntax could be

    lo_api->method( exporting iv_bname ?= iv_name ).

    or

    lo_api->method( exporting iv_bname = (?) iv_name ),

    (since for a one-parameter or preferred-parameter method the formal parameter name may not be specified, the simplified syntax would then be possible:

    lo_api->method( (?) iv_name )

    or even implicit (no special annotation at all – but this variant would bring the backward compatibility issue):

    lo_api->method( exporting iv_bname = iv_name ).

    to indicate that a move should be performed to an ad-hoc data object of the appropriate data type.

    Regards,

    Rüdiger

    (0) 
    1. Horst Keller Post author

      Will there be a simplification for this kind of call too?

      Yes.

      lo_api->method( exporting iv_bname = CONV #( lv_bname ) ).

      CONV is one of the new constructor operators, that I will introduce in an upcoming blog.

      Best

      Horst

      (0) 
    2. Volodymyr Shcheglov

      Rüdiger Plantiko wrote:

                             

      Great news, Horst! Looking forward to 7.40 being shipped.

      …<skipped>

      One example: Suppose I have a field in a STRING variable iv_name. Just because some other developer didn’t use the general type ‘CSEQUENCE’  for his API import parameter, I have to add a data declaration for it and to move the STRING content to it before calling his API:

      data: lv_bname type BNAME.

      lv_bname = iv_name.  ” (… supposing iv_name is a STRING field )

      lo_api->method( exporting iv_bname = lv_bname ). ” Doesn’t accept the iv_name, bc. it’s a STRING.

      Will there be a simplification for this kind of call too?

      …<skipped>

      Regards,

      Rüdiger

                         

      This example is slightly irrelevant because you can use string templates. Since ABAP version 7.02 one is able to write the code such as:

      data: lv_str type string value ‘initial’.

      lo_api->method( iv_bname = |{ lv_str }| ).

      This is applicable only to methods. Unfortunately, you can’t do the same for function or procedures calls.

      By the way, question to Horst: why isn’t that feature implemented for older constructions such as CALL FUNCTION or PERFORM, or macro calls? Was it too difficult, or it was just assumed there was no demand for such things?

      As for Rüdiger‘s example, the real problem is with other convertible types, say, P or N. For example, we have a method’s parameter is defined as IV_PAR(8) TYPE P DECIMALS 4, and an actual parameter, as LV_VAL(6) TYPE P DECIMALS 2. It’s obvious that this variable can be easily converted to the formal parameter’s type but right now you will get a syntax error trying to call this method as:

      LO_OBJ->METHOD( IV_PAR = LV_VAL ).

      So,

      CONV #( lv_bname ) would be handy.

      (0) 
  2. Custodio de Oliveira

    Hi Horst,

    In your very first example you have:

    Before 7.40

    DATA text TYPE string.
    text = `…`.

    With 7.40

    DATA(text) = `…`.

    But currently we can have in one line already:

    DATA text TYPE string VALUE ‘…’.

    Still a nice improvement.

    Cheers,

    Custodio

    (0) 
    1. Horst Keller Post author

      Hi Custodio,

      so let me rewrite my first example:

      data(var) = rhs.

      Where rhs is

      • any data object
      • a functional method call
      • a call of a built-in function
      • an arithmetic expression
      • a string expression
      • a bit expression
      • a constuctor expression
      • a table expression

      Now it’s your turn again 😉

      (0) 
      1. Volodymyr Shcheglov

        Horst Keller wrote:

                               

        Hi Custodio,

        so let me rewrite my first example:

        data(var) = rhs.

        Where rhs is

        • any data object
        • a functional method call
        • a call of a built-in function
        • an arithmetic expression
        • a string expression
        • a bit expression
        • a constuctor expression
        • a table expression

        Now it’s your turn again 😉

                           

        Dear Horst, will there be an ability to make a call like this:

        CALL FUNCTION ‘MY_FUNC’

          EXPORTING

            IV_PAR = lo_object->method( )

        ? This is really a demanding feature. Otherwise we need to declare a variable just to store a returning parameter and then pass it to a function call. Moreover, this feature is fully supported in method calls.

        (0) 
  3. Suhas Saha

    Release 740 is perhaps the most revolutionary change i’ve seen in 7.5yrs as an ABAPer!

    I think the ABAP code will look more like Java in days to come. Any insider tip when will 740 be shipped to custom developers? 😏

    BR,

    Suhas

    (0) 
    1. Horst Keller Post author

      I think the ABAP code will look more like Java in days to come.

      Yep, and I’m sure not erverybody will like that. For me, the only spoiler is that you can mix very old and quaint stuff with very modern features inside one program unit now. You can create quite queer effects by this. I’m a great fan of a version flag for ABAP programs that would allow new stuff in new programs only, where obsolete stuff would be prohibited then. But alas, no majority for this.

      Any insider tip when will 740 be shipped to custom developers?

      Uh, oh, I’m not insider enough to answer such a question, better ask one of the PM guys blogging about 7.40 …

      Best

      Horst

      (0) 
      1. Brian O'Neill

        Horst Keller wrote:

                               

        I’m a great fan of a version flag for ABAP programs that would allow new stuff in new programs only, where obsolete stuff would be prohibited then. But alas, no majority for this.

                           

        Wow! That’s too bad!

        I could see where the version flag could just be set lower for people that are still stuck in the past. The problem is that some people writing outdated code may not know that they are doing anything wrong or even obsolete.

        I think there is a hidden cost to not adopting modern ABAP that is not really promoted. There are good developers graduating right now who may have the potential to be great ABAP developers. You can do a lot without utilizing modern ABAP, but outdated techniques will not inspire a typical millennial developer.

        It’s hard enough to find experienced ABAP developers as is. By not utilizing modern ABAP, it’s going to be hard to inspire the next generation of ABAP developers.

        (0) 
        1. Uwe Fetzer

          If you setup your ATC (ABAP Test Cockpit) the right way, you can prevent depreciated ABAP code from beeing transported.

          (just one way to force your devs to use modern code)

          (0) 
  4. Oliver Russinger

    Very cool innovations.

    specially

    lo_api->method( exporting iv_bname = CONV #( lv_bname ) ).

    would be something long awaited ;-).

    is there a complete reference in the snd anywhere as it will time time ’till we will geht 740 i guess.

    best regards oliver

    (0) 
    1. Horst Keller Post author

      is there a complete reference in the snd anywhere

      I’m planning to introduce all innovations step by step in blogs (always when I find some time to write one) and to produce one TOC in the end. Furthermore, I will try to get the offline version of the 7.40 keyword documentation into the sap help portal this year …

      (0) 
  5. B. Meijs

    First the changes in WAS702, now this. Surely an improvement to the language. Can’t wait to get my hands on a was740 server (and some customers to upgrade to a was740 based system)

    One question: Is there an backward compatibility check option available to check if my was702/703/740 abap code will run in older versions like was700?

    (0) 
    1. Horst Keller Post author

      Is there an backward compatibility check option available to check if my was702/703/740 abap code will run in older versions like was700

      Don’t know of any but trying to compile the code in 7.00.

      (0) 
      1. Nigel James

        You raise a good point Ben Meijs . If I was a partner delivering custom code and I was targeting several systems, would I have to write to the LCD or could I do checking to see if my new code could run, in a similar way to javascript testing for browser features.

        Nigel

        (0) 
      2. B. Meijs

        Horst Keller wrote:

                               

        Is there an backward compatibility check option available to check if my was702/703/740 abap code will run in older versions like was700

        Don’t nkow of any but trying to compile the code in 7.00.

                           

        Well, actually we run the code inspector on older releases, because we also ran into problems with functions missing specific formal parameters.

        (0) 
  6. Paul Hardy

    All jokes and songs aside, this is wonderful stuff. I only had access to a 7.02 system this time last year, and it will be five years at least before I can start programming in version 7.40, but already I can’t wait….

    Going back to header lines, I reckon one of the reasons a lot of people kept using them, was because people includign me are lazy, and you had to explictly define a work area and it was far easier just to say LOOP AT ITAB. In 7.40 there is no excuse whatsoever you neatly solve the problem of not having two data objecst with the same name, coupled with not having to explicitly define the work area.

    Declaring the variables whilst receiving it is brilliant as well.

    Well done. As I said, I can’t wait.

    Cue Ray Davies: – “Yeah, you really got be going, you got me so I can;t sleep at night, you really got me, you really got me…”

    (0) 
    1. Rüdiger Plantiko

      Applause to the solution to declare the field symbol at the point where it is actually needed – and not some miles above (for those of us who like large methods)…

      The most robust way would be to have the scope limited to the loop itself, throwing the dreaded GETWA_NOT_ASSIGNED as soon as one tries to access the field-symbol outside of the loop.

      I don’t know, however, whether limited scope is planned in ABAP.

      We would then be very similar to the range-based for in C++11 (http://www.stroustrup.com/C++11FAQ.html#for ). In

      for (auto entry : addressBook ) {

         …

          }

      not only the type of the variable “entry” will be automatically (type “auto”) be inferred from the addressBook, but also the scope of the symbol “entry” is limited to the for {} loop (which is nothing new). This way it is made sure that “entry” cannot be mis-used later on.

      (0) 
      1. Horst Keller Post author

        Rubbing salt into the wound eh?

        In fact:

        LOOP AT  ASSIGNING FIELD-SYMBOL(<fs>).

        ENDLOOP.

        LOOP AT  ASSIGNING FIELD-SYMBOL(<fs>).

        ENDLOOP.

        gives a syntax error 😥

        (0) 
        1. Suhas Saha

          LOOP AT itab1  ASSIGNING FIELD-SYMBOL(<fs1>).

          ENDLOOP.

          LOOP AT itab2  ASSIGNING FIELD-SYMBOL(<fs2>).

          ENDLOOP.

          So this is the correct syntax then? I think the visibility remains the same as in previous releases.

          (0) 
  7. Uwe Fetzer

    First of all: you know I like ABAP 7.40 !!!!!

    But in this case the declaration of SFLIGHT_LH should result in a table (with all ‘LH’ entries) and not a flat structure (with just the first entry):

    DATA sflight_all  TYPE SORTED TABLE OF sflight WITH UNIQUE KEY carrid connid fldate.

    SELECT * FROM sflight

      INTO TABLE sflight_all

      ORDER BY PRIMARY KEY.

    DATA(sflight_lh) = sflight_all[ carrid = ‘LH’ ].

    And it would be great if this inline declaration would work also in the next release:

    SELECT * FROM sflight

      INTO TABLE DATA(sflight_all)

      ORDER BY PRIMARY KEY.

    Maybe ABAP 7.50/8.00?

    (0) 
    1. Oliver Russinger

      i guess this is  DATA(sflight_lh) = sflight_all[ carrid = ‘LH’ ]. a read table not a loop…endloop construct.

      but they (sap) have a lot to do to meet users expectations.

      remember r/3 3.1 , there was no alv at all, only write statements and you have to do your sorting by yourself 😉

      (0) 
      1. Horst Keller Post author

        Ten points for the right answer.

        All table expressions that we have up to now are shortcuts for READ and return a table line. I’ll ask the creator of those, if he will think about LOOP-expressions.

        SELECT is already under consideration.

        (0) 
          1. Thomas Porcham

            This is good news. SELECT on itabs could replace many loops. I think it is more readable too. I would also like a MAP functionality on itabs similar to the map function in functional languages like Haskell or Scala.

            Thanks for inline declarations.
            It will make my ABAP code less verbose. I think it will remove “noise” from my programs helping me to concentrate on the business logic.

            (0) 
            1. Horst Keller Post author

              SELECT on itabs could replace many loops

              No, it was about SELECT FROM dbtab INTO TABLE DATA(itab).

              But from 7.40 SP05, we will have Table Comprehensions!

              E.g.:

              DATA(itab) = VALUE itab(

                    FOR x IN array INDEX INTO idx

                       LET off = factor * idx IN

                      ( col1 = x col2 = x * x col3 = x + off ) ).

              I would also like a MAP functionality on itabs

              Again 7.40, SP05:

              itab2 = CORRESPONDING #( itab1 MAPPING … = …  … = … ).

              (0) 
              1. Paul Hardy

                Hooray!

                In 7.02 in about 2011 you let ABAP have expressions ike A = ( B + C)  a construct BASIC could do in 1981, so if someone had jumped forward 30 years in a “Back to the Future” car they would have been comfortable with the language.

                Now in 2013 we have the good old FOR NEXT LET  LOOPS which again look just like the ZX81 syntax when I started to program when I was 14. I’m 45 now.

                I’m not complaining – this is a good thing, I loved programming the ZX81 even if it only had 1K of memory. It was better than the VIC20 even if that had much more memory the programming language onthe VIC20 had lots of holes in it, and workarounds e.g. PEEK / POKE.

                Programming languages must be like the fashion industry. Don’t throw anything out of your wardrobe, in thirty years it will be fashionable again.

                Cheersy Cheers

                Paul

                (0) 
                1. Rüdiger Plantiko

                  [START DIGRESSION]

                  >It was better than the VIC20 even if that had much more memory

                  Disagree. The VIC20 had the better CPU: MOS Technology’s 6502 was by far the best CPU available in the market in those days – better then the 8080, and much better than the Z80 anyway. It’s a pity that the x8… family won the race, although its concepts are worse. BTW – see here an impressive page of a 6502 fan, emulating the currents inside the processor during execution! http://www.visual6502.org/JSSim/

                  [END DIGRESSION]

                  (0) 
  8. Ronaldo Aparecido

    Congratulations Mrs Keller.

    I did try install the 7.4 version ,but the message that was not released in Brazil.

    It´s important for studies.

    what date to come in Brazil?

    THanks ,useful Post.

    (0) 
  9. solen dogan

    THats great Horst

    Abap gets more eloguent as each release is shipped

    I am so excited to try the new release

    look forward to try the new release

    (0) 
  10. Kamal el Marzgioui

    I was already very charmed by the new features of the new ABAP Workbench and Debugger (in 7.00 and 7.02)!

    This…this is just great stuff!! Can’t wait 🙂

    @Horst, thank you for the post. Please keep sharing!

    Regards,
    Kamal.

    (0) 
  11. Adi Sieker

    Curently doing my first project on a 7.4 AS ABAP and inline declarations are just so nice. Been waiting for that for ages.

    Also method chaining and being able to directly access fields of a structure when calling functional methods are just awesome.

    Thank you ever so much, for the work in tht respect. Makes working with ABAP just so much more fun.

    (0) 
  12. Christian Dr. Drumm

    Hi Horst,

    is there any way to use inline declaration together with field-symbols where the field-symbol is itself an itab?

    What I’m trying to do is the following. I’ve got a structure example_order which has a component example_order_items. Now I want to access this data using only field symbols and inline declarations. What I tried is the following:

    ASSIGN COMPONENT ‘ITEMS’ OF STRUCTURE <example_order> TO FIELD-SYMBOL(<items>).

    LOOP AT <items> ASSIGNING FIELD-SYMBOL(<item>).

    This doesn’t seem to work. At least I could’t figure out how. If I use an explicitly declared field symbol instead everything works fine:

    FIELD-SYMBOLS <items> TYPE ANY TABLE.

    ASSIGN COMPONENT ‘ITEMS’ OF STRUCTURE <example_order> TO <items>

    LOOP AT <items> ASSIGNING FIELD-SYMBOL(<item>).

    However, I’d like to use the inline declarations even in this case. Is this possible?

    Best,

    Christian

    (1) 
    1. Horst Keller Post author

      Hi Christian,

      Since you are using the dynamic ASSIGN COMPONENT, the field symbol declared inline has the generic type ANY (how should the compiler know the type of the assigned component). A field symbol (or formal parameter) that is not typed as an internal table cannot be used at operand positions for internal tables.

      Your example would work with a static ASSIGN:

      ASSIGN COMPONENT <example_order>-items TO FIELD-SYMBOL(<items>).

      LOOP AT <items> ASSIGNING FIELD-SYMBOL(<item>).

      If this is not possible, you have to declare and type the field symbol explicitly with TYPE ANY TABLE as you’ve shown it. An inline declaration is not possible then.

      Best,

      Horst

      (1) 
  13. Lokesh Balajee

    Hello Horst,

    I recently started exploring 7.4 features and its fascinating.

    This works well

    PARAMETERS: p_from   TYPE /sapapo/locno.

    DATA(g_from) = p_from

    DATA(g_from1) = g_from.

    But, when I try this, I get an error.

    PARAMETERS: p_from   TYPE /sapapo/locno.

    DATA(g_from1) = DATA(g_from) = p_from.

    Also one more question,

    Though Inline declarations can be used always, what is its effect on the global variables in Data declaration Includes?. The reason for this question is all the clients have coding standards and most of them will go for 1 Data declaration include and 1 Subroutine includes.

    Lokesh

    (0) 
    1. Horst Keller Post author

      Hi Lokesch,

      But, when I try this, I get an error.

      PARAMETERS: p_from   TYPE /sapapo/locno.

      DATA(g_from1) = DATA(g_from) = p_from.

      Yes, inline declarations are forbidden in multiple assignments.

      what is its effect on the global variables in Data declaration Includes

      Global variables shouldn’t be declared at all … 😛

      But seriously:

      Inline declarations influenced the way we looked at those things and we tackled that as follows

      Programming Guideline for Inline Declarations

      Horst

      (0) 
      1. Adi Sieker

        Horst Keller wrote:

        Inline declarations influenced the way we looked at those things and we tackled that as follows

        Programming Guideline for Inline Declarations

        ideally the scope of inline declarations would be the block they were defined for. 🙂

        That was the first question most colleagues asked me when I showed them inline declarations.

        i.e.the following would cause a syntax error on the last line, stating that ls_struc is unknown:

        LOOP AT lt_table into DATA(ls_struc).

             “Do something with ls_struc.

        ENDLOOP.

        add 1 to ls_struc-counter.

        (0) 
          1. Adi Sieker

            Horst Keller wrote:

            I know, I know, I know, …  sigh

            sounds like that discussion as been had before. 🙂

            But since it’s not in there now and because people will start relying on it being the way it is now, it will never happen. 🙁

            (0) 
            1. Horst Keller Post author

              The backward compatibility trap. Building new features in old ABAP. There is no local context in old ABAP and it seems to be impossible to enable it without breaking existing code.

              (0) 
              1. SAP ITR ENTWICKLUNG

                Maybe with a new syntax element for scoping blocks like {… scoping block } as we know it from Java, C++…

                Additionally there could be a scoping version of other block building statements like

                If x = Y {  … }

                loop at lt_a into ls_a {  … }

                case y { when a. … }

                replacing the end[something] with the block closing sign. This would at the same time make the code leaner and more readable.

                (0) 
                1. Paul Hardy

                  I agree there is a big bonus with making code shorter (leaner)

                  In Java and so forth everyone is sed to the good old { }.

                  However for good or for bad ABAP programmers are used to things like ENDIF, and I wonder if } is more readable e.g.

                  is

                        ENDIF.

                      ENDCASE.

                    ENDMETHOD.

                  better than

                     }

                    }

                  }

                  When it comes to the readability stakes?

                  I presume in Eclipse when you hover over a “}” sign it tells you in a hover text the start of the code block the sign is closing? In ABAP the official recommendation used to be to add a comment such as

                  ENDIF.”Is the fish blue?

                  Cheersy Cheers

                  Paul

                  (0) 
                  1. Adi Sieker

                    Paul Hardy wrote:

                    In ABAP the official recommendation used to be to add a comment such as

                    ENDIF.”Is the fish blue?

                    I think, if that is necessary to make the scope (method/function/form) readable then it’s too complicated/long and should be split up.

                    (0) 
                    1. Adi Sieker

                                                  Horst Keller wrote:

                      In ABAP the official recommendation used to be to add a comment such as

                      ENDIF.”Is the fish blue?

                      Used to be, but is it still?

                      See Comments.

                      Quote from “Arrangement in source Code” page:

                      “…

                      End of line comments are suitable for the following situations:

                      • To indicate block ends (separate from indentations) in larger control structures

                      …”

                      and the “Good Example” is formatted as follows:

                        ENDIF. “item_ref IS BOUND AND…
                      ENDLOOP.

                      ENDMETHOD. “main
                      *———————————————————-*
                      ENDCLASS. “application


                      it does imply to use the endcomments.


                      (0) 
  14. Sayantika Bhattacharya

    Hi,

    My system is on Release 7.40 SP006. Inline declaration is giving syntax error.

    My Code:

    REPORT zr_opensql_01_aggregation.

    SELECT bp_id,
            company_name,
            so~currency_code,
            SUM( so~gross_amount ) AS total_gross_amount
       FROM snwd_so AS so
       INNER JOIN snwd_bpa AS bpa
               ON so~buyer_guid = bpa~node_key
       INTO TABLE @DATA(lt_result)
       GROUP BY bp_id, company_name, so~currency_code.

    cl_demo_output=>display_data( value = lt_result ).


    Syntax Error: The inline declaration “DATA(lt_result)” is not possible in this position.


    Any inputs on why this is happening will help a lot.


    Regards,

    Sayantika.

    (0) 
  15. Linkin Pereira

    Thanks for this Blog post Horst. I haven’t started playing around with 7.4 yet but sure looks interesting.

    One thing scares me though is this.

    Rule

    Only use inline declarations locally

    Only make inline declarations in processing blocks that support local data. Use them as if they were local declarations in the current statement block.

    Details

    If used correctly, inline declarations are an excellent way of making programs leaner and easier to understand.

    ——

    The IF used correctly part in the Details.

    ——–

    Don’t you feel the code would be much difficult to maintain if the program is a 1000 lines code and some where way up in the top there is an inline declared variable ( defined globally — by mistake I may add) and you have to debug this chunk of code to figure out where exactly is this variable defined.

    With the usual declaration statement you know that the declaration is made with the TYPE statement.

    — Well one argument would be with inline code you would not have 1000 lines of code anyway.  😛

    Comments.

    (0) 
      1. Adi Sieker

        Horst Keller wrote:

        ‘xactly, see Procedure Volume

        😈

        .

        In my opinion this is the number one thing that should be hammered into every developers head. I also think that 150 executable statements is to much. I think it should be around 50-70.

        (0) 
        1. Horst Keller Post author

          I also think that 150 executable statements is to much. I think it should be around 50-70.

          Yes, that’s the factor you can gain by using expression enabled ABAP 😉 .

          (0) 
        2. Rüdiger Plantiko

          My rule of thumb is: The implementation part of a method should be visible completely on my screen, with the signature part expanded => max. at about 30-40 lines of code per method, including blank lines and comments.

          (0) 
  16. Amol Vakhare

    Hi Horst,

    Wondering how can I declare a packed variable or any other type variable using the Data() operator.

    Data(lv_var1) = 1.  “Integer”
    Data(lv_var2) = ‘1.10’. “Declared as character…

    How to declare lv_var2 as packed type?

    (0) 
    1. Adi Sieker

      Hi Amol,

      DATA(lv_var2) = VALUE packed_type(‘1.10’).

      Where packed_type is any type in the ABAP dictionary and whatever else VALUE allows.

      Cheers

         Adi

      (0) 
      1. Amol Vakhare

        Hi Adi,Though I understood the use of VALUE.. still wondering why following gives compile time error>

        Data(lv_var) = value p(‘123.10’).

        Error: Value of generic type “P” cannot be constructed..

        The documentation says for generic type, additional attributes need to be specified.. so stuck again? Probable alternate is to define a type with a packed component in it and use it to declare.. Thoughts?

        (0) 
          1. Horst Keller Post author

            The problem is, it wouldn’t work too. Syntax error in:

            TYPES pack TYPE p DECIMALS 3.

            DATA(val) = VALUE pack( ‘1.1’ ).

            Why? Operator VALUE is not possible for elementary types except for the initial value.

            Therefore, either initial value:

            TYPES pack TYPE p DECIMALS 3.

            DATA(val) = VALUE pack( ).

            or the conversion operator:

            TYPES pack TYPE p DECIMALS 3.

            DATA(val) = CONV pack( ‘1.1’ ).

            But in fact the question remains, why you want to do this at all?

            If you want to declare a packed number, simply do so by using statement DATA or TYPES. The DATA operator for inline delcarations is designed for being used at declaration positions where the operand type is completely known and not for declaring variables of any types before using them. So if you have e.g. a method retrurning a packed number, of course you write

            DATA(val) = get_packed( … ).

            This is a use case. Constructs as above are possible but more as side effects …

            (0) 
            1. Amol Vakhare

              I see your point now.. With it I am more clearer on its usage. Thank you so much!

              The DATA operator for inline delcarations is designed for being used at declaration positions where the operand type is completely known and not for declaring variables of any types before using them. So if you have e.g. a method retrurning a packed number, of course you write

              DATA(val) = get_packed( … ).

              This is a use case. Constructs as above are possible but more as side effects …

              (0) 
  17. Michael Fritz

    Hi,

    how would one re-write this existing coding using inline declarations?

    DATA: o_reference  TYPE REF TO zcl_customer_service_base.

    IF NOT notif_no IS INITIAL.

    “Create new reference

         CREATE OBJECT o_reference   

              EXPORTING     

                   i_order_number        = ”     

                   i_notif_number         = l_notif_no     

                   i_use_notification     = abap_true     

                   i_no_notif_buffer      = abap_false     

                   i_no_partner_init      = abap_true     

                   i_remove_deleted_items = abap_true.

    ELSE. 

    “Use existing reference 

         o_reference = me.

    ENDIF.

    The following will not work because of error “o_reference is already declared” when using the second inline declaration in the ELSE branch.

    IF NOT l_notif_no IS INITIAL   .

         DATA(o_reference) = NEW zcl_customer_service_base(                           i_order_number        = ”    

              i_notif_number         = l_notif_no    

              i_use_notification     = abap_true    

              i_no_notif_buffer      = abap_false    

              i_no_partner_init      = abap_true    

              i_remove_deleted_items = abap_true    

              ).

    ELSE.

         DATA(o_reference) = me.  “<<– ERROR here

    ENDIF.

    Michael

    (0) 
    1. Horst Keller Post author

      Well that’s simple

      IF NOT l_notif_no IS INITIAL   .

           DATA(o_reference) = NEW zcl_customer_service_base(                           i_order_number        = ”   

                i_notif_number         = l_notif_no   

                i_use_notification     = abap_true   

                i_no_notif_buffer      = abap_false   

                i_no_partner_init      = abap_true   

                i_remove_deleted_items = abap_true   

                ).

      ELSE.

           o_reference = me.  “<<– ERROR here

      ENDIF.

      Note that

      DATA(o_reference) = NEW zcl_customer_service_base(  …

      can be seen as a short cut for:

      DATA o_reference TYPE REF TO zcl_customer_service_base.

      CREATE OBJECT o_reference EXPORTING …

      Therefore, the data declaration is valid for the current context.

      (0) 
      1. Michael Fritz

        Horst,

        thanks again for your reply.

        That means it does not matter if the IF condition is met or not and the program execution immediately jumps into the ELSE branch?

        That’s cool indeed!

        I’ve just checked this and it works perfectly! Never thought of this! At the very end no short dumps or similar when accessing the reference from the ELSE branch.

        Thanks,
        Michael

        (0) 
        1. Horst Keller Post author

          That means it does not matter if the IF condition is met or not and the program
          execution immediately jumps into the ELSE branch?

          Yes, but that’s also kind of dangerous. Because there is no local context for data in control structures, the declared objects are statically visible everywhere behind the declaration and dynamically (ASSIGN (name ) …) even everywhere in front of the declaration. Therefore, it needs some discipline when working with that, see the guideline.

          (0) 
  18. Michael Fritz

    You see I’m still learning those new inline declarations since we recently upgraded our ERP to 7.40 (SP08) 😉

    I wonder how to create an internal table without using any default values.

    Previously I would have written a statement like this:

    data: lt_w3head type STANDARD TABLE OF w3head.

    Now I tried to get the same result using inline declaration, to no avail so far.

    I need this internal table as a result table for a function module, hence I don’t have any rows available so far. The function module will return them instead.

    Would it be possible to use those inline declarations for this purpose, too?

    Michael

    (0) 
    1. Horst Keller Post author

      TYPES: tt_w3head type STANDARD TABLE OF w3head WITH … KEY …

      DATA(data: lt_w3head) = VALUE #( ).

      but see above,

      … the question remains, why you want to do this at all? … The DATA operator for inline delcarations is designed for being used at declaration positions where the operand type is completely known and not for declaring variables of any types before using them.

      See Declaration positions

      (0) 
    1. Horst Keller Post author

      Same, because the same kernel modules are called. It’s mainly the syntax that is different.

      But be aware of performance traps if you use expressions repeatedly in statements or even in loops. The new syntax can be seductive to write things you never wrote with fully blown statements. That can easlily lead to performing the same operations multiply.

      (0) 
  19. Dominik Krämer

    Hi Horst,

    great blog! However I have an question on the following part:

    TYPES: t_scarr   TYPE SORTED TABLE OF scarr
                 WITH UNIQUE KEY carrid.

    TYPES:
       BEGIN OF t_flights,
         scarr TYPE t_scarr,
       END OF t_flights.
    DATA: flights TYPE t_flights.

    SELECT * FROM scarr INTO TABLE flightsscarr.

    ASSIGN COMPONENT ‘SCARR’ OF STRUCTURE flights TO FIELDSYMBOL(<fs_dt_data>).


    Basically the above code works like a charm and in the end the fs_dt_data contains a sorted table with the values from scarr. However when I then want to do a loop over this table I cannot get the code to be activated:

    LOOP AT <fs_dt_data> ASSIGNING FIELDSYMBOL(<fs_data>).
       ...
    ENDLOOP.


    I always receive the following error: “.” expected after “<FS_DT_DATA>”


    Is this a bug or am I doing something wrong?


    Regards,

    Dominik

    (0) 
    1. Horst Keller Post author

      The error message is nonsense (I forward it) but you are trying to loop over something that can’t be statically recognized as table. For that you still need a explicit declaration:

      FIELD-SYMBOLS <fs_dt_data> type STANDARD TABLE.
      ASSIGN COMPONENT ‘SCARR’ OF STRUCTURE flights TO <fs_dt_data>.

      LOOP AT <fs_dt_data> ASSIGNING FIELDSYMBOL(<fs_data>).
         ...
      ENDLOOP.

      (0) 
      1. Dominik Krämer

        Yes that’s my workaround for the moment. Is this something which is planned for a later release, or are there any technical restrictions which prevent this?

        Regards,

        Dominik

        (0) 
        1. Horst Keller Post author

          If no type can be deduced, the field symbol declared inline is typed with ANY. You never could/can use fully generic field field symbols at operand positions of internal tables. That’s the rule.

          (0) 
          1. Dominik Krämer

            Hi Horst,

            not sure if I fully understood your point.

            Untitled.png

            In the debugger the type is correctly deduced as Sorted Table, only in the ABAP editor this is not recognised / throws the error message: “.” expected after “<FS_DT_DATA>”


            Regards,

            Dominik

            (0) 
            1. Horst Keller Post author

              Debugger is runtime. A DESCRIBE FIELD or RTTI would also find the internal table.

              But syntax check happens at compile time and the internal table cannot be deduced statically.

              https://ldcialx.wdf.sap.corp:44318/sap/public/bc/abap/docu?sap-language=EN&sap-client=000&format=STANDARD&object=ABENTYP&hellip;

              “The formal parameter or field symbol can be used as operands anywhere that is
              not excluded by this typing. However, operands that expect particular internal
              tables are an exception to this rule. Here, only formal parameters or field
              symbols typed as internal tables with the appropriate table category are allowed.”

              (0) 
              1. Dominik Krämer

                Thanks for the explanation.

                However it’s a shame that the runtime would give me the expected result and then have the limitation in the compiler, as I would had a use case for this which would have saved me a ton of additional declaration.

                Regards,

                Dominik

                (0) 
      2. Horst Keller Post author

        The nonsense is independent from expressions

        FIELD-SYMBOLS <fs_dt_data> TYPE any.

        DATA wa.
        LOOP AT <fs_dt_data> INTO wa.
         

        ENDLOOP.

        (0) 
  20. Sougata Chatterjee

    Hi Horst,

    Why the following is not supported? I’m in 7.4 SP05 on HDB – is it supported in later SPs?

    CALL FUNCTION ‘function_module’

         EXPORTING

              i_param     = DATA(lv_data)


    Regards,

    Sougata.

    (0) 
    1. Horst Keller Post author

      Since that operand position is a read and not a write position. Declaration positions are always write positions (L-values, not R-values).

      (0) 
    2. Rüdiger Plantiko

      Hi Sougata,

      you want to pass a data object to the called function module, where it will not be changed (since you are EXPORTING it).

      The only case where this makes sense in my eyes would be that the inventor of the function module declared an import parameter to be obligatory although not strictly needed in the code (i.e. although being optional in fact), and you want to pass an initial value to it, just to satisfy the interface.

      You can’t do that with a DATA operator, but you could do it with VALUE:

      call function ‘Z_IMPORT_PAR_OBLIGATORY’

         exporting

           is_vbap       = value vbap( ).


      Here, IS_VBAP is an obligatory import parameter of Z_IMPORT_PAR_OBLIGATORY of type VBAP. You can’t use # here, because for function modules – as opposed to methods – the compiler doesn’t re-read the parameter types.


      Regards,

      Rüdiger

      (0) 
  21. Nishant Bansal

    Dear Horst Keller,

    First of all thanks for sharing this blog.

    I have one question regarding this. Below i am taking one simple example.

    DATA wa like LINE OF itab.
    LOOP AT itab INTO wa.  
      ...
    ENDLOOP.

    With 7.40

    LOOP AT itab INTO DATA(wa).  
      ...
    ENDLOOP
    .



    After defining the WA using DATA(WA).

    Can i define the same name WA variable again in the current session?

    Actually i don’t have system to test it.



    Thanks and Regards,

    Nishant

    (0) 
    1. Michael Fritz

      Hi,

      no you can’t use DATA(wa) again.

      However after writing this inline declaration above, you can now use

      LOOP AT itab INTO wa.

      ENDLOOP.

      anywhere in your coding. No need to declare wa again.

      Michael

      (0) 
  22. Raphael Pacheco

    Horst,

    I have a basic question of this inline declarations.

    How could I create a counter using the inline declarations? It would be possible to make, for example ADD 1 to DATA(i)?


    Warn regards,


    Raphael Pacheco.

    (0) 
    1. Horst Keller Post author

      Would it be possible to write, for example, ADD 1 to DATA(i)?

      No, for two reasons:

      • The statement ADD is not expression enabled.
      • DATA(num) = num + 1 doesn’t work either because you cannot declare a variable inline at LHS that is used at RHS.

      Best

      Horst

      (0) 
  23. Sudhi Karkada

    Thanks for the article and for introducing us to new ABAP.

    I am sorry if this question was already asked and answered in the 5 pages of Q&A, but how does the double-click navigation work with this type of inline declarations? If the variable is a structure, would I be able to eventually navigate (just by double-clicking a few times) to DDIC and see the structure components?

    When I first started programming, many languages were sneered at because they did not require explicit data declarations. Agreed, inline declaration is different, but still the trend is unsettling.

    (0) 
    1. Horst Keller Post author

      Works as before, eg.,

      DATA struct1 TYPE scarr.

      DATA(struct2) = VALUE scarr( ).

      ... struct1 ...

      ... struct2 ...

      Double clicking struct1, struct2 leads to the respective declaration, where you can double click the type.

      (0) 
  24. Dionisio Ambrona

    Dear Horst Keller,

       in a first place, thanks a lot for your blog. I’m just starting to learn about new ABAP 7.40 capabilities, so I find it quite usefull and interesting.

       Regarding your first example in this post,

    Before 7.40

    DATA text TYPE string.
    text = `…`.

    With 7.40

    DATA(text) = `…`.


       I have the following question: what is the data type in the second dclaration with 7.40, is it a type string variable or is it a type c variable? Is there any rule so the system determines this?

       I know by reading the blog that online declarations are specially meant to be used for calling methods avoding extra code lines to declare help variables, but I’m just curious.

       Thank you and best regards!

       Dionisio

    (0) 
  25. Marc Cawood

    Any way to dynamically cast inbound parameters of FUNCTION modules?

    DATA: lv_s TYPE STRING VALUE ‘0000’.

    CALL FUNCTION SOMETHING

    EXPORTING vkorg = lv_s. ” Dumps here expecting CHAR4

    or even better:

    CALL FUNCTION SOMETHING

    EXPORTING vkorg = myStringMethod( ). ” Function returning string

    Currently the compiler says “No type can be derived from the context for the operator “CONV”.”

    (0) 
    1. Marc Cawood

      OK, gottit:

      CALL FUNCTION ‘LOAD_SOMETHING’

                 EXPORTING

                   customer  = CONV kunnr( get( ‘customer’ ) )


      Pretty cool – ABAP arrivoing in the 21st century at last!

      (0) 
    2. Horst Keller Post author

      See EXPORTING:

      a1, a2, … are general expression positions. In other words, functions and expressions can be passed as actual parameters, alongside data objects.

      -> You can pass your myStringMethod( ).

      Unlike in method calls, types cannot be specified generically (#) when a constructor expression is specified. This is because the typing of the parameters is not determined until runtime.

      -> That’s why you have to specify type string explicitly (or call a method).

      (0) 
      1. Marc Cawood

        How do you mean “constructor”?

        Theoretically, if the parameters of a Function Module are strongly typed then the compiler can know the type…

        (0) 
        1. Horst Keller Post author

          How do you mean “constructor”?

          I mean constructor expression.

          Theoretically, if the parameters of a Function Module are strongly typed then the compiler can know the type…

          Problem is that calling a function module other than calling a method is always a dynamic call. You specify the function module as a field and never directly. The compiler doesn’t know the function module.Specifying it with a literal is pseudo static so to say. Then, some test tools (extended program check) evaluate the name, but not the compiler.

          (0) 
          1. Suhas Saha

            Problem is that calling a function module other than calling a method is always a dynamic call. You specify the function module as a field and never directly.

            Wow!!! Made my day 😎

            Is this somewhere specified in the ABAP documentation? I mean although this (You specify the function module as a field) is quite obvious, but still most of the ABAPers wouldn’t have thought about it 🙂

            (0) 
            1. Paul Hardy

              I suppose that is why, when you pass the wrong type of variable into a function module, the normal syntax check doesn’t say boo, but you get a short dump when you run the program.

              And if a function module has three compulsory IMPORTING parameters and I only specify two, no syntax error either.

              I know function modules are obsolete now, and we should not be using them, but I have always thought this (lack of syntax check warnings) was a bug as opposed to anything else i.e. too difficult for SAP to fix so document it, and then pretend it is not a bug. To be fair Microsoft do this sort of thing all the time.

              It is rather like the FOR ALL ENTRIES with a blank table doing a full table scan as opposed to not bringing back anything. That is an obvious bug as well, but will never be fixed, just more and more code inspector warnings.

              I am sure I will now be told millions of reasons why the code inspector / extended syntax check can do this, but not the standard syntax check, that’s impossible, to which I will reply “it’s a bug, bug, bug, bug, bug”

              (0) 
              1. Horst Keller Post author

                I suppose that is why, when you pass the wrong type of variable into a function module, the normal syntax check doesn’t say boo, but you get a short dump when you run the program.

                And if a function module has three compulsory IMPORTING parameters and I only specify two, no syntax error either.

                Yes,, sure …

                I will now be told millions of reasons

                One important reason is, its too old to be changed. There are miilions lines of code lying around worldwide that call FMs wrongly but are never executed (dead, dead, dead), Now you go and make syntax errors out of that and have some fun.


                In fact, there are also fans of the lax checking. The reason is that you do not break a system if you change an interface in an incompatible way as you do with methods. A true story that happened to me: I changed the interface of a method of CL_ABAP_DOCU and forgot to adjust its callers before activating. As result callers became syntactically incorrect and since CL_ABAP_DOCU is called far deeper in the system than I ever expected, you could not logon to the system any more due to syntax errors in central modules. Unfortunately, there was a customer presentation running on that system in SAP Arena, Mannheim that afternoon – no lie! The presenter wasn’t amused. And correcting the error also wasn’t to simple (remember, no logon  possible). Admin had to import the old version of the class. Wouldn’t have happened with a function module.  I am not a fan of lax checking but had to learn to be more careful the hard way.





                (0) 
                1. Marc Cawood

                  In theory you (ABAP Product Management) could make a new keyword (or syntax variant) to call FMs “strongly”.

                  EXEC FUNCTION ‘MYFUNC’.

                  or

                  CALL FUNCTION MYFUNC. ” Instead of ‘MYFUNC’

                  It’s always been a weaknes of FMs that they compile when they’re definitely gonna dump.

                  (0) 
                  1. Horst Keller Post author

                    In theory, yes. But good ol’ function modules are “different”.

                    Remember, there’s no ABAP syntax for defining a function module’s interface as it is for methods.

                    (0) 
                2. Paul Hardy

                  I understand that SAP tries to be 100% downward compatible i.e. an upgrade will cause very few syntax errors to existing code. I have usually found if custom code passes the extended syntax check then you are very safe indeed,

                  However in the past I have found that after an upgrade some custom code shows a yellow warning in the standard syntax check, where it did not on the old version of ABAP.

                  As a fine example, the most annoying thing in the world “work area has more fields than selected”.

                  Show me a bus queue of programmers who thought that was a wonderful innovation.

                  If SAP were to add a yellow warning “your function module is going to dump, and here is why” it would not cause syntax errors in any old code, which would continue to behave as before (dump) and not cause any grief being being transported or whatever.

                  In fact you can add this new syntax check warning and ditch the “work area has more fields” one if you want.

                  Oh yes, and in regard to the current situation … it’s a bug, bug, bug, bug, bug.

                  (0) 
                  1. Horst Keller Post author

                    In fact you can add this new syntax check warning and ditch the “work area has more fields” one if you want.


                    Unfortunately (or maybe fortunately) I can’t 😉



                    (0) 
              2. Suhas Saha

                I suppose that is why, when you pass the wrong type of variable into a function module, the normal syntax check doesn’t say boo, but you get a short dump when you run the program.

                I always wondered, why do FMs get such love from the syntax checker? 😕 Now i know

                (0) 
  26. Mike Pokraka

    Small annoyance: Why are multiple inline declarations not permitted in the same block? Would it not make sense to just ignore something that’s already declared and reuse it?

    My main use case is an inline declaration that can be conditional and/or inside a loop.

    if some condition.

       read table x into data(wa) index 1.

    endif.

    if some other condition.

      read table x into data(wa) where …

    endif.

    It just seems awkward to have to use wa1 and wa2 here.

    Regards,

    Mike

    (0) 
    1. Horst Keller Post author

      I’d rather say big annoyance. An inline declaration is like a DATA statement at this position with all well known drawbacks. But as already discussed here from time to time, ABAP does not support local data contexts in control structures and I’m afraid it will never do.

      (0) 
      1. Mike Pokraka

        Understood, and that is exactly how an old-style DATA statement works.

        But, since ‘it will never do’, can/should an inline DATA not be made more tolerant and just be local to the method all the time?

        If a global element of the same name already exists, it will be overshadowed (current behaviour).

        If a local declaration (method, function module) already exists, the same element will be reused – as if the data() statement were not there.

        Just an idea that, to me at least, makes more sense than ever with inline declarations.

        (0) 
        1. Mike Pokraka

          Actually I’ve just thought about it some more and it could be a risk if one isn’t careful:

          loop at itab into data(wa).

              if some condition.

                  read table things into data(wa).

                  ==> oops!

          endloop.

          But then again this is down to the developer to manage. The same mistake is possible with a regular DATA declaration up front.

          (0) 
  27. Marc Cawood

    Any idea why this doesn’t work?

    /ui2/cl_json=>deserialize(

       EXPORTING json = request->get_cdata( )

       CHANGING data = CONV zmy_data_tt( lt_data )

    ).

    Instead of declaring DATA lt_data TYPE zmy_data_tt.

    (0) 
    1. Horst Keller Post author

      There are several reasons, CONV is not an inline declaration, CHANGING is neither a declaration position nor  general expression position. How should that work at a read/write position?

      (0) 
      1. Marc Cawood

        OK, I get the CHANGING argument – not the inline one since I use CONV successfully for EXPORTING parameters.

        One more for you – should this not work?

        DATA ls_ret TYPE zmytype.

        APPEND ls_ret TO data( lt_ret ).

        Is lt_ret not implicitly TYPE STANDARD TABLE OF zmytype?


        It seems to me this is corollary of LOOP AT itab INTO DATA(wa)

        (0) 
  28. Jacques Nomssi

    Hello Horst,

    given a method declaration

    get_selected_rows EXPORTING et_rows TYPE lvc_t_row

    RETURNING VALUE(rv_flag) TYPE flag,

    I can write either

    get_selected_rows( IMPORTING et_rows = DATA(lt_rows) ).

    or

    CHECK get_selected_rows( IMPORTING et_rows = lt_rows ).

    but I cannot combine those like this

    CHECK get_selected_rows( IMPORTING et_rows = DATA(lt_rows)).


    Why not? Is the method call then considered part of an expression ?

    JNN

    (0) 

Leave a Reply