Similar to SQL's GROUP BY,
there is also a GROUP BY
for working with internal tables that can be used behind LOOP AT itab
or in the form IN GROUP
in a table iteration with
FOR.
It replaces the clumsy group level processing with statements AT NEW ...
that relies on the order of table columns and content that is sorted
respectively.
Thi cheat sheet explains the grouping of internal tables step by step
using a very simple case of an internal table spfli_tab that
is filled with data from the database table SPFLI. The
following steps show how the content of the internal table can be
grouped using LOOP AT GROUP BY.
The simplest form of grouping is by one column without explicitly specifying the output behavior of the group loop:
LOOP AT spfli_tab INTO wa
GROUP BY wa-carrid.
... wa-carrid ...
ENDLOOP.Within the loop, there is access to the work area wa, in
particular to the component wa-carrid that is used for
grouping. The work area wa contains the first line of each
group and represents the group in the loop. This is called
representative binding.
To access the members of a group, a member loop can be inserted into the group loop:
LOOP AT spfli_tab INTO wa
GROUP BY wa-carrid.
...
LOOP AT GROUP wa INTO DATA(member).
... member-... ...
ENDLOOP.
...
ENDLOOP.The member loop is executed using the group represented by wa
and its members are assigned to member and are available in
the member loop.
To group by more than just one criterion, a structured group key is defined as follows. In the simplest case, the grouping criteria are columns of the internal table:
LOOP AT spfli_tab INTO wa
GROUP BY ( key1 = wa-carrid key2 = wa-airpfrom ).
... wa-carrid ... wa-airpfrom ...
ENDLOOP.This is also a representative binding in which the work area
wa is reused in the group loop to access the group key.
To access the members of the groups, the exact same member loop can be inserted as when grouping by one column.
By explicitly specifying an output area for the group key, a group key binding can be defined explicitly instead of the representative binding in which the output area of the group loop is reused:
LOOP AT spfli_tab INTO wa
GROUP BY wa-carrid
INTO DATA(key).
... key ...
ENDLOOP.The difference to the example with representative binding is the
INTO addition after GROUP BY. Instead of reusing
wa, an elementary data object key represents the
group. This can be generated inline. The additions GROUP SIZE,
GROUP INDEX,
and WITHOUT MEMBERS
can only be used in the group key binding, which gives it more functions
than the representative binding. If these are not required, the
representative binding can be used. The group key binding can also be
used to make the use of the group key in the loop more explicit.
Inserting a member loop works in the same way as in the representative
binding, with the difference that a group is now addressed by
key instead of wa.
LOOP AT spfli_tab INTO wa
GROUP BY wa-carrid
INTO key.
...
LOOP AT GROUP key INTO member.
... members ...
ENDLOOP.
...
ENDLOOP.Finally, the group key binding for structured group keys:
LOOP AT spfli_tab INTO wa
GROUP BY ( key1 = wa-carrid key2 = wa-airpfrom )
INTO DATA(key).
... key-key1 ... key-key2 ...
ENDLOOP.Here, key is a structure with the components key1
and key2. A member loop can be inserted in exactly the same
way as when grouping by one column.
If the group members are not relevant, the addition NO MEMBERS
can be used to save time and memory.
LOOP AT spfli_tab INTO wa
GROUP BY ( key1 = wa-carrid key2 = wa-airpfrom
index = GROUP INDEX size = GROUP SIZE )
WITHOUT MEMBERS
INTO DATA(key).
... key-key1 ... key-key2 ... key-index ... key-size ...
ENDLOOP.It is no longer possible to use a member loop here. Instead, the group
key was enriched with optional components for further information using
GROUP INDEX
GROUP SIZE.
Note the steps outlined here about how to import and run the code.