Skip to content

AT NEW, AT END OF, SUM — control breaks

Level: 201 · working knowledge

One line: Control-break statements inside a LOOP AT fire when a component's value changes from the previous row (AT NEW), is about to change (AT END OF), or at the first and last row — with SUM filling the numeric fields with group totals — and every one of them assumes the table is sorted by the break fields, which nothing enforces.

The shape

SORT lt_rows BY region city.
LOOP AT lt_rows INTO DATA(ls_row).
  AT NEW region.            " first row of a new region value
  ENDAT.
  AT END OF region.         " last row of that value
    SUM.                    " numeric fields now hold the region totals
  ENDAT.
ENDLOOP.

AT FIRST and AT LAST bracket the whole loop. SUM inside AT LAST gives the grand total. It all reads naturally, and the rules underneath are why GROUP BY exists:

  • Adjacency, not value. A break fires when the value differs from the previous row's. An unsorted table produces a break per change, not per group.
  • The break field and everything to its left. AT NEW city fires when city or any component before it changes. Component order in the structure is part of the logic.
  • Masking. Inside AT NEW/AT END OF, every component after the break field is overwritten with asterisks in the work area. Reading ls_row-city inside AT NEW region gives **********, not the city.
  • SUM needs a work area: INTO, not ASSIGNING, and it sums all numeric components — including an id you did not mean to add up.

What replaced it

LOOP AT … GROUP BY (7.40) groups by value, needs no sort, masks nothing, and totals with a REDUCE over the group members. Read AT NEW fluently — it is in every report older than 2015 — and write GROUP BY.

z_kw_at_new.prog.abap — pasted here by tools/check_examples.py. Syntax-checked, never run: abaplint parses and type-checks it against the release in abaplint.json, but no system has produced output for it, so this page shows none.

REPORT z_kw_at_new.

TYPES ty_amount TYPE p LENGTH 9 DECIMALS 2.
TYPES: BEGIN OF ty_row,
         region TYPE c LENGTH 2,
         city   TYPE c LENGTH 10,
         amount TYPE ty_amount,
       END OF ty_row.
TYPES ty_rows TYPE STANDARD TABLE OF ty_row WITH EMPTY KEY.

DATA(lt_rows) = VALUE ty_rows( ( region = 'EU' city = 'Paris'  amount = '10.00' )
                               ( region = 'US' city = 'Boston' amount = '20.00' )
                               ( region = 'EU' city = 'Rome'   amount = '30.00' ) ).

" A control break compares with the PREVIOUS row, so the table must be sorted
" by the break fields, in that order, or the breaks fire at random.
SORT lt_rows BY region city.

LOOP AT lt_rows INTO DATA(ls_row).
  AT FIRST.
    WRITE: / 'start'.
  ENDAT.

  AT NEW region.
    " Inside AT NEW, every component AFTER the break field is masked with
    " asterisks in the work area: ls_row-city is not readable here.
    WRITE: / 'region', ls_row-region.
  ENDAT.

  WRITE: / '  ', ls_row-city, ls_row-amount.

  AT END OF region.
    " SUM fills the numeric components of the work area with the group totals.
    SUM.
    WRITE: / '  subtotal', ls_row-amount.
  ENDAT.

  AT LAST.
    SUM.
    WRITE: / 'grand total', ls_row-amount.
  ENDAT.
ENDLOOP.

If you are coming from another language

  • SQL. GROUP BY with SUM( ) — and the AT NEW model is what a report writer did before the database could do it.
  • Python. itertools.groupby, which has exactly the same adjacency rule and the same "sort first" footnote.
  • awk. The pattern $1 != prev { … } { prev = $1 } is AT NEW written by hand.

See also