Fixed-length records

So far every dataset has been LINE SEQUENTIAL: a text file with a line break after each record. That is convenient on a PC, but it is not how z/OS stores data. A mainframe sequential dataset is a string of fixed-length records with nothing between them. No line breaks, no delimiters: record 2 starts at byte 81 of an 80-byte-record dataset simply because record 1 is 80 bytes long.

RECFM and LRECL

Every z/OS dataset carries its format in the catalog:

• RECFM=FB: fixed-length records, grouped into blocks. (VB means variable-length records, each with a small length prefix.)
• LRECL=80: the logical record length, in bytes.
• BLKSIZE: how many bytes are read from disk at once. The system picks a good value if you let it.

When a program opens the dataset, z/OS checks the program's record description against the catalog. If the FD says 100 bytes and the dataset has LRECL=80, the OPEN fails with file status 39.

In COBOL

Use plain ORGANIZATION IS SEQUENTIAL (it is also the default), and say how long the records are:

       IDENTIFICATION DIVISION.
       PROGRAM-ID. FIXDEMO.
       ENVIRONMENT DIVISION.
       INPUT-OUTPUT SECTION.
       FILE-CONTROL.
           SELECT CODE-FILE ASSIGN TO CODES
               ORGANIZATION IS SEQUENTIAL
               FILE STATUS IS WS-STATUS.
       DATA DIVISION.
       FILE SECTION.
       FD  CODE-FILE
           RECORD CONTAINS 10 CHARACTERS.
       01  CODE-REC.
           05  CODE-KEY       PIC X(3).
           05  CODE-TEXT      PIC X(7).
       WORKING-STORAGE SECTION.
       01  WS-STATUS          PIC XX.
       PROCEDURE DIVISION.
           OPEN INPUT CODE-FILE
           READ CODE-FILE
           PERFORM UNTIL WS-STATUS NOT = "00"
               DISPLAY "[" CODE-KEY "] [" CODE-TEXT "]"
               READ CODE-FILE
           END-PERFORM
           DISPLAY "STATUS " WS-STATUS
           CLOSE CODE-FILE
           STOP RUN.

The dataset CUR.CODES is 30 bytes on one line, with no line breaks:

EUREURO   GBPPOUND  USDDOLLAR 

Output:

[EUR] [EURO   ]
[GBP] [POUND  ]
[USD] [DOLLAR ]
STATUS 10

RECORD CONTAINS 10 CHARACTERS documents the length and lets the compiler check it against the 01 level. On the mainframe you will also often see BLOCK CONTAINS 0 RECORDS, which means "use the block size from the dataset", and RECORDING MODE IS F.

When the layout is wrong

With no delimiters, the only thing that marks where a record starts is arithmetic. If the program's layout is one byte out, every record after the first drifts. The same program reading a dataset whose records are really 9 bytes long prints:

[EUR] [EURO  G]
[BPP] [OUND US]
[DDO] [LLAR US]
STATUS 10

No error, just garbage. z/OS catches the simple case with status 39 at OPEN, but not a copybook that has the right total length and the wrong fields inside. That is why the record layout (usually a shared copybook) is treated as a contract between every program that touches the file.

Numbers in fixed-length records

Fixed-length records often hold numbers that are not readable text at all: COMP-3 packed decimal and COMP binary fields, which you will meet later. In this lesson all fields are DISPLAY digits, so you can still read the data with your eyes.