Debugging, performance, JSON/XML and maintainable code

From abend to failing statement

When a COBOL program hits a program check, LE writes a message to the job log and, depending on run-time options, a CEEDUMP. The workflow is the same every time: read the message, find the program and offset, then use the listing.

CEE3207S The system detected a data exception (System Completion Code=0C7).
         From compile unit PAYCALC at entry point PAYCALC at
         compile unit offset +00000A3E at entry offset +00000A3E
         at address 1A40CA3E.

1. Note the program (compile unit) name and the offset, here PAYCALC at +A3E.
2. Open the listing for exactly the version of PAYCALC that ran. A listing from another compile is useless.
3. Use the LIST or OFFSET section to find the statement covering that offset.
4. Look at the fields that statement uses. In the CEEDUMP, find them in the program's storage section and check their contents in hex.
5. Decide whether the code or the data is wrong. For S0C7 it is usually the data.

The CEEDUMP traceback shows the chain of calls that led to the failure, which matters when the abend is in a shared subprogram: you need to know which caller passed the bad data.

Performance-minded coding

COBOL performance problems are usually about doing unnecessary work millions of times, not about clever tricks. Habits that pay off:

• Use binary (COMP or COMP-5) fields or indexes for subscripts and loop counters, not display numerics.
• Keep arithmetic in one data type. Mixing display, packed and binary forces conversions on every statement.
• Declare packed (COMP-3) fields with an odd number of digits so they fill whole bytes.
• Use SEARCH ALL (binary search) on large sorted tables instead of a sequential SEARCH.
• Do not INITIALIZE a large table inside a loop when only a few entries change.
• Let the system choose block sizes for sequential files; tiny blocks multiply I/O.

JSON and XML in COBOL

Recent Enterprise COBOL releases can convert between COBOL data structures and JSON or XML directly, which is how many batch and CICS programs now feed APIs. Availability of each statement depends on your compiler version.

All of these can fail on bad input or a too-small receiving field, so code ON EXCEPTION and check the result. Code pages matter: JSON and XML exchanged with other platforms are normally Unicode, while your COBOL data is usually EBCDIC. See JSON GENERATE / JSON PARSE.

Testing and maintainability

• One paragraph, one job. Name paragraphs with a number and purpose, such as 2100-VALIDATE-ACCOUNT.
• Avoid GO TO jumping between paragraphs; structured PERFORM and EVALUATE are easier to follow.
• Use 88-levels for conditions so tests read like business rules.
• Keep interfaces in copybooks and change them deliberately, with every user recompiled.