Abend

1964 to the 1990s is not the life of the language. RPG is older than both documents here, having started on IBM machines before the System/360, and this page does not date it. 1964 is the Principles of Operation, where the packed decimal format, the digit and sign codes and the data exception itself come from: the architecture this fault is a fault of. The cycle drawn above is from the 1968 RPG manual. The 1990s end is the cycle outliving both, through the System/3, the /34, the /38 and the AS/400 of 1988, still running code written against the 1968 description.

A job could end abnormally for a dozen reasons: no space on the volume, a file not allocated, a divide by zero, an operator cancelling it. This page shows the one that turns on the bytes themselves, because that is the one where the data looks fine and the arithmetic does not. The codes are from the System/360 Principles of Operation, and the sign table is checked against the System/370 manual as well, twelve years later, because a table typed from memory needs a witness that is not this page.

On System/360 and its descendants a data exception on a decimal instruction came back as a system completion code of 0C7, and everyone who worked on that hardware knows the number. It is not on this page. What the operator saw depended on the operating system, the release and the language: a 0C7 under OS/360 and MVS, a different message entirely under the System/38 and AS/400 that came later, and RPG's own halt indicators before either. The mechanism below the message is the same everywhere, so the mechanism is what this shows.

The manual says operands are fetched right to left, which would make the sign the first thing examined, but it does not promise an order in which faults are reported and a real machine may check a whole field at once. So the table marks every half-byte that is wrong rather than claiming to know which one stopped the program.

They are not a paraphrase. Each item's description is taken from the November 1968 RPG language manual, form C24-3337-3, which has a real text layer, so no OCR is involved. The test suite requires a distinctive clause from every one of the fifteen to appear in the archived copy, so a step reworded to read better would turn it red. Figure 31 in the manual is the flowchart these numbers refer to; this page does not reproduce the figure, only the numbered text beside it.

RPG supports control levels L1 through L9, nested, so a report can break on branch inside region inside division. This page drives two, L1 and L2, because one level never shows why they are numbered: it takes a second to see a group total print inside a division total, and to see that breaking at L2 turns L1 on with it. The engine carries all nine indicators and sets them all on at end of file, which is what the manual says happens; levels three to nine are simply never reached by this deck.

The calculations the cycle runs at items 6, 7 and 11 are a fixed toy: accumulate a value, print a group total on a control break, print a final total at the end. The specification sheet lower down is a real sheet with the manual's real column numbers, and it is checked the way the compiler checked one, but it does not compile: editing it does not change what the cycle above computes. It changes how many days the deck took to come back, which is the other half of the subject.

Real turnaround was not reliably a day. It was a day at a well-run shop with an overnight batch window, and it was two hours if you had the machine to yourself, and it was three days if the operator dropped your deck or the tape you needed was mounted on another job. The model here says one submission a day and that faults of one severity class come back together. The second half is a fact about how a compiler pass works; the first is a round number standing in for a range.

The compiler graded diagnostics and returned the worst severity it found, and what its table grades is the outlook rather than the output: 4 is minor errors with successful execution probable, 8 is errors with unsuccessful execution possible, 12 is serious errors with it probable, 16 is critical errors and normal execution impossible. Whether a later step in the job ran on a given code was the job’s own COND parameter, not the compiler’s decision. This page used to say 8 and above meant a listing and no program, which is neither what the table says nor how a job worked. At 4 you got a program, because the compiler had assumed something on your behalf and carried on, and its diagnostics say so in as many words. That is the failure this page would put first if it could only keep one: not the day you lost to an error, but the day you did not lose, because you got output back and it was the wrong output.

Which faults are errors and which are warnings comes from the compiler's own return-code scale, and the column ranges every check works from are quoted on the sheet itself, out of the manual's prose. The sentences the page prints are not. The manual's diagnostic message text is in its Appendix G, and the scan's text layer is too damaged there to quote, so each message here is described in this page's own words. Nine checks, where a real compiler had hundreds: this one can see a missing operation, an operation code it does not know, a missing factor or result, a length or decimal position that is not a number, more decimals than the field has positions, a control level that is not one, a result field defined twice or not at all, and a name nothing defines. It cannot see anything about the other specification sheets, because they are not here.

JCL, which was its own language and its own class of overnight failure. Matching and secondary files. Table lookup. Exception output. Halt indicators are here only as a switch to show the branch to item 12. And nothing about the machine underneath: no storage estimates, no core, no operator. The subject is one loop.