[squid-protocol]

Fixes #4669, Fixes #4670

Part of #4270. P-scaled PICTUREs in the det translator / runtime, measured against the oracle (cobc 3.1.2 -std=ibm) item by item.

Modelled

A P is a scaling position, never stored. Item.digits now counts a numeric item's 9s and Item.scale counts the Ps: 99PP / 99PPV are 2 digits at scale -2, VPP99 / PP99 are 2 digits at scale 4. Nothing else in the runtime needed to change for these: BigDecimal.setScale(-2) already truncates the way cobc does.

• DISPLAY (zoned) right-P and left-P items, signed and unsigned, and left-P binary items. Covered: VALUE, MOVE into them (literals truncated at both ends, from an item, from an alphanumeric), MOVE out to numeric-edited, COMPUTE / ADD receivers (ROUNDED, ON SIZE ERROR), operands, and comparisons of the bare item.
• A right-P item MOVEd to an alphanumeric gets one zero per P (1300 ). A left-P binary item gets its Ps as leading digits. Both match cobc (Cobol.sourceText).
• det translator: a PIC SVPP99 COMP-3 item crashes with ValueError instead of a refusal by name #4669's PIC SVPP99 COMP-3 no longer crashes the layout with a ValueError: its image is IBM's 2 bytes. det runtime: MOVE 1200 into a 99PP item throws in Cobol.store #4670's MOVE 1200 into 99PP now stores 12 at scale -2.

Refused by name (the oracle has no usable answer)

• P-scaled COMP-3: cobc 3.1.2 puts digits into the sign nibble (MOVE 5600 TO PIC S99PP COMP-3 reads back as <600).
• Right-P or COMP-5 binary: cobc loops forever on MOVE 0 into PIC 99PP COMP.
• DISPLAY of a P-scaled item: cobc writes 99PP VALUE 1200 as 0012 and S99PP -3400 as 0034-.
• A P-scaled operand in a condition's arithmetic expression: cobc says IF PR / 7 > 171 and IF PR * 2 > 2399 are false for 1200.
• A right-P operand in a COMPUTE with a division or more than two operands: cobc computes (PR / 7) * PR with truncated quotient digits, and I5 + PR * 0.5 and B0 * .03 * PS as if the product were 0.

The #4668 refusal (a P-scaled operand in a numeric-edited comparison) is not touched.

Shared functions touched (for the det/cobolrt train: #4654, #4677, #4663, #4666, #4668)

• layout.py: Item.digits, Item.scale, new Item.p_scaled, Item.elementary_size (one line)
• gen.py: new module function p_unmodelled, Gen.resolve (3 lines at its end), Gen._init_one (3 lines at its top), Gen.display_operand (3 lines), Gen.rel (numeric branch, 2 lines), new method Gen.p_scaled_expr, the COMPUTE branch of the statement translator (1 line)
• cobolrt/Cobol.java: sourceText (its two returns), new helper pZeros

Evidence

• Tests (tests/cobol_mainframe/test_det_programs.py), all 12 fail on origin/main and pass here:
  • PSCALE: an end-to-end program run in bytes, typed and groups modes. It has about 140 output lines over 8 P-scaled items and is equal to GnuCOBOL.
  • 8 refusal cases.
  • A layout unit test for digits / scale / size and the packed bytes.
• Random differential fuzz: the det-4655 generator plus 6 P-scaled operands and 3 P-scaled receivers, 40 programs of 100 statements, bytes and typed modes. The generator leaves out the refused forms.
  • This fuzz is noisy on main without any P items: about 10 differences per program, from arithmetic precision and binary overflow (det runtime: a zero divisor without ON SIZE ERROR throws in the port; the oracle leaves the receiver unchanged #4655's family).
  • So each program was also run with every P picture replaced by its unscaled equivalent, and only differences found in the P version alone count.
  • Result: 0 P-specific differences. The one candidate turned out to reproduce with the unscaled picture and is filed as det runtime: a result overflowing a binary receiver's bytes (no ON SIZE ERROR) is stored unlike GnuCOBOL #4684.
• det_port.py check vs origin/main: all 68 ports change, but only in the runtime's Cobol.java (17 lines). No service changed.
• proof_sweep.py --det-only: 65 PROVED, and 3 NOT PROVEN that are already in the baseline (carddemo-cardupdate, carddemo-intcalc-generated, mortgage-mpmt). Result "sweep: as expected".
• evidence_report.py --check rc 0. test_evidence.py and test_evidence_report.py: 71 passed.
• pr_gates.py --fast: 5/5 pass. pr_gates.py --ratchets: 7 pass, 0 fail, 0 skipped.

Follow-ups filed

• det: a relation's NOT at the end of a line, its operator on the next, crashes the PROCEDURE DIVISION parse #4682: a relation's NOT at the end of a line with its operator on the next crashes the PROCEDURE DIVISION parse (found by the fuzz, not P-related).
• det runtime: a result overflowing a binary receiver's bytes (no ON SIZE ERROR) is stored unlike GnuCOBOL #4684: a result overflowing a binary receiver's bytes with no ON SIZE ERROR is stored unlike cobc (not P-specific).

🤖 Generated with Claude Code

[github-actions]

🐦‍⬛ Muninn Security Scan

✅ No security issues found.

🐦‍⬛ Powered by Muninn · Skald Lab