equivalence: non-digit zoned data as MOVE sender, operand, comparand; the oracle's sign rewrite (#4662)

[squid-protocol]

Fixes #4662. Part of #4270.

stacked on #4663 (→ #4653) — merge those first.

What. The rest of register C11: a zoned item holding non-digits (spaces, letters, low-values, X'FF') is modelled as GnuCOBOL 3.1.2 (-std=ibm) reads it, per shape, measured against cobc over unsigned / SIGN TRAILING / LEADING / TRAILING SEPARATE / LEADING SEPARATE senders. IBM documents no result for such data; C11 says so and lists each shape with its libcob function.

Shape	libcob	Status
sign byte read (GET_SIGN) and rewritten (PUT_SIGN) on every MOVE-from / arithmetic read, digit-only too (0123 → 012C); ! +1, X'FF' +0, space stays a space	cob_get_sign / cob_put_sign	modelled
MOVE zoned → zoned (1A3 stays 1A3)	store_common_region	modelled
MOVE → numeric-edited (spaces stay spaces)	cob_move_display_to_edited	modelled
MOVE → packed	cob_move_display_to_packed	modelled
COMPUTE / ADD / SUBTRACT operand	cob_decimal_set_display (skip leading low-nibble-0 bytes, u64 wrap; first byte FF / 00 = ±10^size)	modelled; 19+ digits refused by name
ADD x TO y / SUBTRACT x FROM y, x ≤ 9 digits (not COMP under TRUNC(STD))	cob_add_int (y, cob_get_int (x))	modelled
IF x = 0, >, <, IS POSITIVE	cob_cmp_numdisp (unsigned / trailing sign, no decimals) or the decimal	modelled
MOVE → binary (#4652)	cob_move_display_to_binary	now also ! / X'FF' signs; a bad SEPARATE sign still refused
zoned with a P position as sender; receiver that itself holds non-digits (ADD 1 TO spaces)	—	not modelled, #4692

Tests: tests/cobol_mainframe/test_det_nondigit_zoned.py — per sender shape x ~24 operations x cases, port == nondigit_zoned_expected.txt under TRUNC(STD) and TRUNC(BIN); with EQUIVALENCE_E2E=1 the file is re-checked against cobc (26 pass with Docker). All 10 shape tests fail on the base, pass here.

Shared functions touched (siblings #4654 / #4677 / #4686 / #4666 / #4668 / #4685):

• Cobol.move(Field,Field,Charset) (split into move + moveBytes; the sign GET/PUT around it), Cobol.nonDigitToBinary (reads a Zoned.Raw), Cobol.num (zoned operand), Cobol.compare(Field,Field), compare(Field,BigDecimal), compareFigurative(ZEROS) (via cmpNum), new Cobol.intOperand / cmpNum.
• Editing.format: an overload with rawDigits / negative; the old signature delegates (3 changed lines inside).
• gen.py: arith (ADD/SUBTRACT int path), rel and the sign-class condition (cmp_num), new int_operand.
• New cobolrt/Zoned.java. Not touched: Cobol.store (binary path, det runtime: a result overflowing a binary receiver's bytes (no ON SIZE ERROR) is stored unlike GnuCOBOL #4684).

Evidence. det_port.py check vs #4663: all 79 ports change (runtime file), 30 services change (ADD/SUBTRACT/compare calls); det_port.py run of those 30: 28 proven, 2 not proven — carddemo-cardupdate, carddemo-intcalc-generated, both KNOWN_UNPROVEN already. No verdict regression (GenApp LGTESTP1-4 ran on local Db2). pr_gates.py --fast 5/5; --ratchets 7 pass, 0 fail, 0 skipped; evidence_report.py --check clean.

Follow-ups filed: #4690 (hex literal X'0A'/X'0D' breaks the emitted Java), #4691 (DISPLAY of bytes > X'7F': UTF-8 vs raw), #4692 (non-digit receiver, P-scaled sender). Note: digit-only programs change too — a signed zoned sender's sign byte is rewritten as the oracle does.

🤖 Generated with Claude Code

[github-actions]

🐦‍⬛ Muninn Security Scan

✅ No security issues found.

🐦‍⬛ Powered by Muninn · Skald Lab