[squid-protocol]

Part of #5076 (the issue stays open: the unproven programs below keep their named blockers).

What changed

• Screen compare reads every OCCURS occurrence (tests/tools/equivalence_cics.py): screen_layout no longer refuses a symbolic map with OCCURS; it lays out every occurrence by subscript (ACCOUNTO(3)), as the COMMAREA compare does (Equivalence: COMMAREA OCCURS fields are compared at the first occurrence only; det LINK of an OCCURS DTO passes the first occurrence's extent #4765). screen_key names each one as the generated screen's screenValues() does (ACCOUNT.3); the SEND-MAP decode and map_subfields use it. An OCCURS DEPENDING ON in a map, and a RECEIVE typed into one occurrence, are refused by name.
• Found by the new compare, fixed in the det translator (det/cics.py): Generated.screen_fields read the screen class with a regex that missed a repeated field (values.put("ACCOUNT" + "." + (i + 1), ...)), so the port's SEND MAP silently left every ACCOUNT line out (first run: 6 BNK1CCA scenarios DIFFERED, java: None). SEND MAP and RECEIVE MAP now cover each occurrence (screen_occurs, screen_slots).
• New case cbsa-bnk1cca (BNK1CCA, map BNK1ACC with OCCURS=10): 20 scenarios (AIDs, valid / zero / unknown customer, 0 / 1 / 3 / 10 / 12 accounts, LINK INQACCCU PGMIDERR and NOTAUTH). Proven: 25/25 paragraphs, 29/41 branches (the rest: failure branches after SEND/RETURN no fault can reach, a non-numeric NUM field).
• New case cbsa-delcus (DELCUS): proof_blockers on HEAD showed it as the one burned CICS program still with no case. 16 scenarios, 22/22 paragraphs, 13/16 branches. Not proven (see below).
• Tests: test_call_compare_4778.py (OCCURS tests replaced: every occurrence listed, a differing later occurrence is a difference, subfields per occurrence, ODO and occurrence-typed input refused), new test_det_screen_occurs_5076.py.

Class table (burned estates; proof_blockers on HEAD before the new cases, --no-census: 98 CICS programs translate whole, 53 proven)

Program	Status
BNK1CCA	case added, proven
DELCUS	case added, not proven: file change after INQACCCU's SYNCPOINT ROLLBACK (cbsa-delcus, in det_sweep_baseline.json, register Q3). Every event, COMMAREA, ACCOUNT, PROCTRAN and ABNDFILE equal; 3 scenarios differ in the CUSTOMER file
BNK1UAC, BNK1TFN, BNK1CCS	cased, unproven: java-facade RETURN LENGTH differs from the record (#5089); stays refused, not loosened (#5147's policy)
CRECUST (via BNK1CCS)	CEEDAYS picture, #4952 / #5064
ZECS000, ZECS003	run-time file names, #4771 (ZECS003 in det_sweep_baseline.json)
ECS001, ASYNCPNT	cased and proved by #5087 / #5147
BANKDATA	batch Db2 loader, no EXEC CICS: out of scope
After this PR no burned CICS program carries a no case gap (proof_blockers on HEAD before the DELCUS case listed only DELCUS; BANKDATA and the CardDemo batch programs DBUNLDGS, PAUDBLOD, CBSTM03B have no EXEC CICS). Non-burned estates untouched. The full 16-program list of the issue was measured on an older commit; this table is the re-measure on HEAD (burned estates only, by policy).

Evidence

• det_port.py run cbsa-bnk1cca: PROVED (20 scenarios, java-facade equal too). det_port.py check: 116/117 ports unchanged (only the new case).
• proof_sweep.py --det-only --cases cbsa-delcus: not proven, "as expected" against the baseline entry.
• pr_gates.py --fast: 6/6 pass. pr_gates.py --ratchets: 8 pass, 0 fail, 0 skipped.
• pytest: test_det_translate, test_call_compare_4778, test_occurs_compare_4765, test_equivalence_cics, test_det_cics_map_names_4270, test_bms_screens, test_proof_blockers all pass.
• Golden masters not touched (no scan-path change).

Sibling gaps (not in this PR)

• A recoverable FILE changed after a SYNCPOINT ROLLBACK is lost on the Java side (the harness marks the whole file transaction rollback-only; Db2 was fixed in COBOL translator: CICS command breadth (START/RETRIEVE, channels/containers, GETMAIN, LOAD, CONVERSE, INQUIRE, ASSIGN, counters, HANDLE ABEND PROGRAM) #4270). Needs the task's file transaction restarted at the rollback: a harness mechanism of its own. Blocks cbsa-delcus.
• Typing into one occurrence of a repeated field on RECEIVE MAP is refused by name (no case needs it).

Lessons for agents

• cics_census.py survey --baseline surveys an OLD commit; for HEAD use survey --no-census --out DIR --label head, then proof_blockers.py DIR --label head --no-census.
• A pgrep -f wait loop matches its own shell and never ends; wait on a log marker instead.

🤖 Generated with Claude Code

[squid-protocol]

No description provided.

[squid-protocol]

No description provided.

[squid-protocol]

No description provided.

[squid-protocol]

No description provided.

[github-actions]

🐦‍⬛ Muninn Security Scan

✅ No security issues found.

🐦‍⬛ Powered by Muninn · Skald Lab