det: LINK RESP2; RETURN/LINK COMMAREA LENGTH past the DTO passes the bytes; opaque bytes passed on (#4679)

[squid-protocol]

Stacked on #4678 -- merge that first (once it merges, a rebase drops its commit). Fixes #4679. Part of #4270.

Decisions (IBM refs)

• (a) LINK RESP2. A failed LINK stored RESP2 0. It now stores DetCics.linkResp2(condition): LENGERR 11 (COMMAREA length < 0 or > 32763), PGMIDERR 1 (no installed definition, autoinstall off). IBM EXEC CICS LINK conditions. These are the only two CicsTask.link and the stub's GGCLINK (ggcics.c) raise, with the same RESP2. The region never raises IBM's other LINK conditions (PGMIDERR 2/3, NOTAUTH 101, INVREQ, ROLLEDBACK, SYSIDERR, TERMERR), so they are not mapped. The XCTL mapping from det: XCTL COMMAREA LENGTH past the DTO passes those bytes; failed XCTL sets RESP2 (#4501) #4678 is reused, not duplicated.
• (b) RETURN TRANSID COMMAREA LENGTH past the DTO passes the LENGTH bytes (DetCics.commareaOut), as det: XCTL COMMAREA LENGTH past the DTO passes those bytes; failed XCTL sets RESP2 (#4501) #4678's XCTL does. IBM EXEC CICS RETURN: the next program's EIBCALEN is LENGTH. On the harness side, equivalence_cics.java_commarea reads a byte[] COMMAREA (Jackson base64, CCSID 037) the way it reads the stub's RETURN area: in the case's data page, by the case's COMMAREA layout (decode_record). Bytes past the layout are compared on neither side. This is what broke the CardDemo ports in det: XCTL COMMAREA LENGTH past the DTO passes those bytes; failed XCTL sets RESP2 (#4501) #4678 (from_java cannot read a byte[]).
• LINK LENGTH past the target's DTO lays that many bytes over the caller's storage (Cobol.commarea, capped at 32763), not just the DTO's size. IBM EXEC CICS LINK: LENGTH is the COMMAREA length, passed by reference. Example: GenApp's LGxxVS01 LINK ERROR-MSG LENGTH 101 to LGSTSQ, whose DTO is 99 bytes.
• LINK to a target no DTO types (cics-crucible ca-link-lengths undefined-program: CALINK -> CAGONE, not in the CSD) is no longer a hole. The COMMAREA travels as its bytes (DetCics.commareaBytes), so the case now proves PGMIDERR RESP2 1 end to end.
• (c) Opaque bytes passed on. A receiver keeps the bytes past its own DFHCOMMAREA record in Storage.beyond (from a byte[] COMMAREA or a LINK's area). A further RETURN, XCTL or LINK of that area with the LENGTH passes them on (commareaOut / Cobol.commarea read them). A LINK target's writes there go back by reference (Cobol.commareaBack, the receiver's caBack).
• oracle_assumptions.md X10: the det: XCTL COMMAREA LENGTH past the DTO passes those bytes; failed XCTL sets RESP2 (#4501) #4678 note is extended with all of the above.

Touched

det/cics.py (LINK branch, commarea_out), det/program.py (DFHCOMMAREA receiver), cobolrt/Storage.java (beyond), cobolrt/Cobol.java (beyond, commarea, commareaBack), cobolrt/cics/DetCics.java (linkResp2, commareaBytes, commareaOut), tests/tools/equivalence_cics.py (java_commarea), docs/language_status/oracle_assumptions.md, tests in test_det_translate.py (translator shapes plus a javac/java runtime test for beyond pass-through and write-back) and test_equivalence_cics.py.

Evidence

• det_port.py check vs origin/fix/4501-xctl-commarea-length: all 68 ports change (runtime files). Every one was re-proved with proof_sweep.py --det-only (Db2 included): 65 PROVED, "sweep: as expected". The 3 not proven are exactly det_sweep_baseline.json: cardupdate (Porting loop at scale: baseline-reuse proofs, run-many, adopt; COCRDUPC recorded as not proven #4085, only commarea.CARD-UPDATE-CVV-CD differs), intcalc-generated (D1), mpmt (COBOL oracle/translator: floating point and compiler options (COMP-1/2 hex float, NUMPROC, ARITH(EXTEND), TRUNC(OPT), INTDATE, COMP-5 byte order) #4271).
• 8 CardDemo ports now RETURN bytes (cardlist, cardview, cardupdate, acctview, acctview-generated, acctupdate, cotrtlic, cotrtupc), plus mortgage-cmort. All prove except cardupdate (Porting loop at scale: baseline-reuse proofs, run-many, adopt; COCRDUPC recorded as not proven #4085, unchanged), java-facade side included. Example: cardlist's RETURN event is a 2000-byte base64 COMMAREA.
• The later no-DTO LINK commit only adds the unused commareaBytes helper to these 68 ports (a second det_port.py check shows no service change), so they were not re-swept.
• cics-crucible det ports (cics_case_ports.py + cics_crucible.py --sides cobol-stub java-ported): ca-link-lengths 8/8, 0 holes (was 1 hole and 1 fail), ca-xctl-versions 8/8, hc-abend-link 10/10.
• Evidence: evidence.py status --ci passes. No record's port, case, pin or declared differences changed; harness/generator staleness is left to the scheduled refresh (An evidence record per ported program: the claim, the oracle, provenance, approval, staleness #4048). evidence_report.py --refresh produced no diff, --check passes, test_evidence.py + test_evidence_report.py 71 passed.
• test_det_translate.py 260 passed (JDK 17, no skips); test_equivalence_cics.py 102 passed, 2 skipped (E2E).
• pr_gates.py --fast 5/5. --ratchets 7 pass, 0 fail, 0 skipped (JDK 17 env).

Overlaps

• equivalence: CICS proofs compare DISPLAY output and keep low-values apart from spaces (#4635) #4660 (equivalence_cics COMMAREA compare): java_commarea calls decode_record(data, ca_fields, enc) exactly as the COBOL side's RETURN/XCTL decode does. When equivalence: CICS proofs compare DISPLAY output and keep low-values apart from spaces (#4635) #4660 lands, give the byte[] path the same exact=True so LOW-VALUES are not spaces and only trailing spaces drop. Expect a small textual conflict in _java_events.
• det runtime: browse/DELETE RESP2 as IBM states it; signon model port LOW-VALUES (#4657, #4659) #4661 (task.resp2()): LINK's RESP2 comes from its condition, as XCTL's does. If CicsTask grows a per-command RESP2, LINK can switch to it.
• Follow-up filed: det facade: handleTransaction answers raw EBCDIC bytes when a RETURN's COMMAREA LENGTH is past the DTO (#4679 follow-up) #4695. Generated facades of the RETURN-bytes programs now answer a byte[] from handleTransaction.

🤖 Generated with Claude Code

[github-actions]

🐦‍⬛ Muninn Security Scan

✅ No security issues found.

🐦‍⬛ Powered by Muninn · Skald Lab