Fix #4242: record a quoted PROGRAM-ID (PROGRAM-ID. 'EPSCSMRD'.)

[squid-protocol]

Resolves #4242.

What was broken

The COBOL class_start rule (languages/cobol.py) requires \b right after the separator that follows PROGRAM-ID.. A quote is not a word character, so a program named by a literal, PROGRAM-ID. 'EPSCSMRD'. or "NAME", never matched. DBB epscsmrd.cbl has 13 quoted PROGRAM-IDs in column 9, and it recorded no class row.

Changes

• class_start now allows an optional quote on each side of the name. The capture stays the bare name, and the \b guard still holds: ' is not in the sequence-area character class, so that prefix can't swallow it. The rule's comment block gains entry 4 explaining this.
• Fixtures (test_cobol.py):
  • valid: single-quoted in Area B, double-quoted, quoted with IS INITIAL PROGRAM;
  • invalid: a commented-out quoted PROGRAM-ID, and PROGRAM-ID. inside a MOVE literal;
  • pathological: the quoted name on the next line.
• Field testing: defect D041 (engine, program_id, round 7, found by the fix(answer-key): model multi-program sources (sibling / nested programs) (#4206) #4226 census). The report and fact_channel_confidence.json are regenerated; program_id's clean-round count resets, as it should.
• Answer-key README: the multi-program paragraph no longer says the engine records no PROGRAM-ID for epscsmrd.cbl.

What moved, and why

• Golden masters (31 differences per mode, scoped in full, not from the truncated crucible_check view): NIST CCVS85 PDC0001.cbl (Program-Id. 'P1'.) gains its class, and every other difference rolls up from that one file:

  • Entity Declarations 0 → 1;
  • its architectural-profile coordinates;
  • DNA hits +1 and the GPU / graph digests;
  • batch entry 287 → 288 (P1 is a batch program no JCL step runs) and its missing-inputs example;
  • the LLM brief.

  Nothing outside that file moved.

• DBB completeness (test_completeness.py): transactions (2, 3) → (2, 4). EPSCSMRD is now a recognized CICS program, and no transaction in the repository reaches it. Re-pinned with a comment.

• Full-corpus refraction differential (DBB): epscsmrd.cbl is compared for the first time, because before this fix the file had no program. That gives 33 unexplained deltas, blessed with a note:

  • 7 paragraph/old: the forge reads the sibling programs' IDENTIFICATION / DATA DIVISION headers as units of the first program;
  • 26 schema_column/db: the same names as this file's 26 forge_flat_schema record deltas.

  Both come from reading a multi-program source as one program, which is owned by COBOL forge/differential reads a multi-program source as one program, blocking EPSCSMRD onboarding #4243 (with COBOL engine and answer key read only the first program's DATA DIVISION in a multi-program source #4245 for the data divisions). Re-bless when that lands.

Verification

• epscsmrd.cbl scanned alone: class_data has 13 rows (EPSCSMRD … EPSCSMRE); before, 0.
• tests/cobol_mainframe + tests/cics_crucible + COBOL extraction (strict included), with every pinned corpus fetched: all pass after the re-pin and the bless (1631 passed before those two; the two DBB tests are the only ones that moved).
• Ground-truth ledger: clean (epscsmrd.cbl is still in excluded_programs).
• crucible_check.py --update --yes: both fixtures re-blessed, scoped as above.
• ruff / mypy / dead-key / ast-accuracy audits: all clear. Format clean.

Not done

• PROGRAM-ID as a function boundary (cobol func_start: free-format source reads data items as paragraphs and misses PROGRAM-ID units (CobolCraft) #4203, needs a design call).
• Other quoted-name readers in cobol_to_cobol tools (cobol_dag_architect._PROGRAM_ID, cobol_compiler_forge) don't take a quote either. They are outside this issue; cobol_jcl_forge already does.

🤖 Generated with Claude Code

[github-actions]

🐦‍⬛ Muninn Security Scan

✅ No security issues found.

🐦‍⬛ Powered by Muninn · Skald Lab