det translator: model DECIMAL-POINT IS COMMA; document the refusals #4462 keeps (#4462)

[squid-protocol]

Fixes #4462

What

Slice 5, the close-out of #4462: SPECIAL-NAMES DECIMAL-POINT IS COMMA is modelled. Every other remaining translator-refuses-* cause is kept on purpose, refused by name, and listed with its reason in docs/language_status/det_port_design.md.

• det/source.py: comma_literals passes each numeric literal's decimal comma to the grammar as a point, keeping every column in place (VALUE 1000,00, MOVE 0,5 TO T-WERT (2), -12,5). A comma between digits is the decimal point, as cobc reads it. A separator comma is followed by a space. PICTUREs, literal text and names are left alone. Under the clause, a literal written with a decimal point (1.5) is refused by name.
• det/layout.py: Item.decimal_comma. An edited PICTURE's , is its decimal point (scale) and its . is an insertion character.
• det/gen.py: field factories get .decimalComma(). NUMVAL / NUMVAL-C / TEST-NUMVAL(-C) read their argument with . and , swapped.
• Runtime (cobolrt/Field.java, Editing.java, Cobol.java): Field.decimalComma is used by editing (Editing.format already had the parameter; shape now has it too). It is also used when an alphanumeric sender is MOVEd into a numeric or edited receiver (alnumToDisplay, parseAlnum: , is the point and . is skipped, as libcob reads it). Touches Cobol.java in move()'s NUMERIC / edited branches, source(), alnumToDisplay, parseAlnum, edit() and num(). It does not touch align or arithmetic (det runtime: truncate arithmetic intermediates as cobc's ARITHMETIC-OSVS does (#4287) #4654) or the binary paths (equivalence: model the oracle's MOVE of a non-digit zoned item to a binary item (#4652) #4663), but equivalence: model the oracle's MOVE of a non-digit zoned item to a binary item (#4652) #4663's spot next to move() is nearby, so a rebase may touch the same lines. Touches gen.py in factory() and _func(), not the arithmetic code det runtime: truncate arithmetic intermediates as cobc's ARITHMETIC-OSVS does (#4287) #4654 changes.

Decisions per cause

cause	decision
DECIMAL-POINT IS COMMA (ZINSBER)	modelled (above). ZINSBER now stops at its next gap: a national letter in the data names GEBÜHR / BETRÄGE. It is refused by name, and #4664 is filed for it
national / DBCS text (KYUY*)	kept. #4272 owns it. KYUYJP refuses by name with no UnicodeEncodeError (test_det_source_4462 pins it)
IDMS (LNIDMS01)	kept, #4532
source defect (GenApp polloo2.cpy)	kept: the source is defective
vendor copybooks (MQ CMQ*, LE CEEIGZCT)	kept: the estate doesn't ship them
COPY ORDPRICE / COPY SHPRATE (ORDMAIN, SHPINQ, SHPINQO)	a correct refusal by design. estate-crucible horror H-0053's key says "the estate holds no copybook ORDPRICE / SHPRATE, so no compiler finds it". Only a program source of that name exists. Engine copy resolution (#4468) reports the gap. The cause's side changes from translator to definitional

Ledger (fact_crosscheck.py update --corpus estate-crucible, then assign)

• 342 → 342 disagreements. translator-refuses-*: idms 1, national-text 4, source-defect 1, vendor-copybook 4 (unchanged counts), plus translator-copy-program-member 3.
• ZINSBER's entry "DECIMAL-POINT IS COMMA is not modelled" is gone. It is replaced by its next gap, "the name GEBÜHR holds a national letter", under translator-refuses-national-text, whose summary was refreshed (it now references det translator: a data name with a single-byte national letter (cp273 BETRÄGE / GEBÜHR) is refused: estate-crucible ZINSBER #4664).

Evidence

• New end-to-end program test_det_programs.py::DPCOMMA matches GnuCOBOL (cobc -std=ibm -fsign=EBCDIC) byte for byte in bytes, typed and groups modes. It runs ZINSBER's statements with ASCII names, plus zero suppression, a fixed and a floating sign, check protection, BLANK WHEN ZERO, de-editing, alphanumeric→zoned / packed / edited moves, NUMVAL and NUMVAL-C.
• test_det_source_4462.py: unit tests for the literal rewrite and for layout and statements, plus the 1.5 refusal.
• With EQUIVALENCE_E2E=1: test_det_programs + test_det_layout + test_det_translate gave 387 passed and 0 skipped. test_det_funcs gave 2 passed. test_det_sweep_fast, test_det_survey, test_det_port_check, test_det_parity, test_det_unmask and test_data_moves_decimal_comma_4355 all passed.
• pr_gates.py --fast: 5/5 pass. pr_gates.py --ratchets: 7/7 pass, none skipped (cics-spec, ports, port-surface, estate, fact-crosscheck, completeness, ground-truth). Engine scans are untouched, so no golden master moves.

Found on the way

• det runtime: numeric-edited item compared with a numeric literal numerically; GnuCOBOL compares it as text #4665: when an edited item is compared with a numeric literal, the det runtime compares numbers and GnuCOBOL compares text. This is independent of the clause, so the comparison was left out of DPCOMMA.

🤖 Generated with Claude Code

[github-actions]

🐦‍⬛ Muninn Security Scan

✅ No security issues found.

🐦‍⬛ Powered by Muninn · Skald Lab