COBOL units on DSF's GML programs: SPECIAL-NAMES read as a unit, all-digit paragraph names missed, forge loses paragraphs

[squid-protocol]

Found while pinning navikt/DSF as a keyed corpus (#3491). Its 7 COBOL programs under src/GML/*.cobol expose three unit-extraction defects. The answer key and the blind reading agree against the engine or forge in each case:

1. Engine: SPECIAL-NAMES is a procedure unit. In FRMERK, FRREDIG, PLUKKFR and PLUKKFRN, the ENVIRONMENT DIVISION paragraph SPECIAL-NAMES. (followed by C01 IS PAGE1.) becomes a function unit, and is then also reported as dead.
2. Engine: all-digit paragraph names are missed. R001BYDL's 0000., 1000., 8000., 9000. and 9999. are legal COBOL paragraph names (a procedure name may be all digits) and form no unit.
3. Forge: paragraphs lost. In PLUKKFR/PLUKKFRN, the forge finds none of LESE, TEST-TRKNR, LES-TRK, FEIL-TRANS, SKRIV-UT or SLUTT; the header is written PROCEDURE DIVISION. with extra blanks. In FO04D1X1 it misses unit B.

All three are recorded in the ground-truth ledger under dsf_cobol_unit_extraction against this issue.

🤖 Generated with Claude Code