Repository navigation
Zoned overpunch table is the US code page's: German / Nordic / French / Italian data throws on every value ending in 0 #3826
Description
What: {, } and A–R as sign characters hold only for CP037/500/1047/284/285. Under a national code page, bytes 0xC0 and 0xD0 are other characters (probe):
| Code page | 0xC0 (positive zero) | 0xD0 (negative zero) |
|---|---|---|
| CP273 (German) | ä |
ü |
| CP277 (Danish/Norwegian) | æ |
å |
| CP278 (Finnish/Swedish) | ä |
å |
| CP280 (Italian) | à |
è |
| CP297 (French) | é |
è |
Where:
cobol_to_java_repository_forge.py:139(POSITIVE = "{ABCDEFGHI").tests/tools/equivalence_common.py_OVERPUNCH.equivalence_inputs.encode_field.
Impact: The Java codec throws invalid zoned sign, and the harness decodes <invalid>.
Test: A German S9(5)V99 value of +100.00 whose last byte shows as ä. Expected: 100.00 on both sides.
Fix: Derive the sign table from the configured code page (0xC0–C9 and 0xD0–D9 decoded through it), or decode zoned fields from bytes. Shares the code-page configuration with #3816.
Found by a cultural-assumptions audit (2026-09-26): confirmed in code or by probe (JDK 17 locales, IBM code-page decoding, the harness's GnuCOBOL image) unless marked plausible.
🤖 Generated with Claude Code
Activity
#3826: the zoned sign table follows the data's code page (#3842)
#3816 (part 1): the Western European EBCDIC code pages, and raw EBCDI…