Shapefile DBF language drivers¶

A valid .cpg takes precedence over the language driver ID (LDID, byte 29 of the .dbf header). Without a usable .cpg, the following inventory describes RustGIS's LDID decoding. All numbers in the marker column are hexadecimal. These are language-driver identifiers, not the numeric code-page labels accepted in a .cpg.

Decoded markers¶

Code page	LDID markers
DOS CP437	`01`, `09`, `0B`, `0D`, `0F`, `11`, `15`, `18`, `19`, `1B`
DOS CP850	`02`, `0A`, `0E`, `10`, `12`, `14`, `16`, `1A`, `1D`, `25`, `37`
DOS CP852	`1F`, `22`, `23`, `40`, `64`, `87`
DOS CP861	`67`
DOS CP865	`08`, `17`, `66`
DOS CP866	`26`, `65`
Windows-874 (Thai)	`50`, `7C`
CP932 (Shift JIS)	`13`, `7B`
CP936 (GBK)	`4D`, `7A`
CP949 (Korean)	`4E`, `79`
CP950 (Big5)	`4F`, `78`
Windows-1250	`C8`
Windows-1251	`C9`
Windows-1252	`03`, `57`, `58`, `59`
Windows-1253	`CB`
Windows-1254	`CA`
Windows-1255	`7D`
Windows-1256	`7E`
Windows-1257	`CC`
Mac Roman (10000)	`04`
Mac Cyrillic (10007; see below)	`96`
UTF-8 (65001)	`F0`
No encoding metadata: UTF-8 default	`00`

57 deliberately means Windows-1252 in RustGIS. GDAL interprets it as ISO-8859-1; changing to that interpretation would change bytes 80–9F, such as the euro sign and an en dash, into control characters. The regression test uses 80 96 9F → €–Ÿ.

Mac Cyrillic is decoded with encoding_rs's x-mac-cyrillic (Mac Ukrainian) decoder. The shared Cyrillic letters are supported; this is not an exact mapping of every Mac Cyrillic character. The test uses 8A E8 E5 E2 → Киев, which both mappings share.

Recognized but not decoded¶

Code page	LDID markers
DOS CP863 (French Canadian)	`1C`, `6C`
DOS CP860 (Portuguese)	`24`
DOS CP895 (Czech, Kamenický)	`68`
DOS CP620 (Polish, Mazovia)	`69`
DOS CP737 (Greek)	`6A`, `86`
DOS CP857 (Turkish)	`6B`, `88`
Mac Central European (10029)	`97`

These markers emit a diagnostic naming the marker and code page, then use lossy UTF-8. Unknown markers also emit a diagnostic and use lossy UTF-8. Valid UTF-8 survives, malformed bytes become replacement characters, and the records remain available. An unusable .cpg adds its own diagnostic before falling back to the LDID. Supplying a valid .cpg bypasses the LDID and its unsupported-code-page diagnostic.

The existing transitive yore 1.4.0 package has CP737, CP857, CP860 and CP863 decoders, but RustGIS's current dbase adapter does not expose them. Making these available would require decoder integration and a coordinated direct dependency declaration; their presence in the lockfile does not make the corresponding LDIDs supported today. No local conversion tables or system programs are needed for the supported decoders above.

References and regression coverage¶

Marker inventory: GDAL 3.11.4's GetEncodingFromLDIDNumber, plus 09, 7D, 7E and F0 from dbase 0.8.0's CodePageMark.
GDAL's Shapefile encoding documentation describes .cpg precedence and its different 57 interpretation.
Character mapping references: Unicode's Microsoft code-page mappings and Apple mappings.
yore's CodePage implementations list the available legacy decoders and lossy decoding API.

crates/data/tests/shapefile_encoding.rs specifies fixed bytes and Unicode expectations independently of RustGIS's mapping. Its matrix covers every marker above, names and values, ZIP and loose-file imports, a genuinely sampled streamed read, a complete read, and ordered chunk traversal. It also checks .cpg precedence, fallback diagnostics, and malformed multibyte text without discarding records. Test files are generated locally; the tests do not download data or call external converters.