kobold-arrow

Map COBOL fixed-record layouts to equivalent Apache Arrow schemas.

Given a declared COBOL record layout (fields tagged with their physical COBOL encoding -- PIC X, DISPLAY numeric, COMP-3 packed decimal, COMP/COMP-5 binary, COMP-1/COMP-2 float), kobold-arrow produces the equivalent Arrow schema: each field's name, Arrow logical type, and nullability. This lets a legacy fixed-record dataset be landed into the Arrow / columnar world with a faithful, lossless type mapping.

The schema is emitted as data -- a serializable description whose type names mirror Arrow's own logical vocabulary (Utf8, Int32, Int64, Decimal128, Float32, Float64, ...). The crate deliberately does not link the heavyweight arrow crate; it depends only on serde/serde_json and stays light.

Part of KOBOLD -- a forensic archaeology and evidence system for legacy COBOL estates. Independently-authored tooling; contains no GnuCOBOL source.

Mapping rules

• PIC X(n) -> Utf8
• DISPLAY / COMP-3 numeric with scale > 0 -> Decimal128(precision = digits, scale) (kept exact)
• Integer DISPLAY / COMP-3 (scale == 0): digits <= 9 -> Int32; 10..=18 -> Int64; > 18 -> Decimal128(precision = digits, scale = 0)
• Binary (COMP / COMP-5): bytes <= 4 -> Int32, otherwise Int64
• COMP-1 -> Float32; COMP-2 -> Float64

COBOL fixed records have no SQL-style NULL, so mapped fields are non-nullable.

CLI

kobold-arrow map <layout.json> [--pretty]

Reads a CobolLayout JSON and prints the equivalent ArrowSchema JSON. Example layout:

{"name":"CUSTOMER-RECORD","fields":[
  {"name":"cust-name","encoding":{"alphanumeric":{"len":30}}},
  {"name":"cust-id","encoding":{"display-numeric":{"digits":5,"scale":0,"signed":false}}},
  {"name":"balance","encoding":{"packed":{"digits":9,"scale":2,"signed":true}}}]}

Library

use kobold_arrow::{CobolField, CobolLayout, CobolEncoding, map_layout, to_json};

let layout = CobolLayout {
    name: "REC".into(),
    fields: vec![CobolField {
        name: "amount".into(),
        encoding: CobolEncoding::Packed { digits: 7, scale: 2, signed: true },
    }],
};
let schema = map_layout(&layout);          // -> ArrowSchema
let json = to_json(&schema).unwrap();       // Decimal128(precision: 7, scale: 2)

Future work

Building actual Arrow RecordBatch columns -- decoding the fixed-record bytes into Arrow arrays on top of the real arrow crate -- is a documented future feature, kept behind an optional dependency so the schema mapper stays light by default.

License

Apache-2.0 (see LICENSE).