fastdbf 0.4.1

A Python package for reading and writing DBF files, powered by a Rust core

fastdbf

fastdbf is a Rust-based reimplementation of the core ideas from ethanfurman/dbf, exposed as a Python package through PyO3.

The current focus is a practical Python-first package for reading and writing .dbf files, including Visual FoxPro-style nullable fields.

Status

This project is in an early but usable state.

Implemented today:

• read .dbf files from Python
• inspect field metadata and field types
• create new tables from DBF-style field specifications
• append rows from dictionaries or tuples
• write tables back to disk
• Visual FoxPro null-flag support for nullable fields
• a Rust core with test coverage for read/write roundtrips

Not implemented yet:

• full memo file support (.dbt / .fpt)
• full original dbf compatibility surface
• record objects with attribute access like record.name
• indexing, relations, PQL, and most advanced helpers from the original project

Installation

Create or sync the development environment with uv:

uv sync

Install into the current environment with uv:

uv pip install .

Editable install:

uv pip install -e .

Run tests:

uv run pytest

Build a wheel:

uv build

Quick Start

Read an existing DBF file:

import fastdbf

with fastdbf.Table("people.dbf").open("r") as table:
    print(table.kind)
    print(table.field_names)
    print(table.record_count)
    print(table.row(0))

    for field in table.fields():
        print(field["name"], field["type_code"], field["nullable"])

    for row in table:
        print(row)

Create and write a new DBF file:

import fastdbf

specs = "name C(25) null; age N(3,0) null; birth D null; active L null"
with fastdbf.Table("people.dbf", specs, dbf_type="vfp") as table:
    table.append({
        "NAME": "Spunky",
        "AGE": 23,
        "BIRTH": "1989-07-23",
        "ACTIVE": True,
    })

    table.append({
        "NAME": None,
        "AGE": None,
        "BIRTH": None,
        "ACTIVE": None,
    })

Field Types

Currently supported field types:

• C Character
• D Date
• L Logical
• N Numeric
• F Float
• I Integer
• B Double
• T / @ DateTime
• Y Currency
• M / G / P as reference values

Nullable fields are supported through null or nullable modifiers in the field specification:

"name C(25) null; amount N(10,2) nullable; created T null"

Nullable fields should be used with dbf_type="vfp" for Visual FoxPro-compatible null flags.

pandas Example

import pandas as pd
import fastdbf

with fastdbf.Table("input.dbf").open("r") as table:
    df = pd.DataFrame(list(table))
    df["NAME"] = df["NAME"].str.upper()

    field_specs = []
    for field in table.fields():
        code = field["type_code"]
        nullable = " null" if field["nullable"] else ""
        if code == "C":
            field_specs.append(f'{field["name"]} C({field["length"]}){nullable}')
        elif code in ("N", "F"):
            field_specs.append(
                f'{field["name"]} {code}({field["length"]},{field["decimals"]}){nullable}'
            )
        else:
            field_specs.append(f'{field["name"]} {code}{nullable}')

dbf_type = "vfp" if any(f["nullable"] for f in table.fields()) else "db3"
specs = "; ".join(field_specs)

with fastdbf.Table("output.dbf", specs, dbf_type=dbf_type) as out:
    for row in df.to_dict(orient="records"):
        out.append(row)

Performance & Columnar I/O (Arrow)

For maximum performance, especially with large datasets, fastdbf provides columnar read/write interfaces that avoid the high overhead of Python object allocation.

1. Apache Arrow Interface (Zero-Copy) — Fastest

Leverages the Arrow PyCapsule Interface to exchange data directly between Rust and Pandas / Polars / PyArrow without copying.

Read into Pandas via Arrow:

import fastdbf
import pyarrow as pa

with fastdbf.Table("data.dbf").open("r") as table:
    # Arrow Batch -> Pandas DataFrame
    df = pa.Table.from_batches([pa.record_batch(table.to_arrow())]).to_pandas()

Write from Pandas via Arrow:

import fastdbf
import pyarrow as pa

# Create Arrow batch from DataFrame
batch = pa.RecordBatch.from_pandas(df)

with fastdbf.Table("output.dbf", field_specs="NAME C(20); AGE N(10,2)") as table:
    table.extend_arrow(batch)

2. Columnar Interface (to_columns, extend_columns)

Reads/writes data as a dictionary of lists (one list per column). Faster than row-by-row processing but still bound by GIL limits.

Bulk Columnar Read:

import pandas as pd
import fastdbf

with fastdbf.Table("data.dbf").open("r") as table:
    cols = table.to_columns()
    df = pd.DataFrame(cols)

Bulk Columnar Write:

import pandas as pd
import fastdbf

# Assuming df ist das Pandas DataFrame
# Optionale interne Meta-Spalten wie '_deleted' entfernen
clean_data = {col: df[col].tolist() for col in df.columns if col != "_deleted"}

with fastdbf.Table("output.dbf", field_specs="NAME C(20); AGE N(10,2)") as table:
    table.extend_columns(clean_data)

---

Overview: Read/Write Methods compared

Method	Implementation	Pros
Row-by-Row	table.row(), table.append()	Easiest to use
Bulk Columnar	to_columns(), extend_columns()	No heavy dependencies
Zero-Copy Arrow	to_arrow(), extend_arrow()	Direct memory exchange

Documentation

Full Python API documentation:

• docs/PYTHON_API.md

Changelog:

• CHANGELOG.md

Rust Example

use fastdbf::{Date, Table, Value};

fn main() -> Result<(), Box<dyn std::error::Error>> {
    let mut table = Table::new("name C(25); age N(3,0); birth D; qualified L")?;

    let mut record = table.new_record();
    record.insert(table.fields(), "name", Value::Character("Spunky".into()))?;
    record.insert(table.fields(), "age", Value::Numeric(23.0))?;
    record.insert(table.fields(), "birth", Value::Date(Some(Date::new(1989, 7, 23))))?;
    record.insert(table.fields(), "qualified", Value::Logical(Some(true)))?;
    table.push_record(record)?;

    table.write_to_path("example.dbf")?;
    Ok(())
}

License

This project is licensed under the Apache License 2.0.