dbflib 1.1

Create, mutate, and index Visual FoxPro DBF tables (DBF + FPT memo + CDX index), byte-compatible with files written by Visual FoxPro Advanced (32-bit)

dbflib

Create, mutate, and index Visual FoxPro DBF tables — the .dbf table, its .fpt memo file, and its .cdx compound index — from pure Python, with no dependencies outside the standard library.

The correctness bar is not "a reader can parse it": files produced by dbflib are byte-compatible with files written by Visual FoxPro Advanced (VFPA, 32-bit) — the target environment — verified against a corpus of several hundred VFPA-created production tables. That includes the quirky parts — VFP's numeric overflow "squeeze", cp1252 with undefined code points preserved, x87 extended-precision arithmetic in index key expressions, machine collation, and deleted records remaining indexed. The files should also be compatible with classic Visual FoxPro, but 32-bit VFPA is what the output is verified against.

Installation

pip install dbflib

Requires Python 3.9+. No runtime dependencies.

Quick start

from dbflib import create_dbf, insert_rows, delete_rows, add_dbf_column

create_dbf('out/items.dbf',
    columns=[('name', 'C (30)'), ('qty', 'N (8, 2)'), ('note', 'M')],
    rows=[{'name': 'widget', 'qty': 1.5}],
    indexes={'BYNAME': 'UPPER(name)'})

insert_rows('out/items.dbf', [{'name': 'gadget'}])
delete_rows('out/items.dbf', lambda record: record['QTY'] == 0)
add_dbf_column('out/items.dbf', 'price', 'N (10, 2)', fill=0)

Every mutating call automatically rebuilds the table's structural .cdx from the index expressions stored inside it, so the index never goes stale.

Column types

Columns are (name, type) pairs using compact schema notation:

Notation	VFP type
C (n)	Character, width n
N (n, d)	Numeric, width n, d decimals
F (n, d)	Float (stored like Numeric)
M	Memo (creates the .fpt)
L	Logical
I	Integer (binary int32)
D	Date
T	DateTime
Y	Currency
B	Double

Indexes

indexes maps CDX tag names to a FoxPro key expression string, or to a dict with key and optional filter / unique / descending:

from dbflib import build_cdx

build_cdx('out/items.dbf', {
    'BYNAME':  'UPPER(name)',
    'POSONLY': {'key': 'UPPER(name)', 'filter': 'qty > 0'},
    'UNAME':   {'key': 'UPPER(name)', 'unique': True},
})

Key and filter expressions are real FoxPro expressions, compiled to Python and evaluated with VFPA's exact semantics so the resulting key bytes match what VFPA would write.

The expression compiler supports a subset of FoxPro syntax, not the full language. Supported: field references, string/numeric/date/logical literals, comparison and logical operators (.AND./.OR./.NOT. and symbol forms), string concatenation, numeric arithmetic, and these functions:

UPPER LOWER TRIM RTRIM LTRIM ALLTRIM SPACE PADR PADL SUBSTR LEFT RIGHT STR STRTRAN CHRTRAN CHR LEN AT VAL ABS MOD MAX MIN MLINE RECNO DELETED DTOS DTOC CTOD CTOT DTOT TTOD TTOC TIME YEAR MONTH DAY HOUR MINUTE SEC EMPTY IIF ICASE EVL

Anything outside this subset raises FoxProExpressionError when the index is built — nothing is silently miscomputed. Support for more of the language can be added as needed; it just isn't feature-complete today. One structural constraint carried over from how VFP sizes index keys: every character-typed expression must have a statically known width, so e.g. STRTRAN(...) must be wrapped in PADR(...) to be usable as a key.

Reading

from dbflib import open_dbf_table, read_cdx, read_cdx_definitions

table = open_dbf_table('out/items.dbf')
try:
    rows = table.read_rows()        # list of dicts of Python values
finally:
    table.close()

tags, order = read_cdx('out/items.cdx')            # tag properties + ordered (key, recno) entries
definitions = read_cdx_definitions('out/items.cdx')  # recover index definitions from the CDX itself

API summary

Function	Purpose
create_dbf(path, columns, rows=(), indexes=None)	Create a table (replacing existing files), optionally with rows and a structural CDX
insert_rows(path, rows, truncate=False)	Append row dicts; truncate=True empties the table first
delete_rows(path, where)	Soft-delete by 1-based record numbers or a row predicate
add_dbf_column / drop_dbf_column / alter_dbf_column	Column operations; rebuild the table in place, preserving deleted flags and the CDX
build_cdx(path, indexes)	Build the structural compound index
build_cdx_per_tag(path, definitions)	Build one single-tag CDX file per index
refresh_cdx(path)	Rebuild the structural CDX from its own stored expressions
read_cdx / read_cdx_definitions	Read an existing CDX back into tags/entries or index definitions
open_dbf_table(path)	Open a table read-write (VfpTable)
DbfTableData(path)	Low-level binary reader used for fast index builds

Fidelity notes

Where Visual FoxPro's behavior conflicts with Python convention, dbflib sides with VFP (as observed in 32-bit VFPA):

• Oversized numeric values silently "squeeze" (drop decimals, then scientific notation, then asterisks) instead of raising.
• **** numeric overflow reads back as +inf and re-encodes as asterisks.
• Deleted records stay in index tags, exactly as VFP keeps them.
• Text is cp1252 with surrogateescape, so undefined bytes round-trip.
• Index keys use VFP machine collation and its exact numeric/date/datetime key encodings.

License

MIT