Python Paradox database reader

paradox.py

# coding: utf-8

"""
Converts Paradox databases to Python objects or CSV.

You don't need any dependency (except Python) to make this module work.
This module is incomplete but reads most Paradox `.DB` files.
If this module is not fast or complete enough for you, consider using pxview.
"""

from __future__ import division
import csv
from datetime import date, datetime, timedelta
from struct import unpack

try:
    force_text = unicode
except NameError:
    force_text = str


__author__ = 'Bertrand Bordage'
__copyright__ = 'Copyright © 2014 Bertrand Bordage'
__license__ = 'MIT'


field_types = {
    1: 'A',   # string
    2: 'D',   # date
    3: 'S',   # short integer
    4: 'I',   # integer
    5: '$',   # money
    6: 'N',   # number
    9: 'L',   # logical (boolean)
    12: 'M',  # memo
    20: 'T',  # time
    21: '@',  # timestamp
    22: '+',  # autoincrement
}


def unpack_signed(fmt, s, complement):
    v = unpack(fmt, s)[0]
    if v == 0:
        return
    if v < 0:
        return v + complement
    return v - complement


I_COMPLEMENT = 1 << (4*8 - 1)
H_COMPLEMENT = 1 << (2*8 - 1)
B_COMPLEMENT = 1 << (1*8 - 1)


def unpack_i(s):
    return unpack_signed('>i', s, I_COMPLEMENT)


def unpack_h(s):
    return unpack_signed('>h', s, H_COMPLEMENT)


def unpack_b(s):
    return bool(unpack_signed('>b', s, B_COMPLEMENT))


def unpack_d(s):
    return -unpack('>d', s)[0]


def to_date(s):
    ordinal = unpack_i(s)
    if ordinal is None:
        return
    try:
        return date.fromordinal(ordinal)
    except ValueError:
        return


def to_time(s):
    seconds = unpack_i(s)
    if seconds is None:
        return
    seconds //= 1000
    return (datetime.min + timedelta(seconds=seconds)).time()


SECONDS_PER_DAY = 60*60*24


def to_datetime(s):
    seconds = int(unpack_d(s) // 1000)
    ordinal, seconds = divmod(seconds, SECONDS_PER_DAY)
    try:
        d = datetime.fromordinal(ordinal)
        return d + timedelta(seconds=seconds)
    except ValueError:
        return


def decode_data(t, v, input_encoding):
    if t == 'A':
        return v.strip('\x00').decode(input_encoding)
    if t == 'S':
        return unpack_h(v)
    if t == 'L':
        return unpack_b(v)
    if t == 'I':
        return unpack_i(v)
    if t == 'D':
        return to_date(v)
    if t == 'T':
        return to_time(v)
    if t == '@':
        return to_datetime(v)
    if t in ('N', '$'):
        return unpack_d(v)
    if t == 'M':
        return ''  # TODO: handle memos
    if t == '+':
        return unpack_i(v)
    return v


def read(filename, dict_output=False, input_encoding='iso-8859-1'):
    """
    Converts a Paradox .DB file to Python objects.
    """
    db_file = open(filename)
    header = db_file.read(4)
    header_size = unpack('>H', header[2:4])[0]
    db_file.seek(0)
    header = db_file.read(header_size*1024//4)
    block_size = ord(header[5])
    n_fields = ord(header[33])
    parts = header.split(filename.split('/')[-1])
    headers, data = parts
    defs = headers[120:]
    columns = data.strip('\x00').split('\x00')[:n_fields]
    fields = []
    for i, column in enumerate(columns):
        t = field_types[ord(defs[i*2])]
        length = ord(defs[i*2+1])
        fields.append((column, t, length))
    record_size = sum([length for _, _, length in fields])
    blank_record = '\x00' * record_size
    # FIXME: I don't know why, but this is not working without dividing
    #        header size by 4.
    data_offset = header_size*1024//4
    rows = []
    block_index = 0
    previous_record = None
    while True:
        offset = data_offset + block_index * block_size * 1024
        db_file.seek(offset)
        block_header = db_file.read(6)
        offset += 6

        next_block_offset = data_offset + (block_index+1) * block_size * 1024
        while offset + record_size < next_block_offset:
            db_file.seek(offset)
            record = db_file.read(record_size)
            if len(record) < record_size or record == blank_record:
                break
            if record == previous_record:
                offset += record_size
                continue
            previous_record = record
            row = []
            for k, t, size in fields:
                db_file.seek(offset)
                v = db_file.read(size)
                offset += size
                v = decode_data(t, v, input_encoding)
                row.append((k, v))
            rows.append(row)
        
        is_last_block = unpack('>H', block_header[0:2])[0] == 0
        if is_last_block:
            break
        block_index += 1
    if dict_output:
        rows = map(dict, rows)
    return fields, rows


def to_csv(filename, output_filename=None):
    """
    Converts a Paradox .DB to a CSV file.
    """
    fields, data = read(filename)
    fieldnames = [c for c, _, _ in fields]
    if output_filename is None:
        output_filename = filename + '.csv'
    d = csv.DictWriter(open(output_filename, 'w'), fieldnames)
    d.writeheader()
    d.writerows([dict([(k, force_text(v).encode('utf-8')) for k, v in row])
                 for row in data])

Hi,
Thanks for this useful code. But how about if the paradox value a signed double like -111,00 (whereas 111,00 is fine). I didn't manage to add a fix.
Thank you

According to the specification, the field type for floats is N. How I dealt with this data type is written line 69. I didn’t use unpack_signed in this case because of some inconsistency with the floating point standard. If I remember well, the sign was the opposite of what it should be or something like this. I ended up doing a bad workaround, but since there was no negative float in the only Paradox DB I saw in my life, I let it like this.

So you have to find a solution by yourself, and it’s line 69 that everything happens.

The solution is explained here : http://stackoverflow.com/questions/27360892/python2-7-difficulty-to-convert-a-binary-signed-double-to-ascii/27416776#27416776. Thanks to you.

Thanks for this solution, unfortunately it didn't work for me. I ended up implementing complete Python bindings to the pxlib library: https://github.com/mherrmann/pypxlib

Install with

pip install pypxlib

Then you can for instance do:

from pypxlib import Table
table = Table('my-paradox.db')
try:
    for row in table:
        print(row)
finally:
    table.close()

i think the line 131 should change into db_file = open(filename, 'rb'). read as binary file to solve the problem that can not read specific length. if not, the contents of some line will lost (break, at line 165)

Thanks for the gist. I had to extract data from paradox database this helped a lot. I applied the previous fix suggested by @zicowarn and that worked in my case.

Hey guys! Inspired by @mherrmann:
https://github.com/javifr/paradoxdbsplitter

@BertrandBordage, can you help me? My Language Driver is set to ANSII850, I try to change the line 127 of your code to
def read(filename, dict_output=False, input_encoding='ansii850'):
But it does not work. You know what can be wrong?

edit in my case I'm geting the error:

@danielpcamara It’s a very old script I wrote probably for Python 2, which opened in binary mode by default if I remember correctly. pass 'rb' to open and that should do it.

Thanks @BertrandBordage, this still didn't work for me, but I was able to find out another way to extract my DataBase in CSV using the app Paradox Data Editor.

how to write rows of table in pandas dataframe?

@BertrandBordage any idea why I get the following error
"TypeError: ord() expected string of length 1, but int found"
for line 136? I already added the "rb" as suggested above.

Thanks for this solution, unfortunately it didn't work for me. I ended up implementing complete Python bindings to the pxlib library: https://github.com/mherrmann/pypxlib

Install with

pip install pypxlib

Then you can for instance do:

from pypxlib import Table
table = Table('my-paradox.db')
try:
    for row in table:
        print(row)
finally:
    table.close()

This is what worked for me