What is Paradox Database

Points that matter for recovery:

• The data is in .db and .mb. You can export a table without its indexes, but not without its .mb file if it has memo or BLOB fields: long text is cut off at the part stored inline in the record.
• .db is not unique to Paradox. SQLite databases, Windows Thumbs.db and several other programs use the same extension. dBASE files are .dbf, a different format.
• The table format changed between Paradox 3, 4, 5 and 7. Older readers may refuse newer tables.
• Text is stored in a DOS or Windows code page, not Unicode. Accented characters come out wrong unless you convert from the right one.
• A table can be password-protected, in which case its contents are encrypted and you need the password.

Status today

• Paradox for Windows is sold only as part of WordPerfect Office Professional. The 2021 edition includes it. It is a 32-bit Windows application.
• The BDE is deprecated and 32-bit only. Embarcadero has recommended against using it for many years.
• Microsoft Access used to import and link Paradox tables. Microsoft dropped that support; current versions of Access cannot open them.
• Windows still carries the old 32-bit Jet "Microsoft Paradox Driver (*.db)" ODBC driver, visible only in the 32-bit ODBC administrator (C:\Windows\SysWOW64\odbcad32.exe). It handles older table levels, is unreliable with Paradox 7 and later tables unless the BDE is installed, and cannot be used from 64-bit applications.

So there is no supported, modern way to keep using Paradox tables as a live database. The realistic goal is a one-time migration.

Before you start: copy the folder

Work on a copy, never on the originals:

cp -a /mnt/oldserver/accounts /home/me/paradox-copy

Copy the entire directory, not just the .db files, so that .mb and .px files travel with their tables. If the application is still in use, copy while everyone is out of it; a table copied in the middle of a write can be inconsistent.

Option 1: pxview on the command line

pxview is a small open-source program built on the pxlib library. It reads .db, .px and .mb files without Paradox or the BDE. It is plain C and builds on Linux and other Unix-like systems (on Windows, use WSL). Debian and Ubuntu package the pxlib library; pxview itself is usually built from the source on the project site.

Inspect a table first:

pxview --info customers.db
pxview --schema customers.db

--info prints the table header, including the file version, the number of records and the field list; add --verbose for more. If this fails, the file is either not a Paradox table or is damaged.

Export to CSV:

pxview --csv --output-file=customers.csv customers.db

Export as SQL, including memo data and converting text to UTF-8:

pxview --sql --tablename=customers --blobfile=customers.mb \
       --recode=UTF-8 --output-file=customers.sql customers.db

The SQL output is a CREATE TABLE followed by INSERT statements. Add --use-copy to get PostgreSQL COPY format, or --skip-schema to write only the data. --recode works only if pxlib was compiled with iconv or recode support; pxview --help reports which.

Option 2: Python with pypxlib

pypxlib wraps the same library for Python and is the easier route when you have many tables:

pip install pypxlib

import csv
from pypxlib import Table

with Table("customers.db", encoding="cp1252") as table, \
     open("customers.csv", "w", newline="", encoding="utf-8") as out:
    names = list(table.fields)
    writer = csv.writer(out)
    writer.writerow(names)
    for row in table:
        writer.writerow([row[name] for name in names])

print("done")

Table looks for a matching .mb file next to the .db automatically. The encoding argument is the code page of the stored text; it defaults to cp850 (DOS Western European). Try cp1252 for data entered in Windows applications and cp437 for US DOS data. Wrap the body in a loop over pathlib.Path(".").glob("*.db") to convert a whole directory.

The package was last released in 2020, so test it on your Python version before relying on it. pxview is the fallback.

Option 3: Paradox itself

If you have a licensed copy of Paradox for Windows, open the table and use File > Export to write delimited text. This is the only route that can also open the forms, reports and ObjectPAL code, which you will want to read to understand what the application did, even though none of it can be converted automatically.

Loading into a current database

Create the table with proper types and load the CSV. In PostgreSQL:

CREATE TABLE customers (
    customer_id integer PRIMARY KEY,
    name        varchar(80) NOT NULL,
    balance     numeric(15,2),
    created     date,
    notes       text
);

psql -d legacy -c "\copy customers FROM 'customers.csv' WITH (FORMAT csv, HEADER true)"

How Paradox field types map:

Verify the migration

• Compare row counts: the record count from pxview --info against SELECT count(*).
• Compare a sum on each numeric column and the minimum and maximum of each date column.
• Search the loaded text for the replacement character � or sequences like é, which mean the code page was wrong.
• Check that the longest memo values are complete, which confirms the .mb file was read.

Common problems

• "Not a Paradox file" or garbage output. The file is another program's .db. Check with pxview --info.
• Memo fields truncated or empty. The .mb file is missing, renamed or in a different directory.
• Keys and rules did not come across. Primary keys live in .px, and validity checks and referential integrity in .val. Exports carry data only. Recreate keys, constraints and indexes in the new database by hand.
• Deleted records reappear. Do not pass --output-deleted to pxview unless you are deliberately recovering deleted rows.
• Table reports as corrupt. File-shared Paradox tables were prone to index damage after crashes. The data in .db is often still readable when the .px is not, which is another reason to export from .db directly.

Further reading

• pxlib: library to read and write Paradox files
• pxview documentation
• pypxlib on PyPI
• PostgreSQL documentation: COPY

How do I insert a row only if it doesn't exist, without race conditions?

Select-then-insert breaks under concurrency. Use a unique constraint plus one atomic statement: ON CONFLICT, ON DUPLICATE KEY UPDATE, or UPDLOCK/SERIALIZABLE in SQL Server.

#concurrency#schema-design#transactions