csv-conflict

Fix a CSV file that git left full of <<<<<<< / ======= / >>>>>>> markers. The markers may sit in the header or inside a quoted multiline field, so the file is not valid CSV. csv-conflict rebuilds the local and remote versions first, parses each as CSV, then either keeps one side or merges cell by cell. Best effort, no common ancestor needed. Python 3.9+, stdlib only.

pip install csv-conflict

csv-conflict data.csv            # merge: differing cells hold both values
csv-conflict data.csv --local    # keep our side
csv-conflict data.csv --remote   # keep their side
csv-conflict data.csv --key id   # pair rows by id instead of by position

A differing cell becomes one quoted field:

<<< HEAD
local value
===
remote value
>>> feature

Behaviour

• Merge (default): columns matched by name (remote order, then local-only columns); rows paired by position, or by --key COL (repeatable; unique, nonempty; rows without a partner are kept as-is). Cells compared literally; empty vs nonempty counts as a difference.
• --local / --remote: only the chosen side must be valid CSV; the other may be broken.
• --no-header: headerless files; positional merge, same column count required. With --key use 1-based numbers.
• --separator ';', --encoding windows-1252 (strict, never guesses), LF/CRLF taken from the first line ending.
• Input is replaced atomically (use -o OUT to keep it). Empty lines are skipped.
• Warns on stderr with the number of cells that still hold conflicts (--fail-on-conflicts: exit 1).
• Exit 2 and file untouched on: malformed/nested/unclosed markers, diff3 ancestor sections, invalid header (empty/duplicate names), ragged rows, bad quoting, bad key, bad encoding/separator, unreadable/unwritable file.

Testing

25 pytest tests, pass on Python 3.9 and 3.14. Clean-venv install of sdist and wheel checked.

Donate

If it saved you time: https://paypal.me/RafalBaranowski

MIT license.