Compare two Arrow or Feather files — privately, in your browser
To compare two Apache Arrow files, drop both into Diffesq: it reads each file's IPC schema on your device and shows a three-layer diff — schema changes, per-column statistics, and key-matched row differences — without uploading either file. Feather v2 is the Arrow IPC file format, so .arrow and .feather both land here.
Files are identified by their magic bytes rather than their extension, so an Arrow file saved with an unusual suffix is still recognised — and a file that only looks like one is rejected with a specific reason instead of a wrong answer.
How it works
- Open the Diffesq workbench and drop an
.arrowor.featherfile on each side. - Diffesq reads the IPC schema and column vectors on your device, then compares schema, statistics, and key-matched rows.
- Review the three layers, adjust the key, numeric tolerance, or ignored columns, and export the result — all locally.
What the comparison covers
- Schema diff from the file's declared Arrow types: columns added, removed, retyped, reordered; nullability changes.
- Dictionary-encoded columns compare by their values — the encoding is a storage detail, so a dictionary-encoded string column and a plain one are not reported as a type change.
- Profile diff over every row: null rates, distinct counts, min/max, mean and standard deviation, with changed metrics flagged.
- Key-matched row diff with composite keys, absolute/relative numeric tolerance, ignore-case, trim, and ignore-column options.
- Cross-format comparison against CSV, Parquet, SQLite, and JSON Lines files.
- Feather v1 is detected and refused with instructions for re-saving as v2, rather than being misread.
Free vs Pro
Structured-file comparisons run in full on your device on the free tier, with about a third of the results shown as a preview; Pro unlocks the complete results and clean, unwatermarked exports (HTML, Markdown, JSON, PDF).
See Pricing for plans — $2.97/month, $27.99/year, or $49 once.
Privacy
Every comparison runs locally in your browser. Your files are never uploaded — there is no server that receives them, and the page keeps working with Wi-Fi turned off. Details in the Privacy Policy.
Frequently asked questions
Does it support Feather files?
Feather v2 files compare exactly like any other Arrow file, because Feather v2 is the Arrow IPC format. Feather v1 is a different, older container: Diffesq detects it and tells you to re-save as v2 instead of guessing.
How large a file can it handle?
Arrow's reader materialises the whole table, so the practical limit is your browser's memory. Past roughly 2 million cells the row diff samples and says so; past ~20 million it steps aside while the schema and column profiles still cover the full table.
Do I need an account?
No. Comparisons run without any signup. Pro features unlock with a license key — still no account or login.
Does it work offline?
Yes — once the page has loaded, parsing, comparing, and rendering all happen on your device. Turn off Wi-Fi and it still works.