jdelta

A structural diff for two JSON files. git diff and diff work on lines — so reformatting, reordered keys, or a changed indent drown the one value you actually care about in red-and-green noise. jdelta compares the data, not the text, and tells you exactly which values changed, addressed by path. Zero dependencies, no network.

pip install jdelta

$ jdelta before.json after.json

Added (2)
  + email         "alice@corp.com"
  + user.tags[2]  "d"

Removed (1)
  - legacyId      1001

Changed (3)
  ~ user.age      30 → 31
  ~ user.role     "viewer" → "admin"
  ~ user.tags[1]  "b" → "c"

+2  -1  ~3

This is the Python build. A behavior-equivalent Node build is on npm: npx jdelta (https://github.com/jjdoor/jdelta).

Why

You're reviewing a config change, an API-response snapshot, a tsconfig, a locale file — and the diff is unreadable because someone ran a formatter or the serializer reordered keys. You don't care that line 40 moved to line 12; you care that auth.required flipped to false. jdelta ignores key order and whitespace entirely and reports the actual value-level delta by path.

Usage

jdelta old.json new.json            # human-readable, grouped by added/removed/changed
jdelta old.json new.json --json     # machine-readable
jdelta old.json new.json --quiet    # just the +a -r ~c summary line
jdelta old.json new.json --exit-code  # exit 1 if they differ (CI gate)

Options

How it reads the diff

• Paths. Object keys use dot notation (user.profile.age); array elements use index notation (items[2].price). Keys that aren't plain identifiers — containing dots, spaces or hyphens, or starting with a digit — fall back to quoted brackets: ["order-id"], ["123"].
• Added / Removed / Changed. A key present on only one side is added or removed; a key on both with a different value is changed. A type change (number → string, object → array) is reported as a single changed entry tagged with the kinds — not as an add + remove.
• Arrays are compared by index. xs[2] is compared to xs[2]; a length change shows up as added/removed trailing elements. (Inserting at the front of an array therefore reads as "everything shifted" — index diffing is simple and predictable rather than guessing at moves.)
• Numbers are compared by value (1 and 1.0 are equal). Two caveats follow from how each runtime parses JSON numbers:
  • Float display. Floats are rendered by each runtime's native serializer, so an integral float can show as 1 (Node) or 1.0 (Python), and exotic floats (e.g. 1e-7) may differ in formatting. The detected change is the same.
  • Large integers. JavaScript has no bigint in JSON, so the Node build parses every number as an IEEE-754 double: integers beyond ±2^53 (snowflake IDs, int64 database keys) lose precision and can compare equal when they aren't — so the Node build may miss a change to such a value (and --exit-code won't fire). The Python build keeps integer precision exactly — prefer it for large-integer data.
• Long values are abbreviated in the human view (truncated with … past ~72 characters). --json output is never truncated.

--json shape

{
  "added":   [{ "path": "email", "value": "alice@corp.com" }],
  "removed": [{ "path": "legacyId", "value": 1001 }],
  "changed": [{ "path": "user.age", "from": 30, "to": 31 }],
  "summary": { "added": 1, "removed": 1, "changed": 1, "total": 3 }
}

Exit codes

By default jdelta is a viewer and exits 0; add --exit-code to make it a gate (the git diff --exit-code convention).

License

MIT