logicdiff

A whitespace- and reflow-blind diff. A pull request reindents a file and rewraps a few long lines, and now git diff shows 80 changed lines — but did anything actually change? logicdiff answers that: it folds away pure formatting (respacing and line reflow) and shows only the logical changes.

logicdiff old.js new.js
#   only formatting differs - no logical change (a line diff would show 80 changed lines)

logicdiff a.js b.js
#   --- a.js
#   +++ b.js
#   -42:   const total = price * qty;
#   +51:   const total = price + qty;
#
#   1 token removed, 1 added across 2 logical lines (78 lines folded as reflow/whitespace)

Exit 0 when the change is formatting-only (or identical), 1 when there's a real logical change — so CI can ask "is this PR just a reformat?" Zero dependencies, language-agnostic, also on npm (npx logicdiff) — the two builds produce byte-for-byte identical output.

Why not git diff -w?

git diff -w (ignore-all-space) folds respacing — but it is still line-anchored, so it cannot fold reflow. Re-wrap a function signature across three lines and git diff -w still shows 1 removed + 3 added, even though not a single token changed. That exact gap is GitHub discussion #20610 ("Ignore Format Changes in Diff"), open and unanswered for years.

difftastic solves it beautifully with per-language tree-sitter parsing — but it's a multi-megabyte binary, needs a grammar for each language (config/log/DSL files fall back to text), and it's a display tool with no "is this formatting-only?" exit code.

logicdiff is the lightweight middle ground: zero-config, zero-dependency, language-agnostic (works on any text — code, YAML, logs, DSLs), folds both whitespace and reflow, and gives a one-shot CLI answer plus a CI exit code.

How it works

It tokenizes each file into a sequence of tokens — a token is a run of [A-Za-z0-9_] or a single punctuation character, and whitespace is dropped. So a+b, a + b, and a +\n b all become the same token stream [a, + , b]: respacing and line breaks become invisible. It then runs the canonical Myers diff on the token streams. If the streams are equal, the change is formatting-only. If not, the changed tokens are mapped back to their line numbers and shown.

Because it has no language parser, whitespace inside string literals is also ignored — x = "a b" and x = "a b" are "formatting only", exactly like git diff -w. That's a deliberate, documented limitation, not a bug.

Usage

logicdiff old new            # human diff (or "only formatting differs")
logicdiff old new --stat     # just the counts, machine-friendly key=value
logicdiff old new --json     # structured output (byte-identical both builds)
logicdiff old new -q         # no output, exit code only (the CI gate)
cat new | logicdiff old -     # - reads stdin

--color=auto|always|never, --max-tokens N (bail over N tokens, default 2,000,000). Two wildly dissimilar inputs (a huge edit distance) also bail with exit 2 instead of risking the heap — logicdiff is for spotting a real change inside a reformat, not for diffing unrelated files.

Exit codes: 0 identical or formatting-only · 1 logical changes · 2 error.

# CI: warn when a PR is more than a reformat
- run: logicdiff "$BASE" "$HEAD" -q || echo "::warning::real code change, review carefully"

Install

pip install logicdiff # or pipx run logicdiff
npm i -g logicdiff    # Node build, identical behaviour

Python ≥ 3.8 or Node ≥ 18. No dependencies.

License

MIT