Text Diff

Details and how to verify this yourself: Security & privacy.

The tool sits in a window that works like a desktop app: menus on the top row and the full-screen icon at its right (Esc leaves full screen). The set-up options described below are in Settings → Options…, an overlay, and the Settings menu also lists the common choices directly. To keep the options in view instead, choose View → Inline config; the page remembers that choice. Help → Keyboard shortcuts… lists the keys.

1. Paste the original into Before and the changed version into After.
2. Choose how to compare:
   • Lines for code, configuration and logs;
   • Words for prose;
   • Characters for short strings, identifiers or a single changed line.
3. Optionally tick Ignore whitespace or Ignore case.
4. Read the result:
   • the summary line counts what was added and removed;
   • Changes shows both texts merged, with removed text struck through in red and added text underlined in green;
   • Unified diff is the standard patch format. Copy it or download it as changes.diff.

To apply the unified diff to a file, set both labels to the file’s path (e.g. src/app.ts) and run patch -p0 < changes.diff. With a/src/app.ts and b/src/app.ts labels, use git apply changes.diff or patch -p1.

Swap exchanges the two texts, which reverses the diff.

Algorithm. jsdiff’s implementation of Myers’ O(ND) difference algorithm, which finds a shortest edit script. When several shortest scripts exist, jsdiff’s choice is used.

Modes.

Options.

Without ignore_whitespace, whitespace is significant in every mode: in words mode, a b and a b differ.

Structured output.

• changes: every piece of both texts, in order, typed added, removed or unchanged, with its size (count) in the mode’s unit. A replacement is a removed piece followed by an added piece.
• Rebuilding the texts: concatenating unchanged + removed values reproduces before; unchanged + added reproduces after. When an ignore option is set, unchanged pieces carry the after text.
• identical: true when changes contains nothing but unchanged, i.e. equal under the options.
• exact_match: compares the raw strings.

Statistics.

• added, removed, unchanged: totals in the mode’s unit.
• change_blocks: number of contiguous runs of changed pieces.
• lines_added, lines_removed: + and - lines of the unified diff, as diffstat counts them.
• chars_added, chars_removed: code points in the added and removed pieces.

Unified diff.

• Granularity: always line-based, whatever the mode, using the same ignore_whitespace and ignore_case. Context lines shown when ignoring come from after.
• Headers: --- <from_label> and +++ <to_label>, then hunks @@ -<start>,<count> +<start>,<count> @@ with context unchanged lines around each change (default 3, as GNU diff). Changes separated by at most twice context unchanged lines (6 by default) share one hunk.
• Counts in headers: jsdiff always writes the count, even when it is 1 (@@ -11,1 +11,1 @@; GNU diff writes @@ -11 +11 @@). patch and git apply accept both.
• Empty side: an empty range is written as start 0, count 0 (@@ -0,0 +1,1 @@).
• Missing final newline: a line without one is followed by \ No newline at end of file.
• Labels: a label containing non-ASCII characters, control characters, " or \ is quoted C-style, as Git does ("caf\303\251.txt").
• No differences: unified is an empty string and hunks is empty.

hunks holds the same data as unified, with the header parsed into numbers and each line kept with its , - or + prefix (line breaks removed).

Limits.

• Text size: each text may be up to 200,000 characters (UTF-16 code units).
• Edit size: a comparison stops with a 422 error once the edit script would exceed 5,000 lines (lines, and the unified diff in every mode) or 2,000 tokens (words, characters). Myers’ algorithm costs time proportional to text length × number of edits, so these caps bound the worst case. For heavily rewritten texts use lines mode or compare smaller sections.