Text Diff & Compare

Both texts are split into tokens — lines, words or characters — and compared with Myers' shortest-edit-script algorithm (1986), the same idea behind diff and Git. It finds the smallest number of insertions and deletions that turn the original into the changed text; everything it does not touch is the longest common subsequence. The common start and end are trimmed first so a small change in a large file costs almost nothing.

Added and removed are token counts from the script; unchanged is the common subsequence. Similarity is 2 × unchanged ÷ (length of A + length of B), so identical texts score 100 % and texts with nothing in common 0 %. A line edited in place counts as one removed and one added — the tool does not guess that they are "the same line, modified", though the view places them together.

With the ignore options on, tokens are compared after collapsing whitespace runs or lower-casing, but the text shown is always the original; a line that differs only in those ways is listed as unchanged with both versions. In line mode the unified diff (three lines of context, @@ -a,n +b,m @@ hunks) can be copied into a review or applied with patch.

A shortest script is not unique: other alignments with the same edit count exist, and a moved block always appears as a deletion plus an insertion. Very large differences (over 4,000 edits between the common start and end) are shown as one removed and one added block, and the page says so. Inputs up to 2 MB each are compared on your own machine; nothing is uploaded. The same four anonymous usage counts as the rest of the site apply.