Compare two texts

What counts as the same line

Comparison happens on a key rather than on the line itself. Each line is reduced to a key according to whichever ignore options are switched on, the diff runs over the keys, and what you are shown is always your original text — so turning on "ignore capitalisation" changes which lines match without altering a character of what is displayed. There are two separate whitespace options and they differ more than their labels suggest: one trims the ends of a line, the other first collapses runs of spaces and tabs down to a single space and then trims. Neither removes whitespace altogether, and the collapsing looks for literal spaces and tabs only, so an ideographic space or a non-breaking space in the middle of a line is left alone and still registers as a difference.

Ignoring blank lines is the option with a consequence nobody mentions: the blank lines are removed before the comparison runs, so the line numbers beside the result count only the lines that survived, and stop matching the line numbers in your file. With the option off, they match exactly. The word-level highlighting inside a changed line follows the ignore-capitalisation setting and nothing else — the whitespace options do not reach it, so spacing differences inside a line that was changed for some other reason are still painted as changes.

The copyable result is a unified diff in every way that matters for reading it — the two file headers, minus lines, plus lines, three lines of context around each change — with one exception worth knowing before you rely on it. Its hunk markers carry no line ranges, so it is made for a bug report, a code review comment or an email, and the patch and git apply commands will refuse it. If what you need is something to apply rather than something to read, produce it with git.

When you want something else

Structured formats compare badly as lines, and the failure is loud rather than subtle: two JSON documents differing only in the order of their keys, or two CSVs with one column moved, come back as a wall of changes that tells you nothing. Diff them with something that understands the shape instead — sort the keys of both JSON files first and compare the results, or use daff, which diffs CSV as rows and columns and will tell you a column moved rather than that every line changed.

Two directories, or two points in a repository's history, are git's question rather than this page's. And for prose where the paragraphs have been rewrapped, a line-based view is the wrong granularity entirely — git diff --word-diff ignores where the line breaks fell and shows you only the words that moved, which is the comparison you actually wanted when a document has been reflowed.