Code diff that ignores the reformat

Tick the filters for the changes you did not make by hand: whitespace-only, indentation-only, comment-only and trailing-comma. Each one normalises both sides before the comparison rather than hiding rows afterwards, so a file that Prettier or Black reindented from two spaces to four, added trailing commas to and re-wrapped comments in comes out showing only the logic you actually edited.

Patience diff matches the lines that occur exactly once in both files first, and only then compares what lies between those anchors. Bram Cohen described it in 2006 after noticing that minimal diffs align the wrong braces: a closing } is identical to every other closing }, so the shortest edit script happily pairs your new one with someone else's and reports a change that starts halfway up the previous function. A unique line — a function signature, an import, a distinctive string — cannot be mismatched that way, so anchoring on those produces hunks that start where a human would say the change starts. It is rarely the smallest possible diff, and that is the trade: git ships the same idea as --patience and its refinement --histogram for exactly this reason. This page anchors first and falls back to Myers between anchors, or wherever no unique line exists.

Yes — that is what the violet blocks are. After the line comparison, every deleted line and every inserted line is re-matched against the other side, and any run of three lines or more that turns up intact somewhere else is relabelled as a move with the destination line number attached. A line diff has no concept of a move on its own, which is why lifting a 40-line function to the top of a file normally reads as 40 deletions and 40 unrelated insertions. Git added --color-moved in 2.15 for the same reason. Moves are matched on the same normalised text as the diff, so a block that was moved and reindented still registers as a move when the whitespace filter is on.

No, because the comment filter is a small scanner rather than a regular expression. A regex that deletes everything after // also deletes half of const endpoint = "https://api.example.com/v2", which would quietly change what is being compared — the worst possible failure for a diff. The scanner tracks single quotes, double quotes, template literals and backslash escapes, and only treats a comment opener as one when it is outside all of them. It also carries block-comment state from line to line, so a /* that opens on line 12 and closes on line 18 removes those seven lines and nothing else.

Not with the whitespace filter on, which is its default state. Both sides get every run of spaces and tabs collapsed to one space and both ends trimmed before they are compared, so a file converted from tabs to four spaces compares as unchanged. If you want the opposite — to see indentation changes but not count them as rewrites — untick whitespace and tick indentation-only instead: that trims the leading and trailing whitespace only, and the summary then tells you how many of the changed lines differ in nothing else.

Fifteen, plus a plain-text mode that highlights nothing: JavaScript and TypeScript, JSX and TSX, JSON, Python, Go, Rust, Java and C#, C and C++, PHP, Ruby, shell, YAML, SQL, CSS and SCSS, and HTML and XML. Highlighting is per line and keyword-based rather than a full parse, which is why it costs about 12 milliseconds for 5,000 lines and never gets in the way of the comparison. The language choice matters more for the comment filter than for the colours — it selects between four comment grammars: // with /* */, #, --, and <!-- -->.

No, and for this page that is the point: proprietary source is exactly what people are nervous about pasting into a web tool. Both revisions are compared, highlighted and rendered by JavaScript inside this tab, no request is made while you work, and nothing survives closing it.