Text Compare (Diff Checker)

What counts as a difference, and how it is found

A diff answers one question: what is the smallest set of removals and additions that turns the original into the changed text? Everything the two versions share stays plain; everything else is marked. The search behind it is the algorithm Eugene Myers published in 1986, the one git uses by default, and short of the work limit described below, it finds a shortest answer rather than merely a plausible one.

A few rules decide what you see, and they are worth knowing before you trust a result:

• A changed line is a pairing, not a fact. When a block of lines is replaced, each old line is matched with the new line that shares the most wording, in order. Matched pairs are shown old above new with the differing words darker; lines with no good match are plain removals and additions.
• Line endings never count. A file saved on Windows (CRLF) and the same file saved on a Mac (LF) compare as identical, and so do two texts that differ only in a final line break.
• Words include punctuation. In word mode a comma or a full stop is a word of its own, so adding one is counted. Chinese and Japanese, written without spaces, are compared one character at a time.
• Ignored differences stay out of sight. With case or whitespace ignored, an unchanged line is shown as it appears in the changed text.

Line or word, and which boxes to tick

• Line: code, configuration files, CSV rows, lists, subtitles and logs, anything where a line is a unit that means something on its own.
• Word: prose. Contracts, essays, emails and translations keep a whole paragraph on one line, and line mode can only report that the paragraph changed.
• Ignore whitespace: reindented code, text copied out of a PDF or an email, and in word mode a paragraph rewrapped at a different width.
• Ignore case: SQL written in capitals by one person and in lowercase by another, headings converted to title case, email addresses and host names.
• Hide unchanged lines: long files. Each change keeps three lines of context on either side, the same amount a unified diff shows, and the rest folds into a single line saying how much was hidden.

How to see what someone changed in a document

1. Paste the version you sent into Original text and the one that came back into Changed text. The order matters: additions are what the second box has and the first does not. If the colours look backwards, press Swap.
2. Switch to Word. Prose edits are usually a word or a phrase inside a long paragraph, and word mode marks exactly those.
3. If either version passed through a PDF or an email, tick Ignore whitespace so that different line breaks do not bury the real edits.
4. Read the counts first, then work through the highlights from the top. A number or a date changed inside a sentence is the edit people miss most often when reading two versions side by side.

How to compare two versions of a config file or snippet

1. Stay on Line and keep Hide unchanged lines ticked, so a long file collapses to the parts that differ.
2. Tick Ignore whitespace only when indentation carries no meaning. In Python and YAML it does, and ignoring it can hide a change that breaks the file.
3. For JSON, format both documents the same way first with the JSON formatter, so that a minified version and a pretty one line up.
4. Press Copy diff and paste the result into a pull request comment or a ticket, as described below.

Why the same change can be drawn two ways

A shortest diff is often not unique. Add a new function below one that ends in a closing brace, and the diff can mark your lines as added, or it can treat the old function’s closing brace as the end of your new one and your last brace as the end of the old one. The same number of changes, drawn differently, and both are correct descriptions of the edit. Myers’ method settles such ties the same way every time, which is why git offers alternatives such as --diff-algorithm=patience and histogram that try harder to anchor on lines that occur only once. If a result looks odd, check the counts: the number of changed lines does not depend on how the tie was settled.

Getting the result out of the page

• Copy diff in line mode gives a unified diff: @@ headers with line numbers, then each line prefixed with a space, a minus or a plus, with three lines of context around every change.
• Paste it into a fenced code block marked diff on GitHub or GitLab, and the removed and added lines are coloured for whoever reads the comment. Many other Markdown editors and chat apps do the same at the time of writing.
• Copy diff in word mode writes removals as [-old-] and additions as {+new+}, which survives being pasted into a plain email.
• Selecting by hand leaves out the line numbers and the plus and minus signs, so a copied block of unchanged lines pastes as clean text.
• Nothing goes in the address bar. A copied link shares the tool, never the texts you compared.

The same comparison in Word, Google Docs, VS Code and git

• Microsoft Word: Review, then Compare, then pick the original and the revised document. The result is a third document with the differences as tracked changes, formatting included.
• Google Docs: Tools, then Compare documents. It produces a new document in which the differences appear as suggested edits.
• VS Code: right click a file and choose Select for Compare, then right click the second one and choose Compare with Selected. From a terminal, code --diff old.txt new.txt opens the same view.
• git: git diff --no-index old.txt new.txt compares any two files, even outside a repository. Add --word-diff for word mode, -b to ignore changes in the amount of whitespace, or -w to ignore whitespace entirely.
• Command line: diff -u old.txt new.txt on macOS and Linux prints a unified diff; in the Windows command prompt, fc old.txt new.txt lists the lines that differ.

When to use which

The office suites are the right tool when formatting matters, since they compare bold, links and styles that plain text cannot carry. git and VS Code win for files you already have on disk. A page like this one wins for two snippets that only exist in a chat, an email and a web form.

What a diff cannot tell you

• Invisible characters are real differences. A non-breaking space counts as whitespace and Ignore whitespace handles it; a zero width space does not, and neither does a Cyrillic letter that looks exactly like a Latin one.
• Moves are not detected. A paragraph moved elsewhere shows as removed in one place and added in another.
• Only characters compare. Bold, italics, links, fonts and comments do not survive a paste as plain text, so a change to them is invisible here.
• Changed lines are an interpretation. A line rewritten beyond recognition appears as one removal and one addition, and the counts follow that choice.
• Scripts without spaces compare in long runs. Thai, Lao, Khmer and Burmese have no spaces between words, so word mode treats each run between spaces and punctuation as one word.
• Very different long texts hit a limit. Past a set amount of work the unaligned part is shown as one block out and one block in, with a note, rather than freezing the tab.