How diff works, and why it sometimes looks insane

Every merge conflict you have ever fought came out of one algorithm making a reasonable guess. Understanding the guess makes the conflicts much less mysterious.

Diff is solving longest common subsequence

Given two files, diff finds the longest sequence of lines appearing in both, in order, and calls everything else an insertion or a deletion. It is not comparing meaning — it is finding the biggest shared skeleton and reporting the leftovers.

The classic implementation is Myers' 1986 algorithm, which is what Git uses by default. Its output is minimal in the number of changed lines, which is not always the output a human would call clearest.

Why it mangles prose and moved blocks

Diff is line-based. Reflow a paragraph and every line changes, so a one-word edit is reported as a total rewrite. This is why writers hate version control and programmers barely notice — code is already line-broken in semantically meaningful places.

Moved code has the same problem: a function relocated unchanged shows as a deletion plus an unrelated insertion, because "moved" is not a concept the algorithm has. Git's --color-moved and histogram diff exist to paper over exactly this.

Practical fixes

• Use --word-diff for prose — it compares at word granularity instead of line
• Set diff.algorithm = histogram in Git; it produces noticeably more readable output on real code than the default
• One sentence per line in Markdown makes documentation diff like code
• For anything structured (JSON, XML), normalise formatting before comparing or you will diff the whitespace, not the data

What you could build here

Three directions that would work on this name. None of them is the only answer — it is your name once it is yours.

1. A better diff tool. Fast, private, no upload — everything client-side. Developers share tools like this constantly when they are genuinely good.
2. A format specialist. JSON, CSV, XML and PDF comparison are all real recurring needs that the generic tools handle badly.
3. A funnel for something bigger. Free utility on top, team features underneath. Half the developer SaaS you use started as a single free tool with a memorable name.

Compare something now

• Text Diff OnlineFree side-by-side text comparison

Who you would be buying from

We build and run the sites below. That is the whole pitch: this name is not sitting in a speculator's spreadsheet — it comes from people who make things on the web and know what a good name is worth to someone who is going to use it.

• OMXUSThe studio behind all of this — software, licensing, and the infrastructure the rest of these sites run on.
• DnD Dice RollerA dice tool with no ads, no sign-up and no loading screen. It exists because every other one is exhausting.
• Legend RollerCharacter sheets, campaign rolling and party state, built for people who actually run games weekly.
• HorrorMoviesListCurated horror with real writing behind the picks, not scraped metadata dressed up as a list.
• MyNESGamesYour own NES ROMs, playable in a browser tab, with nothing uploaded anywhere.
• VoteAussieDigital identity with no central authority — Australia already runs elections without ID, so this asks why the internet cannot.
• The Price of WarA 3D memorial that makes the number of dead something you have to move through rather than read.
• NumeroscopeNumerology written honestly — for reflection, not prophecy. Proof that a mystical niche can be done without lying to anyone.
• Text Diff OnlinePaste two things, see what changed. Free, instant, nothing leaves the browser.
• NeuAIs ObservatoryA modular observatory for watching how AI systems actually behave.

Search everything we have made →