Text-Tools

Comparing two blocks of text is a thing people do constantly and almost always at a bad moment: a config that works on one host and not another, a response body that changed between deploys, a contract clause someone edited without tracking changes, a log line that is nearly but not quite the one from yesterday.

The temptation is to eyeball it. That works for three lines and fails silently for thirty, because human attention is extremely good at seeing what it expects. A diff is not a convenience in that situation; it is the difference between finding the change and convincing yourself there isn’t one.

What makes a diff tool good is granularity. Line-level comparison is the right default and is what version control shows you, but it is actively misleading on two classes of input: minified or single-line files, where the entire document is one line and the diff degenerates into “this line changed”; and long prose lines, where a one-word edit paints a whole paragraph as rewritten. Word and character modes exist precisely for those cases, and being able to switch between the three without re-pasting is most of the value.

The other half is that the output has to be portable. A diff you can only look at is worth less than one you can hand to somebody else, so the unified-patch export produces a standard patch with proper hunk headers — the format patch(1) accepts and the format a code-review tool renders correctly when you paste it into a comment.

Text is often the most sensitive thing on this site. A JSON payload is usually machine data; a text diff is frequently a draft contract, an internal memo, an incident timeline or a customer email. That is the reason the comparison happens in your browser rather than on a server: not as a feature, but because uploading that material to an unknown host to answer a question you could answer locally is a bad trade at any price.

If the two things you are comparing are JSON rather than prose, use JSON Diff instead. It compares the parsed structure rather than the text, so reformatting and key reordering — which a text diff reports as hundreds of changes — correctly show as no change at all.