Diff Checker

How the comparison is computed

The alignment uses Myers' algorithm, which is the same one git diff uses by default. Its useful property is that the cost scales with how different the two inputs are rather than how large they are, so two nearly identical thousand-line files are aligned almost instantly. The common prefix and suffix are stripped before the algorithm runs, which reduces a one-line change in a long file to a one-line problem.

The alternative, the textbook dynamic-programming table, would need twenty-five million cells for two five-thousand-line files. That is both slow and roughly a hundred megabytes of memory, in a browser tab, for a diff.

Why modified lines are paired

An edit script is a list of deletions and insertions, and showing them in that raw form is what makes most diff viewers hard to read: a changed line appears as a deletion several rows above its replacement. Adjacent deletions and insertions are paired here into single modified rows, and within each pair a second word-level comparison highlights only the words that actually changed.

That is the difference between seeing that a line changed and seeing what changed in it - which for a long line of configuration or a rewritten sentence is the entire question.

• Side-by-side or unified view, with three lines of context in unified
• Word-level or whole-line highlighting
• Ignore whitespace, for comparing re-indented code
• Ignore case, with a warning if that is the only reason the sides match

The unified output is a real patch

The unified view produces the same format as diff -u and git diff, hunk headers included, so it can be pasted into a code review, an issue or a commit message and will render as a diff wherever that format is understood. The download gives you a .patch file.

What it will not do

It is not a merge tool. It shows you what differs; it will not resolve conflicts, produce a combined version, or apply a patch. It also has no notion of syntax, so it cannot tell you that two versions of a function are semantically equivalent despite differing formatting - turning on whitespace-insensitivity is the closest approximation.

Inputs that differ by more than a few thousand lines are reported as a whole-block replacement rather than aligned line by line, because a per-line alignment of two unrelated documents produces noise rather than information. And where the two sides match only because whitespace or case is being ignored, the tool says so rather than reporting them as identical.