Text Diff

How it works

1. 01

   Paste the original into the left box and the changed version into the right. The diff recomputes on every keystroke.

2. 02

   Read the markers. A leading - is a line only the original has, + is a line only the changed side has, and a space is a line both share. The background tint repeats the same three states, so the marker is never the only signal.

3. 03

   Turn on "Ignore leading and trailing whitespace" when the indentation moved but the content did not, and "Ignore case" when only capitalisation is in question. Both change what counts as a match; every line is still shown exactly as it was typed.

4. 04

   Switch on "Only changed lines" to drop the shared context on a long file, and use Copy diff to take the marked output with you.

5. 05

   Everything runs in this tab. The text is never uploaded, and nothing is saved between visits.

How the comparison works

The two sides are split into lines and matched with a longest common subsequence, computed by dynamic programming. Every line in the LCS is unchanged; every line of the original that is not in it was removed; every line of the changed side that is not in it was added. That is what separates this from comparing line 1 to line 1, line 2 to line 2 and so on — under a zip comparison, inserting a single line at the top reports every line below it as changed, because each one has shifted by one position. The table is O(n x m) in the two line counts, so the identical prefix and suffix are stripped first and the table is built only over the part that actually differs. Inputs are capped at 2,000 lines per side, which bounds the worst case at roughly four million cells.

LCS(i, j) = 1 + LCS(i+1, j+1) if a[i] = b[j], else max(LCS(i+1, j), LCS(i, j+1))

Use cases

Questions

Is anything uploaded?

No. The comparison runs in this tab. There is no upload, no request, no analytics, and nothing is written to storage — reload the page and both boxes are back to their defaults.

How does it decide which lines match?

It computes a longest common subsequence over the lines with the standard dynamic-programming table. That matters: comparing line 1 to line 1, line 2 to line 2 and so on would report every line after a single insertion as changed, because each one has shifted position.

Is there a word-level or character-level diff?

No. This is a line diff only. A line that changed by one character is reported as one removal and one addition, with no highlight inside the line. That is the honest limit of the tool and the main thing it does not do.

How large an input does it take?

The first 2,000 lines of each side. The table is O(n x m) in the two line counts, so two entirely different 2,000-line texts is the slow case; the identical prefix and suffix are stripped before the table is built, which is why a small edit in a large file stays instant. When the cap is hit, the tool says so above the output rather than quietly comparing a fragment.

What exactly does "Ignore leading and trailing whitespace" do?

It trims each line at both ends before comparing, so indentation and trailing spaces stop counting as differences. Whitespace inside a line still counts — two spaces between words is not the same line as one. The lines are always displayed as typed; only the matching changes.

Does it handle Windows line endings?

Yes. CRLF and lone CR are normalised to LF before the text is split, so the same content saved on Windows and on Linux compares as identical instead of as a complete rewrite. Note the flip side: this tool will not tell you that the line endings differ.

Why is there a blank line at the end of my diff?

A trailing newline produces an empty final line, and that is a real difference between a file that ends with a newline and one that does not. If only one side has it, you will see a single empty added or removed line at the bottom.