The git diff format is the evolution of the unified format, proposed to bridge the BSD and GNU diff formats.

The official docs on the hunk header: https://github.com/git/git/blob/master/Documentation/diff-generate-patch.txt#L154-L185

Let's take this example:

# example
@@ -108,3 +123,8 @@ function onInputChanged(text, suggest) {`
# generically
@@ -<old_offset, old_line_count> +<new_offset, new_line_count> @@ <hunk_context>`

It starts with @@ -, though a combined diff (the default) have >2 @s. It'll always have the the -A,B followed by the +C,D block, except for a file deletion where the first block will be just -A without a ,B bit. You can also end up with a hunk like @@@ -105,13 -104,21 +104,23 @@@

The hunk context doesn't necessarily show the source from that line, but will often show the function name. see https://stackoverflow.com/questions/28111035/where-does-the-excerpt-in-the-git-diff-hunk-header-come-from

In Git this mostly terminates in xdiff: https://github.com/git/git/blob/master/xdiff/xemit.c#L128-L149

However there is also a bunch of language specific patterns used in userdiff.c: https://github.com/git/git/blob/ab5d01a29eb7380ceab070f0807c2939849c44bc/userdiff.c#L15-L152

Issues where we dug into the formats:

• Should we change how hunk headers look? · Issue #100 · so-fancy/diff-so-fancy
• line number of diff is incorrect. · Issue #120 · so-fancy/diff-so-fancy