The Diff Model

The diff inside a patch — or any standalone git diff — parses into a slice of FileChange, each with its hunks. Parse does this for you and fills Patch.Files; ParseDiff exposes it directly for diffs with no email around them.

files, err := mailpatch.ParseDiff(diffText)

The shape

type FileChange struct {
	OldPath   string
	NewPath   string
	Type      ChangeType // Modified | Added | Deleted | Renamed | Copied
	IsBinary  bool
	OldMode   string     // e.g. "100644"
	NewMode   string
	Additions int
	Deletions int
	Hunks     []Hunk
}

• Path() returns the current path — NewPath, or OldPath for a deletion.
• Type is inferred from new file mode / deleted file mode / rename/copy headers and from /dev/null on either side.
• IsBinary is set for Binary files differ and GIT binary patch.

Hunks and lines

type Hunk struct {
	OldStart, OldLines int
	NewStart, NewLines int
	Section            string // text after the closing @@
	Lines              []Line
}

type Line struct {
	Kind LineKind // Context | Add | Delete
	Text string   // leading +/-/space removed
}

Walk them to render or analyze:

for _, f := range files {
	fmt.Printf("%s %s (+%d -%d)\n", f.Type, f.Path(), f.Additions, f.Deletions)
	for _, h := range f.Hunks {
		fmt.Printf("  @@ -%d,%d +%d,%d @@ %s\n",
			h.OldStart, h.OldLines, h.NewStart, h.NewLines, h.Section)
		for _, ln := range h.Lines {
			switch ln.Kind {
			case mailpatch.Add:
				fmt.Println("  +", ln.Text)
			case mailpatch.Delete:
				fmt.Println("  -", ln.Text)
			case mailpatch.Context:
				fmt.Println("   ", ln.Text)
			}
		}
	}
}

Diffstat

Patch.Stat (and the counts on each FileChange) are computed from the parsed hunks, not from the textual diffstat git prints — so they reflect what is actually in the diff:

type DiffStat struct {
	FilesChanged int
	Additions    int
	Deletions    int
}

Bare diffs

ParseDiff does not need a git header. A plain --- / +++ / @@ diff parses too, which makes it handy for diffs pasted into issues or produced by tools other than git. Unrecognized lines are ignored, so a diff embedded in surrounding prose still parses.