Fix invalid UTF-8 bug

Author: FollowTheProcess

.typos.toml

@@ -1,2 +1,5 @@
+[files]
+extend-exclude = ["**/testdata/**"]
+
 [default.extend-words]
 decorder = "decorder" # Name of a Go linter

internal/diff/chardiff_test.go

@@ -75,6 +75,21 @@ func TestCharDiff(t *testing.T) {
 			added:          []byte("changed\n"),
 			wantHasChanged: true,
 		},
+		{
+			// Regression: invalid UTF-8 on the added side must not corrupt the join invariant.
+			// \xe2 is the start of a 3-byte sequence with no continuation bytes.
+			name:           "invalid UTF-8 in added side",
+			removed:        []byte("0"),
+			added:          []byte("\xe2"),
+			wantHasChanged: true,
+		},
+		{
+			// Regression: invalid UTF-8 on the removed side.
+			name:           "invalid UTF-8 in removed side",
+			removed:        []byte("\xe2"),
+			added:          []byte("0"),
+			wantHasChanged: true,
+		},
 	}
 
 	for _, tt := range tests {

internal/diff/chars.go

@@ -35,6 +35,13 @@ func CharDiff(removed, added []byte) InlineChange {
 		addedCore = added[:len(added)-1]
 	}
 
+	// Fall back to whole-line diff for invalid UTF-8: converting invalid bytes
+	// to runes replaces them with U+FFFD, making it impossible to reconstruct
+	// the original bytes from the segments.
+	if !utf8.Valid(removedCore) || !utf8.Valid(addedCore) {
+		return fallback(removed, added)
+	}
+
 	oldRunes := []rune(string(removedCore))
 	newRunes := []rune(string(addedCore))
 

internal/diff/testdata/fuzz/FuzzCharDiff/959c728e1ba9028d

@@ -0,0 +1,3 @@
+go test fuzz v1
+[]byte("0")
+[]byte("\xe2")