Fix diff parser: Need to calculate the prefix for the diff tokenizer better

parso/python/diff.py

@@ -388,6 +388,10 @@ class DiffParser(object):
                     typ, string, start_pos, prefix = next(tokens)
                     if '\n' in prefix or '\r' in prefix:
                         prefix = re.sub(r'[^\n\r]+\Z', '', prefix)
+                    else:
+                        assert start_pos[1] >= len(prefix), repr(prefix)
+                        if start_pos[1] - len(prefix) == 0:
+                            prefix = ''
                     yield PythonToken(
                         PythonTokenTypes.ENDMARKER, '',
                         (start_pos[0] + line_offset, 0),

test/test_diff_parser.py

@@ -1012,3 +1012,25 @@ def test_special_no_newline_ending(differ):
     differ.initialize(code1)
     differ.parse(code2, copies=1, parsers=1, expect_error_leaves=True)
     differ.parse(code1, copies=1, parsers=0)
+
+
+def test_random_character_insertion(differ):
+    code1 = dedent('''\
+            def create(self):
+                1
+                if self.path is not None:
+                    return
+                # 3
+                # 4
+        ''')
+    code2 = dedent('''\
+            def create(self):
+                1
+                if 2:
+             x       return
+                # 3
+                # 4
+        ''')
+    differ.initialize(code1)
+    differ.parse(code2, copies=1, parsers=3, expect_error_leaves=True)
+    differ.parse(code1, copies=1, parsers=1)