Fix empty SPPF node from xearley ignore carry-over (fixes #1598)

lark/parsers/xearley.py

@@ -112,10 +112,10 @@ def scan(i, to_scan):
                 else:
                     # Handle items carried over due to ignores
                     new_item = Item(item.rule, item.ptr, item.start)
-                    label = (new_item.s, new_item.start, i + 1)
-                    new_item.node = node_cache[label] if label in node_cache else node_cache.setdefault(label, self.SymbolNode(*label))
-                    if item.node:
+                    if item.node is not None:
                         # new_item.node and item.node both represent the same symbol, so merge their children
+                        label = (new_item.s, new_item.start, i + 1)
+                        new_item.node = node_cache[label] if label in node_cache else node_cache.setdefault(label, self.SymbolNode(*label))
                         for child in item.node.children:
                             new_item.node.add_family(new_item.s, child.rule, new_item.start, child.left, child.right)
 

tests/test_parser.py

@@ -905,6 +905,21 @@ def test_multiple_start_solutions(self):
             tree = l.parse('x')
             assert tree == Tree('start', [Tree('a', ['x'])])
 
+        @unittest.skipIf(LEXER=='basic', 'Ignore carry-over is dynamic-lexer-specific')
+        def test_ignore_carryover_with_priority(self):
+            """Items in to_scan with item.node=None must not produce an empty SPPF
+            node when carried over past an ignored sequence (issue #1598)."""
+            grammar = r"""
+                start: a
+                a.1: NAME
+                NAME: /[a-z]+/
+                %import common.WS
+                %ignore WS
+            """
+            l = Lark(grammar, parser='earley', lexer=LEXER)
+            tree = l.parse('   hello')
+            self.assertEqual(tree, Tree('start', [Tree('a', ['hello'])]))
+
         @unittest.skipIf(LEXER=='basic', 'This scenario only occurs with the dynamic lexers')
         def test_multiple_start_solutions2(self):
             grammar = r"""