BUGFIX Earley: Now yielding a previously repressed ambiguity

lark/parsers/earley.py

@@ -15,7 +15,7 @@
 from ..lexer import Token
 from ..tree import Tree
 from ..exceptions import UnexpectedEOF, UnexpectedToken
-from ..utils import logger, OrderedSet
+from ..utils import logger, OrderedSet, dedup_list
 from .grammar_analysis import GrammarAnalyzer
 from ..grammar import NonTerminal
 from .earley_common import Item
@@ -169,6 +169,7 @@ def predict_and_complete(self, i, to_scan, columns, transitives):
                         items.append(new_item)
 
     def _parse(self, lexer, columns, to_scan, start_symbol=None):
+
         def is_quasi_complete(item):
             if item.is_complete:
                 return True
@@ -281,7 +282,7 @@ def parse(self, lexer, start):
         # If the parse was successful, the start
         # symbol should have been completed in the last step of the Earley cycle, and will be in
         # this column. Find the item for the start_symbol, which is the root of the SPPF tree.
-        solutions = [n.node for n in columns[-1] if n.is_complete and n.node is not None and n.s == start_symbol and n.start == 0]
+        solutions = dedup_list(n.node for n in columns[-1] if n.is_complete and n.node is not None and n.s == start_symbol and n.start == 0)
         if not solutions:
             expected_terminals = [t.expect.name for t in to_scan]
             raise UnexpectedEOF(expected_terminals, state=frozenset(i.s for i in to_scan))
@@ -293,16 +294,21 @@ def parse(self, lexer, start):
             except ImportError:
                 logger.warning("Cannot find dependency 'pydot', will not generate sppf debug image")
             else:
-                debug_walker.visit(solutions[0], "sppf.png")
-
+                for i, s in enumerate(solutions):
+                    debug_walker.visit(s, f"sppf{i}.png")
 
-        if len(solutions) > 1:
-            assert False, 'Earley should not generate multiple start symbol items!'
 
         if self.Tree is not None:
             # Perform our SPPF -> AST conversion
             transformer = ForestToParseTree(self.Tree, self.callbacks, self.forest_sum_visitor and self.forest_sum_visitor(), self.resolve_ambiguity)
-            return transformer.transform(solutions[0])
+            solutions = [transformer.transform(s) for s in solutions]
+
+            if len(solutions) > 1:
+                t: Tree = self.Tree('_ambig', solutions)
+                t.expand_kids_by_data('_ambig')     # solutions may themselves be _ambig nodes
+                return t
+            return solutions[0]
 
         # return the root of the SPPF
+        # TODO return a list of solutions, or join them together somehow
         return solutions[0]

lark/parsers/earley_common.py

@@ -20,13 +20,13 @@ def __init__(self, rule, ptr, start):
             self.s = (rule, ptr)
             self.expect = rule.expansion[ptr]
             self.previous = rule.expansion[ptr - 1] if ptr > 0 and len(rule.expansion) else None
-        self._hash = hash((self.s, self.start))
+        self._hash = hash((self.s, self.start, self.rule))
 
     def advance(self):
         return Item(self.rule, self.ptr + 1, self.start)
 
     def __eq__(self, other):
-        return self is other or (self.s == other.s and self.start == other.start)
+        return self is other or (self.s == other.s and self.start == other.start and self.rule == other.rule)
 
     def __hash__(self):
         return self._hash

tests/test_parser.py

@@ -828,6 +828,22 @@ def test_term_ambig_resolve(self):
             tree = parser.parse(text)
             self.assertEqual(tree.children, ['foo', 'bar'])
 
+        def test_multiple_start_solutions(self):
+            grammar = r"""
+                !start: a | A
+                !a: A
+                A: "x"
+            """
+
+            l = Lark(grammar, ambiguity='explicit', lexer=LEXER)
+            tree = l.parse('x')
+
+            expected = Tree('_ambig', [
+                Tree('start', ['x']),
+                Tree('start', [Tree('a', ['x'])])]
+            )
+            self.assertEqual(tree, expected)
+
         def test_cycle(self):
             grammar = """
             start: start?
@@ -843,16 +859,24 @@ def test_cycle(self):
 
         def test_cycle2(self):
             grammar = """
-            start: _operation
-            _operation:  value
-            value: "b"
-                  | "a" value
-                  | _operation
+            start: _recurse
+            _recurse:  v
+            v: "b"
+                  | "a" v
+                  | _recurse
             """
 
             l = Lark(grammar, ambiguity="explicit", lexer=LEXER)
             tree = l.parse("ab")
-            self.assertEqual(tree, Tree('start', [Tree('value', [Tree('value', [])])]))
+            expected = (
+                Tree('start', [
+                    Tree('_ambig', [
+                        Tree('v', [Tree('v', [])]),
+                        Tree('v', [Tree('v', [Tree('v', [])])])
+                    ])
+                ])
+            )
+            self.assertEqual(tree, expected)
 
         def test_cycles(self):
             grammar = """