Merge pull request #952 from lark-parser/MegaIng-better_repeats_refactor

lark/load_grammar.py

@@ -9,7 +9,7 @@ import pkgutil
 from ast import literal_eval
 from numbers import Integral
 
-from .utils import bfs, Py36, logger, classify_bool, is_id_continue, is_id_start, bfs_all_unique
+from .utils import bfs, Py36, logger, classify_bool, is_id_continue, is_id_start, bfs_all_unique, small_factors
 from .lexer import Token, TerminalDef, PatternStr, PatternRE
 
 from .parse_tree_builder import ParseTreeBuilder
@@ -175,27 +175,136 @@ RULES = {
 }
 
 
+# Value 5 keeps the number of states in the lalr parser somewhat minimal
+# It isn't optimal, but close to it. See PR #949
+SMALL_FACTOR_THRESHOLD = 5
+# The Threshold whether repeat via ~ are split up into different rules
+# 50 is chosen since it keeps the number of states low and therefore lalr analysis time low,
+# while not being to overaggressive and unnecessarily creating rules that might create shift/reduce conflicts.
+# (See PR #949)
+REPEAT_BREAK_THRESHOLD = 50
+
+
 @inline_args
 class EBNF_to_BNF(Transformer_InPlace):
     def __init__(self):
         self.new_rules = []
-        self.rules_by_expr = {}
+        self.rules_cache = {}
         self.prefix = 'anon'
         self.i = 0
         self.rule_options = None
 
-    def _add_recurse_rule(self, type_, expr):
-        if expr in self.rules_by_expr:
-            return self.rules_by_expr[expr]
-
-        new_name = '__%s_%s_%d' % (self.prefix, type_, self.i)
+    def _name_rule(self, inner):
+        new_name = '__%s_%s_%d' % (self.prefix, inner, self.i)
         self.i += 1
-        t = NonTerminal(new_name)
-        tree = ST('expansions', [ST('expansion', [expr]), ST('expansion', [t, expr])])
-        self.new_rules.append((new_name, tree, self.rule_options))
-        self.rules_by_expr[expr] = t
+        return new_name
+
+    def _add_rule(self, key, name, expansions):
+        t = NonTerminal(name)
+        self.new_rules.append((name, expansions, self.rule_options))
+        self.rules_cache[key] = t
         return t
 
+    def _add_recurse_rule(self, type_, expr):
+        try:
+            return self.rules_cache[expr]
+        except KeyError:
+            new_name = self._name_rule(type_)
+            t = NonTerminal(new_name)
+            tree = ST('expansions', [
+                ST('expansion', [expr]),
+                ST('expansion', [t, expr])
+            ])
+            return self._add_rule(expr, new_name, tree)
+
+    def _add_repeat_rule(self, a, b, target, atom):
+        """Generate a rule that repeats target ``a`` times, and repeats atom ``b`` times.
+
+        When called recursively (into target), it repeats atom for x(n) times, where:
+            x(0) = 1
+            x(n) = a(n) * x(n-1) + b
+
+        Example rule when a=3, b=4:
+
+            new_rule: target target target atom atom atom atom
+
+        """
+        key = (a, b, target, atom)
+        try:
+            return self.rules_cache[key]
+        except KeyError:
+            new_name = self._name_rule('repeat_a%d_b%d' % (a, b))
+            tree = ST('expansions', [ST('expansion', [target] * a + [atom] * b)])
+            return self._add_rule(key, new_name, tree)
+
+    def _add_repeat_opt_rule(self, a, b, target, target_opt, atom):
+        """Creates a rule that matches atom 0 to (a*n+b)-1 times.
+
+        When target matches n times atom, and target_opt 0 to n-1 times target_opt,
+
+        First we generate target * i followed by target_opt, for i from 0 to a-1
+        These match 0 to n*a - 1 times atom
+
+        Then we generate target * a followed by atom * i, for i from 0 to b-1
+        These match n*a to n*a + b-1 times atom
+
+        The created rule will not have any shift/reduce conflicts so that it can be used with lalr
+
+        Example rule when a=3, b=4:
+
+            new_rule: target_opt
+                    | target target_opt
+                    | target target target_opt
+
+                    | target target target
+                    | target target target atom
+                    | target target target atom atom
+                    | target target target atom atom atom
+
+        """
+        key = (a, b, target, atom, "opt")
+        try:
+            return self.rules_cache[key]
+        except KeyError:
+            new_name = self._name_rule('repeat_a%d_b%d_opt' % (a, b))
+            tree = ST('expansions', [
+                ST('expansion', [target]*i + [target_opt]) for i in range(a)
+            ] + [
+                ST('expansion', [target]*a + [atom]*i) for i in range(b)
+            ])
+            return self._add_rule(key, new_name, tree)
+
+    def _generate_repeats(self, rule, mn, mx):
+        """Generates a rule tree that repeats ``rule`` exactly between ``mn`` to ``mx`` times.
+        """
+        # For a small number of repeats, we can take the naive approach
+        if mx < REPEAT_BREAK_THRESHOLD:
+            return ST('expansions', [ST('expansion', [rule] * n) for n in range(mn, mx + 1)])
+
+        # For large repeat values, we break the repetition into sub-rules. 
+        # We treat ``rule~mn..mx`` as ``rule~mn rule~0..(diff=mx-mn)``.
+        # We then use small_factors to split up mn and diff up into values [(a, b), ...]
+        # This values are used with the help of _add_repeat_rule and _add_repeat_rule_opt
+        # to generate a complete rule/expression that matches the corresponding number of repeats
+        mn_target = rule
+        for a, b in small_factors(mn, SMALL_FACTOR_THRESHOLD):
+            mn_target = self._add_repeat_rule(a, b, mn_target, rule)
+        if mx == mn:
+            return mn_target
+
+        diff = mx - mn + 1  # We add one because _add_repeat_opt_rule generates rules that match one less
+        diff_factors = small_factors(diff, SMALL_FACTOR_THRESHOLD)
+        diff_target = rule  # Match rule 1 times
+        diff_opt_target = ST('expansion', [])  # match rule 0 times (e.g. up to 1 -1 times)
+        for a, b in diff_factors[:-1]:
+            diff_opt_target = self._add_repeat_opt_rule(a, b, diff_target, diff_opt_target, rule)
+            diff_target = self._add_repeat_rule(a, b, diff_target, rule)
+
+        a, b = diff_factors[-1]
+        diff_opt_target = self._add_repeat_opt_rule(a, b, diff_target, diff_opt_target, rule)
+
+        return ST('expansions', [ST('expansion', [mn_target] + [diff_opt_target])])
+
     def expr(self, rule, op, *args):
         if op.value == '?':
             empty = ST('expansion', [])
@@ -220,7 +329,9 @@ class EBNF_to_BNF(Transformer_InPlace):
                 mn, mx = map(int, args)
                 if mx < mn or mn < 0:
                     raise GrammarError("Bad Range for %s (%d..%d isn't allowed)" % (rule, mn, mx))
-            return ST('expansions', [ST('expansion', [rule] * n) for n in range(mn, mx+1)])
+
+            return self._generate_repeats(rule, mn, mx)
+
         assert False, op
 
     def maybe(self, rule):

lark/utils.py

@@ -187,7 +187,7 @@ def get_regexp_width(expr):
                 return 1, sre_constants.MAXREPEAT
             else:
                 return 0, sre_constants.MAXREPEAT
-            
+
 ###}
 
 
@@ -288,7 +288,7 @@ except ImportError:
 
 class FS:
     exists = os.path.exists
-    
+
     @staticmethod
     def open(name, mode="r", **kwargs):
         if atomicwrites and "w" in mode:
@@ -359,3 +359,29 @@ def _serialize(value, memo):
         return {key:_serialize(elem, memo) for key, elem in value.items()}
     # assert value is None or isinstance(value, (int, float, str, tuple)), value
     return value
+
+
+
+
+def small_factors(n, max_factor):
+    """
+    Splits n up into smaller factors and summands <= max_factor.
+    Returns a list of [(a, b), ...]
+    so that the following code returns n:
+
+    n = 1
+    for a, b in values:
+        n = n * a + b
+
+    Currently, we also keep a + b <= max_factor, but that might change
+    """
+    assert n >= 0
+    assert max_factor > 2
+    if n <= max_factor:
+        return [(n, 0)]
+
+    for a in range(max_factor, 1, -1):
+        r, b = divmod(n, a)
+        if a + b <= max_factor:
+            return small_factors(r, max_factor) + [(a, b)]
+    assert False, "Failed to factorize %s" % n

tests/test_grammar.py

@@ -3,7 +3,7 @@ from __future__ import absolute_import
 import sys
 from unittest import TestCase, main
 
-from lark import Lark, Token, Tree
+from lark import Lark, Token, Tree, ParseError, UnexpectedInput
 from lark.load_grammar import GrammarError, GRAMMAR_ERRORS, find_grammar_errors
 from lark.load_grammar import FromPackageLoader
 
@@ -198,6 +198,53 @@ class TestGrammar(TestCase):
         x = find_grammar_errors(text)
         assert [e.line for e, _s in find_grammar_errors(text)] == [2, 6]
 
+    def test_ranged_repeat_terms(self):
+        g = u"""!start: AAA
+                AAA: "A"~3
+            """
+        l = Lark(g, parser='lalr')
+        self.assertEqual(l.parse(u'AAA'), Tree('start', ["AAA"]))
+        self.assertRaises((ParseError, UnexpectedInput), l.parse, u'AA')
+        self.assertRaises((ParseError, UnexpectedInput), l.parse, u'AAAA')
+
+        g = u"""!start: AABB CC
+                AABB: "A"~0..2 "B"~2
+                CC: "C"~1..2
+            """
+        l = Lark(g, parser='lalr')
+        self.assertEqual(l.parse(u'AABBCC'), Tree('start', ['AABB', 'CC']))
+        self.assertEqual(l.parse(u'BBC'), Tree('start', ['BB', 'C']))
+        self.assertEqual(l.parse(u'ABBCC'), Tree('start', ['ABB', 'CC']))
+        self.assertRaises((ParseError, UnexpectedInput), l.parse, u'AAAB')
+        self.assertRaises((ParseError, UnexpectedInput), l.parse, u'AAABBB')
+        self.assertRaises((ParseError, UnexpectedInput), l.parse, u'ABB')
+        self.assertRaises((ParseError, UnexpectedInput), l.parse, u'AAAABB')
+
+    def test_ranged_repeat_large(self):
+        g = u"""!start: "A"~60
+            """
+        l = Lark(g, parser='lalr')
+        self.assertGreater(len(l.rules), 1, "Expected that more than one rule will be generated")
+        self.assertEqual(l.parse(u'A' * 60), Tree('start', ["A"] * 60))
+        self.assertRaises(ParseError, l.parse, u'A' * 59)
+        self.assertRaises((ParseError, UnexpectedInput), l.parse, u'A' * 61)
+
+        g = u"""!start: "A"~15..100
+            """
+        l = Lark(g, parser='lalr')
+        for i in range(0, 110):
+            if 15 <= i <= 100:
+                self.assertEqual(l.parse(u'A' * i), Tree('start', ['A']*i))
+            else:
+                self.assertRaises(UnexpectedInput, l.parse, u'A' * i)
+
+        # 8191 is a Mersenne prime
+        g = u"""start: "A"~8191
+            """
+        l = Lark(g, parser='lalr')
+        self.assertEqual(l.parse(u'A' * 8191), Tree('start', []))
+        self.assertRaises(UnexpectedInput, l.parse, u'A' * 8190)
+        self.assertRaises(UnexpectedInput, l.parse, u'A' * 8192)
 
 
 if __name__ == '__main__':

tests/test_parser.py

@@ -2204,27 +2204,7 @@ def _make_parser_test(LEXER, PARSER):
             self.assertRaises((ParseError, UnexpectedInput), l.parse, u'AAAABB')
 
 
-        def test_ranged_repeat_terms(self):
-            g = u"""!start: AAA
-                    AAA: "A"~3
-                """
-            l = _Lark(g)
-            self.assertEqual(l.parse(u'AAA'), Tree('start', ["AAA"]))
-            self.assertRaises((ParseError, UnexpectedInput), l.parse, u'AA')
-            self.assertRaises((ParseError, UnexpectedInput), l.parse, u'AAAA')
 
-            g = u"""!start: AABB CC
-                    AABB: "A"~0..2 "B"~2
-                    CC: "C"~1..2
-                """
-            l = _Lark(g)
-            self.assertEqual(l.parse(u'AABBCC'), Tree('start', ['AABB', 'CC']))
-            self.assertEqual(l.parse(u'BBC'), Tree('start', ['BB', 'C']))
-            self.assertEqual(l.parse(u'ABBCC'), Tree('start', ['ABB', 'CC']))
-            self.assertRaises((ParseError, UnexpectedInput), l.parse, u'AAAB')
-            self.assertRaises((ParseError, UnexpectedInput), l.parse, u'AAABBB')
-            self.assertRaises((ParseError, UnexpectedInput), l.parse, u'ABB')
-            self.assertRaises((ParseError, UnexpectedInput), l.parse, u'AAAABB')
 
         @unittest.skipIf(PARSER=='earley', "Priority not handled correctly right now")  # TODO XXX
         def test_priority_vs_embedded(self):