Improve performance by skipping unnecessary normalisation (#3751)

[etc/vim.git] / src / blib2to3 / pgen2 / parse.py

diff --git a/src/blib2to3/pgen2/parse.py b/src/blib2to3/pgen2/parse.py
index dc405264bad454a92469cb613b6dd962c394e837..17bf118e9fcd8053b655b4efc42e26f840b40465 100644 (file)
--- a/src/blib2to3/pgen2/parse.py
+++ b/src/blib2to3/pgen2/parse.py
@@ -9,7 +9,6 @@ See Parser/parser.c in the Python distribution for additional info on
 how this parsing engine works.
 
 """
 how this parsing engine works.
 
 """

-import copy

 from contextlib import contextmanager
 
 # Local imports
 from contextlib import contextmanager
 
 # Local imports


@@ -18,7 +17,6 @@ from typing import (
     cast,
     Any,
     Optional,
     cast,
     Any,
     Optional,

-    Text,

     Union,
     Tuple,
     Dict,
     Union,
     Tuple,
     Dict,


@@ -29,13 +27,13 @@ from typing import (
     TYPE_CHECKING,
 )
 from blib2to3.pgen2.grammar import Grammar
     TYPE_CHECKING,
 )
 from blib2to3.pgen2.grammar import Grammar

-from blib2to3.pytree import NL, Context, RawNode, Leaf, Node
+from blib2to3.pytree import convert, NL, Context, RawNode, Leaf, Node

 
 if TYPE_CHECKING:
 
 if TYPE_CHECKING:

-    from blib2to3.driver import TokenProxy
+    from blib2to3.pgen2.driver import TokenProxy

 
 
 
 

-Results = Dict[Text, NL]
+Results = Dict[str, NL]

 Convert = Callable[[Grammar, RawNode], Union[Node, Leaf]]
 DFA = List[List[Tuple[int, int]]]
 DFAS = Tuple[DFA, Dict[int, int]]
 Convert = Callable[[Grammar, RawNode], Union[Node, Leaf]]
 DFA = List[List[Tuple[int, int]]]
 DFAS = Tuple[DFA, Dict[int, int]]


@@ -46,6 +44,17 @@ def lam_sub(grammar: Grammar, node: RawNode) -> NL:
     return Node(type=node[0], children=node[3], context=node[2])
 
 
     return Node(type=node[0], children=node[3], context=node[2])
 
 

+# A placeholder node, used when parser is backtracking.
+DUMMY_NODE = (-1, None, None, None)
+
+
+def stack_copy(
+    stack: List[Tuple[DFAS, int, RawNode]]
+) -> List[Tuple[DFAS, int, RawNode]]:
+    """Nodeless stack copy."""
+    return [(dfa, label, DUMMY_NODE) for dfa, label, _ in stack]
+
+

 class Recorder:
     def __init__(self, parser: "Parser", ilabels: List[int], context: Context) -> None:
         self.parser = parser
 class Recorder:
     def __init__(self, parser: "Parser", ilabels: List[int], context: Context) -> None:
         self.parser = parser


@@ -53,8 +62,8 @@ class Recorder:
         self.context = context  # not really matter
 
         self._dead_ilabels: Set[int] = set()
         self.context = context  # not really matter
 
         self._dead_ilabels: Set[int] = set()

-        self._start_point = copy.deepcopy(self.parser.stack)
-        self._points = {ilabel: copy.deepcopy(self._start_point) for ilabel in ilabels}
+        self._start_point = self.parser.stack
+        self._points = {ilabel: stack_copy(self._start_point) for ilabel in ilabels}

 
     @property
     def ilabels(self) -> Set[int]:
 
     @property
     def ilabels(self) -> Set[int]:


@@ -62,17 +71,34 @@ class Recorder:
 
     @contextmanager
     def switch_to(self, ilabel: int) -> Iterator[None]:
 
     @contextmanager
     def switch_to(self, ilabel: int) -> Iterator[None]:

-        self.parser.stack = self._points[ilabel]
+        with self.backtrack():
+            self.parser.stack = self._points[ilabel]
+            try:
+                yield
+            except ParseError:
+                self._dead_ilabels.add(ilabel)
+            finally:
+                self.parser.stack = self._start_point
+
+    @contextmanager
+    def backtrack(self) -> Iterator[None]:
+        """
+        Use the node-level invariant ones for basic parsing operations (push/pop/shift).
+        These still will operate on the stack; but they won't create any new nodes, or
+        modify the contents of any other existing nodes.
+
+        This saves us a ton of time when we are backtracking, since we
+        want to restore to the initial state as quick as possible, which
+        can only be done by having as little mutatations as possible.
+        """
+        is_backtracking = self.parser.is_backtracking

         try:
         try:

+            self.parser.is_backtracking = True

             yield
             yield

-        except ParseError:
-            self._dead_ilabels.add(ilabel)

         finally:
         finally:

-            self.parser.stack = self._start_point
+            self.parser.is_backtracking = is_backtracking

 
 

-    def add_token(
-        self, tok_type: int, tok_val: Optional[Text], raw: bool = False
-    ) -> None:
+    def add_token(self, tok_type: int, tok_val: str, raw: bool = False) -> None:

         func: Callable[..., Any]
         if raw:
             func = self.parser._addtoken
         func: Callable[..., Any]
         if raw:
             func = self.parser._addtoken


@@ -86,9 +112,7 @@ class Recorder:
                     args.insert(0, ilabel)
                 func(*args)
 
                     args.insert(0, ilabel)
                 func(*args)
 

-    def determine_route(
-        self, value: Optional[Text] = None, force: bool = False
-    ) -> Optional[int]:
+    def determine_route(self, value: Optional[str] = None, force: bool = False) -> Optional[int]:

         alive_ilabels = self.ilabels
         if len(alive_ilabels) == 0:
             *_, most_successful_ilabel = self._dead_ilabels
         alive_ilabels = self.ilabels
         if len(alive_ilabels) == 0:
             *_, most_successful_ilabel = self._dead_ilabels


@@ -105,10 +129,10 @@ class ParseError(Exception):
     """Exception to signal the parser is stuck."""
 
     def __init__(
     """Exception to signal the parser is stuck."""
 
     def __init__(

-        self, msg: Text, type: Optional[int], value: Optional[Text], context: Context
+        self, msg: str, type: Optional[int], value: Optional[str], context: Context

     ) -> None:
         Exception.__init__(
     ) -> None:
         Exception.__init__(

-            self, "%s: type=%r, value=%r, context=%r" % (msg, type, value, context)
+            self, f"{msg}: type={type!r}, value={value!r}, context={context!r}"

         )
         self.msg = msg
         self.type = type
         )
         self.msg = msg
         self.type = type


@@ -116,7 +140,7 @@ class ParseError(Exception):
         self.context = context
 
 
         self.context = context
 
 

-class Parser(object):
+class Parser:

     """Parser engine.
 
     The proper usage sequence is:
     """Parser engine.
 
     The proper usage sequence is:


@@ -164,6 +188,11 @@ class Parser(object):
         to be converted.  The syntax tree is converted from the bottom
         up.
 
         to be converted.  The syntax tree is converted from the bottom
         up.
 

+        **post-note: the convert argument is ignored since for Black's
+        usage, convert will always be blib2to3.pytree.convert. Allowing
+        this to be dynamic hurts mypyc's ability to use early binding.
+        These docs are left for historical and informational value.
+

         A concrete syntax tree node is a (type, value, context, nodes)
         tuple, where type is the node type (a token or symbol number),
         value is None for symbols and a string for tokens, context is
         A concrete syntax tree node is a (type, value, context, nodes)
         tuple, where type is the node type (a token or symbol number),
         value is None for symbols and a string for tokens, context is


@@ -176,7 +205,9 @@ class Parser(object):
 
         """
         self.grammar = grammar
 
         """
         self.grammar = grammar

+        # See note in docstring above. TL;DR this is ignored.

         self.convert = convert or lam_sub
         self.convert = convert or lam_sub

+        self.is_backtracking = False

 
     def setup(self, proxy: "TokenProxy", start: Optional[int] = None) -> None:
         """Prepare for parsing.
 
     def setup(self, proxy: "TokenProxy", start: Optional[int] = None) -> None:
         """Prepare for parsing.


@@ -203,7 +234,7 @@ class Parser(object):
         self.used_names: Set[str] = set()
         self.proxy = proxy
 
         self.used_names: Set[str] = set()
         self.proxy = proxy
 

-    def addtoken(self, type: int, value: Optional[Text], context: Context) -> bool:
+    def addtoken(self, type: int, value: str, context: Context) -> bool:

         """Add a token; return True iff this is the end of the program."""
         # Map from token to label
         ilabels = self.classify(type, value, context)
         """Add a token; return True iff this is the end of the program."""
         # Map from token to label
         ilabels = self.classify(type, value, context)


@@ -236,8 +267,12 @@ class Parser(object):
                     break
 
                 next_token_type, next_token_value, *_ = proxy.eat(counter)
                     break
 
                 next_token_type, next_token_value, *_ = proxy.eat(counter)

+                if next_token_type in (tokenize.COMMENT, tokenize.NL):
+                    counter += 1
+                    continue
+

                 if next_token_type == tokenize.OP:
                 if next_token_type == tokenize.OP:

-                    next_token_type = grammar.opmap[cast(str, next_token_value)]
+                    next_token_type = grammar.opmap[next_token_value]

 
                 recorder.add_token(next_token_type, next_token_value)
                 counter += 1
 
                 recorder.add_token(next_token_type, next_token_value)
                 counter += 1


@@ -247,9 +282,7 @@ class Parser(object):
 
         return self._addtoken(ilabel, type, value, context)
 
 
         return self._addtoken(ilabel, type, value, context)
 

-    def _addtoken(
-        self, ilabel: int, type: int, value: Optional[Text], context: Context
-    ) -> bool:
+    def _addtoken(self, ilabel: int, type: int, value: str, context: Context) -> bool:

         # Loop until the token is shifted; may raise exceptions
         while True:
             dfa, state, node = self.stack[-1]
         # Loop until the token is shifted; may raise exceptions
         while True:
             dfa, state, node = self.stack[-1]


@@ -257,10 +290,18 @@ class Parser(object):
             arcs = states[state]
             # Look for a state with this label
             for i, newstate in arcs:
             arcs = states[state]
             # Look for a state with this label
             for i, newstate in arcs:

-                t, v = self.grammar.labels[i]
-                if ilabel == i:
+                t = self.grammar.labels[i][0]
+                if t >= 256:
+                    # See if it's a symbol and if we're in its first set
+                    itsdfa = self.grammar.dfas[t]
+                    itsstates, itsfirst = itsdfa
+                    if ilabel in itsfirst:
+                        # Push a symbol
+                        self.push(t, itsdfa, newstate, context)
+                        break  # To continue the outer while loop
+
+                elif ilabel == i:

                     # Look it up in the list of labels
                     # Look it up in the list of labels

-                    assert t < 256

                     # Shift a token; we're done with it
                     self.shift(type, value, newstate, context)
                     # Pop while we are in an accept-only state
                     # Shift a token; we're done with it
                     self.shift(type, value, newstate, context)
                     # Pop while we are in an accept-only state


@@ -274,14 +315,7 @@ class Parser(object):
                         states, first = dfa
                     # Done with this token
                     return False
                         states, first = dfa
                     # Done with this token
                     return False

-                elif t >= 256:
-                    # See if it's a symbol and if we're in its first set
-                    itsdfa = self.grammar.dfas[t]
-                    itsstates, itsfirst = itsdfa
-                    if ilabel in itsfirst:
-                        # Push a symbol
-                        self.push(t, self.grammar.dfas[t], newstate, context)
-                        break  # To continue the outer while loop
+

             else:
                 if (0, state) in arcs:
                     # An accepting state, pop it and try something else
             else:
                 if (0, state) in arcs:
                     # An accepting state, pop it and try something else


@@ -293,14 +327,13 @@ class Parser(object):
                     # No success finding a transition
                     raise ParseError("bad input", type, value, context)
 
                     # No success finding a transition
                     raise ParseError("bad input", type, value, context)
 

-    def classify(self, type: int, value: Optional[Text], context: Context) -> List[int]:
+    def classify(self, type: int, value: str, context: Context) -> List[int]:

         """Turn a token into a label.  (Internal)
 
         Depending on whether the value is a soft-keyword or not,
         this function may return multiple labels to choose from."""
         if type == token.NAME:
             # Keep a listing of all used names
         """Turn a token into a label.  (Internal)
 
         Depending on whether the value is a soft-keyword or not,
         this function may return multiple labels to choose from."""
         if type == token.NAME:
             # Keep a listing of all used names

-            assert value is not None

             self.used_names.add(value)
             # Check for reserved words
             if value in self.grammar.keywords:
             self.used_names.add(value)
             # Check for reserved words
             if value in self.grammar.keywords:


@@ -317,32 +350,38 @@ class Parser(object):
             raise ParseError("bad token", type, value, context)
         return [ilabel]
 
             raise ParseError("bad token", type, value, context)
         return [ilabel]
 

-    def shift(
-        self, type: int, value: Optional[Text], newstate: int, context: Context
-    ) -> None:
+    def shift(self, type: int, value: str, newstate: int, context: Context) -> None:

         """Shift a token.  (Internal)"""
         """Shift a token.  (Internal)"""

-        dfa, state, node = self.stack[-1]
-        assert value is not None
-        assert context is not None
-        rawnode: RawNode = (type, value, context, None)
-        newnode = self.convert(self.grammar, rawnode)
-        if newnode is not None:
+        if self.is_backtracking:
+            dfa, state, _ = self.stack[-1]
+            self.stack[-1] = (dfa, newstate, DUMMY_NODE)
+        else:
+            dfa, state, node = self.stack[-1]
+            rawnode: RawNode = (type, value, context, None)
+            newnode = convert(self.grammar, rawnode)

             assert node[-1] is not None
             node[-1].append(newnode)
             assert node[-1] is not None
             node[-1].append(newnode)

-        self.stack[-1] = (dfa, newstate, node)
+            self.stack[-1] = (dfa, newstate, node)

 
     def push(self, type: int, newdfa: DFAS, newstate: int, context: Context) -> None:
         """Push a nonterminal.  (Internal)"""
 
     def push(self, type: int, newdfa: DFAS, newstate: int, context: Context) -> None:
         """Push a nonterminal.  (Internal)"""

-        dfa, state, node = self.stack[-1]
-        newnode: RawNode = (type, None, context, [])
-        self.stack[-1] = (dfa, newstate, node)
-        self.stack.append((newdfa, 0, newnode))
+        if self.is_backtracking:
+            dfa, state, _ = self.stack[-1]
+            self.stack[-1] = (dfa, newstate, DUMMY_NODE)
+            self.stack.append((newdfa, 0, DUMMY_NODE))
+        else:
+            dfa, state, node = self.stack[-1]
+            newnode: RawNode = (type, None, context, [])
+            self.stack[-1] = (dfa, newstate, node)
+            self.stack.append((newdfa, 0, newnode))

 
     def pop(self) -> None:
         """Pop a nonterminal.  (Internal)"""
 
     def pop(self) -> None:
         """Pop a nonterminal.  (Internal)"""

-        popdfa, popstate, popnode = self.stack.pop()
-        newnode = self.convert(self.grammar, popnode)
-        if newnode is not None:
+        if self.is_backtracking:
+            self.stack.pop()
+        else:
+            popdfa, popstate, popnode = self.stack.pop()
+            newnode = convert(self.grammar, popnode)

             if self.stack:
                 dfa, state, node = self.stack[-1]
                 assert node[-1] is not None
             if self.stack:
                 dfa, state, node = self.stack[-1]
                 assert node[-1] is not None