# Copyright 2016 Grist Labs, Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import abc
import ast
import bisect
import sys
import token
from ast import Module
from typing import Iterable, Iterator, List, Optional, Tuple, Any, cast, TYPE_CHECKING
import six
from six.moves import xrange # pylint: disable=redefined-builtin
from .line_numbers import LineNumbers
from .util import (
Token, match_token, is_non_coding_token, patched_generate_tokens, last_stmt,
annotate_fstring_nodes, generate_tokens, is_module, is_stmt
)
if TYPE_CHECKING: # pragma: no cover
from .util import AstNode, TokenInfo
class ASTTextBase(six.with_metaclass(abc.ABCMeta, object)):
def __init__(self, source_text, filename):
# type: (Any, str) -> None
# FIXME: Strictly, the type of source_text is one of the six string types, but hard to specify with mypy given
# https://mypy.readthedocs.io/en/stable/common_issues.html#variables-vs-type-aliases
self._filename = filename
# Decode source after parsing to let Python 2 handle coding declarations.
# (If the encoding was not utf-8 compatible, then even if it parses correctly,
# we'll fail with a unicode error here.)
source_text = six.ensure_text(source_text)
self._text = source_text
self._line_numbers = LineNumbers(source_text)
@abc.abstractmethod
def get_text_positions(self, node, padded):
# type: (AstNode, bool) -> Tuple[Tuple[int, int], Tuple[int, int]]
"""
Returns two ``(lineno, col_offset)`` tuples for the start and end of the given node.
If the positions can't be determined, or the nodes don't correspond to any particular text,
returns ``(1, 0)`` for both.
``padded`` corresponds to the ``padded`` argument to ``ast.get_source_segment()``.
This means that if ``padded`` is True, the start position will be adjusted to include
leading whitespace if ``node`` is a multiline statement.
"""
raise NotImplementedError # pragma: no cover
def get_text_range(self, node, padded=True):
# type: (AstNode, bool) -> Tuple[int, int]
"""
Returns the (startpos, endpos) positions in source text corresponding to the given node.
Returns (0, 0) for nodes (like `Load`) that don't correspond to any particular text.
See ``get_text_positions()`` for details on the ``padded`` argument.
"""
start, end = self.get_text_positions(node, padded)
return (
self._line_numbers.line_to_offset(*start),
self._line_numbers.line_to_offset(*end),
)
def get_text(self, node, padded=True):
# type: (AstNode, bool) -> str
"""
Returns the text corresponding to the given node.
Returns '' for nodes (like `Load`) that don't correspond to any particular text.
See ``get_text_positions()`` for details on the ``padded`` argument.
"""
start, end = self.get_text_range(node, padded)
return self._text[start: end]
[docs]class ASTTokens(ASTTextBase, object):
"""
ASTTokens maintains the text of Python code in several forms: as a string, as line numbers, and
as tokens, and is used to mark and access token and position information.
``source_text`` must be a unicode or UTF8-encoded string. If you pass in UTF8 bytes, remember
that all offsets you'll get are to the unicode text, which is available as the ``.text``
property.
If ``parse`` is set, the ``source_text`` will be parsed with ``ast.parse()``, and the resulting
tree marked with token info and made available as the ``.tree`` property.
If ``tree`` is given, it will be marked and made available as the ``.tree`` property. In
addition to the trees produced by the ``ast`` module, ASTTokens will also mark trees produced
using ``astroid`` library <https://www.astroid.org>.
If only ``source_text`` is given, you may use ``.mark_tokens(tree)`` to mark the nodes of an AST
tree created separately.
"""
def __init__(self, source_text, parse=False, tree=None, filename='<unknown>', tokens=None):
# type: (Any, bool, Optional[Module], str, Iterable[TokenInfo]) -> None
# FIXME: Strictly, the type of source_text is one of the six string types, but hard to specify with mypy given
# https://mypy.readthedocs.io/en/stable/common_issues.html#variables-vs-type-aliases
super(ASTTokens, self).__init__(source_text, filename)
self._tree = ast.parse(source_text, filename) if parse else tree
# Tokenize the code.
if tokens is None:
tokens = generate_tokens(self._text)
self._tokens = list(self._translate_tokens(tokens))
# Extract the start positions of all tokens, so that we can quickly map positions to tokens.
self._token_offsets = [tok.startpos for tok in self._tokens]
if self._tree:
self.mark_tokens(self._tree)
[docs] def mark_tokens(self, root_node):
# type: (Module) -> None
"""
Given the root of the AST or Astroid tree produced from source_text, visits all nodes marking
them with token and position information by adding ``.first_token`` and
``.last_token``attributes. This is done automatically in the constructor when ``parse`` or
``tree`` arguments are set, but may be used manually with a separate AST or Astroid tree.
"""
# The hard work of this class is done by MarkTokens
from .mark_tokens import MarkTokens # to avoid import loops
MarkTokens(self).visit_tree(root_node)
def _translate_tokens(self, original_tokens):
# type: (Iterable[TokenInfo]) -> Iterator[Token]
"""
Translates the given standard library tokens into our own representation.
"""
for index, tok in enumerate(patched_generate_tokens(original_tokens)):
tok_type, tok_str, start, end, line = tok
yield Token(tok_type, tok_str, start, end, line, index,
self._line_numbers.line_to_offset(start[0], start[1]),
self._line_numbers.line_to_offset(end[0], end[1]))
@property
def text(self):
# type: () -> str
"""The source code passed into the constructor."""
return self._text
@property
def tokens(self):
# type: () -> List[Token]
"""The list of tokens corresponding to the source code from the constructor."""
return self._tokens
@property
def tree(self):
# type: () -> Optional[Module]
"""The root of the AST tree passed into the constructor or parsed from the source code."""
return self._tree
@property
def filename(self):
# type: () -> str
"""The filename that was parsed"""
return self._filename
[docs] def get_token_from_offset(self, offset):
# type: (int) -> Token
"""
Returns the token containing the given character offset (0-based position in source text),
or the preceeding token if the position is between tokens.
"""
return self._tokens[bisect.bisect(self._token_offsets, offset) - 1]
[docs] def get_token(self, lineno, col_offset):
# type: (int, int) -> Token
"""
Returns the token containing the given (lineno, col_offset) position, or the preceeding token
if the position is between tokens.
"""
# TODO: add test for multibyte unicode. We need to translate offsets from ast module (which
# are in utf8) to offsets into the unicode text. tokenize module seems to use unicode offsets
# but isn't explicit.
return self.get_token_from_offset(self._line_numbers.line_to_offset(lineno, col_offset))
[docs] def get_token_from_utf8(self, lineno, col_offset):
# type: (int, int) -> Token
"""
Same as get_token(), but interprets col_offset as a UTF8 offset, which is what `ast` uses.
"""
return self.get_token(lineno, self._line_numbers.from_utf8_col(lineno, col_offset))
[docs] def next_token(self, tok, include_extra=False):
# type: (Token, bool) -> Token
"""
Returns the next token after the given one. If include_extra is True, includes non-coding
tokens from the tokenize module, such as NL and COMMENT.
"""
i = tok.index + 1
if not include_extra:
while is_non_coding_token(self._tokens[i].type):
i += 1
return self._tokens[i]
[docs] def prev_token(self, tok, include_extra=False):
# type: (Token, bool) -> Token
"""
Returns the previous token before the given one. If include_extra is True, includes non-coding
tokens from the tokenize module, such as NL and COMMENT.
"""
i = tok.index - 1
if not include_extra:
while is_non_coding_token(self._tokens[i].type):
i -= 1
return self._tokens[i]
[docs] def find_token(self, start_token, tok_type, tok_str=None, reverse=False):
# type: (Token, int, Optional[str], bool) -> Token
"""
Looks for the first token, starting at start_token, that matches tok_type and, if given, the
token string. Searches backwards if reverse is True. Returns ENDMARKER token if not found (you
can check it with `token.ISEOF(t.type)`).
"""
t = start_token
advance = self.prev_token if reverse else self.next_token
while not match_token(t, tok_type, tok_str) and not token.ISEOF(t.type):
t = advance(t, include_extra=True)
return t
[docs] def token_range(self,
first_token, # type: Token
last_token, # type: Token
include_extra=False, # type: bool
):
# type: (...) -> Iterator[Token]
"""
Yields all tokens in order from first_token through and including last_token. If
include_extra is True, includes non-coding tokens such as tokenize.NL and .COMMENT.
"""
for i in xrange(first_token.index, last_token.index + 1):
if include_extra or not is_non_coding_token(self._tokens[i].type):
yield self._tokens[i]
[docs] def get_tokens(self, node, include_extra=False):
# type: (AstNode, bool) -> Iterator[Token]
"""
Yields all tokens making up the given node. If include_extra is True, includes non-coding
tokens such as tokenize.NL and .COMMENT.
"""
return self.token_range(node.first_token, node.last_token, include_extra=include_extra)
[docs] def get_text_positions(self, node, padded):
# type: (AstNode, bool) -> Tuple[Tuple[int, int], Tuple[int, int]]
"""
Returns two ``(lineno, col_offset)`` tuples for the start and end of the given node.
If the positions can't be determined, or the nodes don't correspond to any particular text,
returns ``(1, 0)`` for both.
``padded`` corresponds to the ``padded`` argument to ``ast.get_source_segment()``.
This means that if ``padded`` is True, the start position will be adjusted to include
leading whitespace if ``node`` is a multiline statement.
"""
if not hasattr(node, 'first_token'):
return (1, 0), (1, 0)
start = node.first_token.start
end = node.last_token.end
if padded and any(match_token(t, token.NEWLINE) for t in self.get_tokens(node)):
# Set col_offset to 0 to include leading indentation for multiline statements.
start = (start[0], 0)
return start, end
[docs]class ASTText(ASTTextBase, object):
"""
Supports the same ``get_text*`` methods as ``ASTTokens``,
but uses the AST to determine the text positions instead of tokens.
This is faster than ``ASTTokens`` as it requires less setup work.
It also (sometimes) supports nodes inside f-strings, which ``ASTTokens`` doesn't.
Some node types and/or Python versions are not supported.
In these cases the ``get_text*`` methods will fall back to using ``ASTTokens``
which incurs the usual setup cost the first time.
If you want to avoid this, check ``supports_tokenless(node)`` before calling ``get_text*`` methods.
"""
def __init__(self, source_text, tree=None, filename='<unknown>'):
# type: (Any, Optional[Module], str) -> None
# FIXME: Strictly, the type of source_text is one of the six string types, but hard to specify with mypy given
# https://mypy.readthedocs.io/en/stable/common_issues.html#variables-vs-type-aliases
super(ASTText, self).__init__(source_text, filename)
self._tree = tree
if self._tree is not None:
annotate_fstring_nodes(self._tree)
self._asttokens = None # type: Optional[ASTTokens]
@property
def tree(self):
# type: () -> Module
if self._tree is None:
self._tree = ast.parse(self._text, self._filename)
annotate_fstring_nodes(self._tree)
return self._tree
@property
def asttokens(self):
# type: () -> ASTTokens
if self._asttokens is None:
self._asttokens = ASTTokens(
self._text,
tree=self.tree,
filename=self._filename,
)
return self._asttokens
def _get_text_positions_tokenless(self, node, padded):
# type: (AstNode, bool) -> Tuple[Tuple[int, int], Tuple[int, int]]
"""
Version of ``get_text_positions()`` that doesn't use tokens.
"""
if sys.version_info[:2] < (3, 8): # pragma: no cover
# This is just for mpypy
raise AssertionError("This method should only be called internally after checking supports_tokenless()")
if is_module(node):
# Modules don't have position info, so just return the range of the whole text.
# The token-using method does something different, but its behavior seems weird and inconsistent.
# For example, in a file with only comments, it only returns the first line.
# It's hard to imagine a case when this matters.
return (1, 0), self._line_numbers.offset_to_line(len(self._text))
if getattr(node, 'lineno', None) is None:
return (1, 0), (1, 0)
assert node # tell mypy that node is not None, which we allowed up to here for compatibility
decorators = getattr(node, 'decorator_list', [])
if not decorators:
# Astroid uses node.decorators.nodes instead of node.decorator_list.
decorators_node = getattr(node, 'decorators', None)
decorators = getattr(decorators_node, 'nodes', [])
if decorators:
# Function/Class definition nodes are marked by AST as starting at def/class,
# not the first decorator. This doesn't match the token-using behavior,
# or inspect.getsource(), and just seems weird.
start_node = decorators[0]
else:
start_node = node
start_lineno = start_node.lineno
end_node = last_stmt(node)
# Include leading indentation for multiline statements.
# This doesn't mean simple statements that happen to be on multiple lines,
# but compound statements where inner indentation matters.
# So we don't just compare node.lineno and node.end_lineno,
# we check for a contained statement starting on a different line.
if padded and (
start_lineno != end_node.lineno
or (
# Astroid docstrings aren't treated as separate statements.
# So to handle function/class definitions with a docstring but no other body,
# we just check that the node is a statement with a docstring
# and spanning multiple lines in the simple, literal sense.
start_lineno != node.end_lineno
and getattr(node, "doc_node", None)
and is_stmt(node)
)
):
start_col_offset = 0
else:
start_col_offset = self._line_numbers.from_utf8_col(start_lineno, start_node.col_offset)
start = (start_lineno, start_col_offset)
# To match the token-using behaviour, we exclude trailing semicolons and comments.
# This means that for blocks containing multiple statements, we have to use the last one
# instead of the actual node for end_lineno and end_col_offset.
end_lineno = cast(int, end_node.end_lineno)
end_col_offset = cast(int, end_node.end_col_offset)
end_col_offset = self._line_numbers.from_utf8_col(end_lineno, end_col_offset)
end = (end_lineno, end_col_offset)
return start, end
[docs] def get_text_positions(self, node, padded):
# type: (AstNode, bool) -> Tuple[Tuple[int, int], Tuple[int, int]]
"""
Returns two ``(lineno, col_offset)`` tuples for the start and end of the given node.
If the positions can't be determined, or the nodes don't correspond to any particular text,
returns ``(1, 0)`` for both.
``padded`` corresponds to the ``padded`` argument to ``ast.get_source_segment()``.
This means that if ``padded`` is True, the start position will be adjusted to include
leading whitespace if ``node`` is a multiline statement.
"""
if getattr(node, "_broken_positions", None):
# This node was marked in util.annotate_fstring_nodes as having untrustworthy lineno/col_offset.
return (1, 0), (1, 0)
if supports_tokenless(node):
return self._get_text_positions_tokenless(node, padded)
return self.asttokens.get_text_positions(node, padded)
# Node types that _get_text_positions_tokenless doesn't support. Only relevant for Python 3.8+.
_unsupported_tokenless_types = () # type: Tuple[str, ...]
if sys.version_info[:2] >= (3, 8):
# no lineno
_unsupported_tokenless_types += ("arguments", "Arguments", "withitem")
if sys.version_info[:2] == (3, 8):
# _get_text_positions_tokenless works incorrectly for these types due to bugs in Python 3.8.
_unsupported_tokenless_types += ("arg", "Starred")
# no lineno in 3.8
_unsupported_tokenless_types += ("Slice", "ExtSlice", "Index", "keyword")
def supports_tokenless(node=None):
# type: (Any) -> bool
"""
Returns True if the Python version and the node (if given) are supported by
the ``get_text*`` methods of ``ASTText`` without falling back to ``ASTTokens``.
See ``ASTText`` for why this matters.
The following cases are not supported:
- Python 3.7 and earlier
- PyPy
- ``ast.arguments`` / ``astroid.Arguments``
- ``ast.withitem``
- ``astroid.Comprehension``
- ``astroid.AssignName`` inside ``astroid.Arguments`` or ``astroid.ExceptHandler``
- The following nodes in Python 3.8 only:
- ``ast.arg``
- ``ast.Starred``
- ``ast.Slice``
- ``ast.ExtSlice``
- ``ast.Index``
- ``ast.keyword``
"""
return (
type(node).__name__ not in _unsupported_tokenless_types
and not (
# astroid nodes
not isinstance(node, ast.AST) and node is not None and (
(
type(node).__name__ == "AssignName"
and type(node.parent).__name__ in ("Arguments", "ExceptHandler")
)
)
)
and sys.version_info[:2] >= (3, 8)
and 'pypy' not in sys.version.lower()
)