cpython/Lib/idlelib/hyperparser.py

314 lines
13 KiB
Python
Raw Normal View History

"""Provide advanced parsing abilities for ParenMatch and other extensions.
HyperParser uses PyParser. PyParser mostly gives information on the
proper indentation of code. HyperParser gives additional information on
the structure of code.
"""
import string
from keyword import iskeyword
from idlelib import pyparse
# all ASCII chars that may be in an identifier
_ASCII_ID_CHARS = frozenset(string.ascii_letters + string.digits + "_")
# all ASCII chars that may be the first char of an identifier
_ASCII_ID_FIRST_CHARS = frozenset(string.ascii_letters + "_")
# lookup table for whether 7-bit ASCII chars are valid in a Python identifier
_IS_ASCII_ID_CHAR = [(chr(x) in _ASCII_ID_CHARS) for x in range(128)]
# lookup table for whether 7-bit ASCII chars are valid as the first
# char in a Python identifier
_IS_ASCII_ID_FIRST_CHAR = \
[(chr(x) in _ASCII_ID_FIRST_CHARS) for x in range(128)]
class HyperParser:
def __init__(self, editwin, index):
"To initialize, analyze the surroundings of the given index."
self.editwin = editwin
self.text = text = editwin.text
parser = pyparse.Parser(editwin.indentwidth, editwin.tabwidth)
def index2line(index):
return int(float(index))
lno = index2line(text.index(index))
if not editwin.context_use_ps1:
for context in editwin.num_context_lines:
startat = max(lno - context, 1)
2006-08-24 23:59:59 -03:00
startatindex = repr(startat) + ".0"
stopatindex = "%d.end" % lno
# We add the newline because PyParse requires a newline
# at end. We add a space so that index won't be at end
# of line, so that its status will be the same as the
# char before it, if should.
parser.set_str(text.get(startatindex, stopatindex)+' \n')
bod = parser.find_good_parse_start(
editwin._build_char_in_string_func(startatindex))
if bod is not None or startat == 1:
break
parser.set_lo(bod or 0)
else:
r = text.tag_prevrange("console", index)
if r:
startatindex = r[1]
else:
startatindex = "1.0"
stopatindex = "%d.end" % lno
# We add the newline because PyParse requires it. We add a
# space so that index won't be at end of line, so that its
# status will be the same as the char before it, if should.
parser.set_str(text.get(startatindex, stopatindex)+' \n')
parser.set_lo(0)
# We want what the parser has, minus the last newline and space.
self.rawtext = parser.str[:-2]
# Parser.str apparently preserves the statement we are in, so
# that stopatindex can be used to synchronize the string with
# the text box indices.
self.stopatindex = stopatindex
self.bracketing = parser.get_last_stmt_bracketing()
# find which pairs of bracketing are openers. These always
# correspond to a character of rawtext.
self.isopener = [i>0 and self.bracketing[i][1] >
self.bracketing[i-1][1]
for i in range(len(self.bracketing))]
self.set_index(index)
def set_index(self, index):
"""Set the index to which the functions relate.
The index must be in the same statement.
"""
indexinrawtext = (len(self.rawtext) -
len(self.text.get(index, self.stopatindex)))
if indexinrawtext < 0:
raise ValueError("Index %s precedes the analyzed statement"
% index)
self.indexinrawtext = indexinrawtext
# find the rightmost bracket to which index belongs
self.indexbracket = 0
while (self.indexbracket < len(self.bracketing)-1 and
self.bracketing[self.indexbracket+1][0] < self.indexinrawtext):
self.indexbracket += 1
if (self.indexbracket < len(self.bracketing)-1 and
self.bracketing[self.indexbracket+1][0] == self.indexinrawtext and
not self.isopener[self.indexbracket+1]):
self.indexbracket += 1
def is_in_string(self):
"""Is the index given to the HyperParser in a string?"""
# The bracket to which we belong should be an opener.
# If it's an opener, it has to have a character.
return (self.isopener[self.indexbracket] and
self.rawtext[self.bracketing[self.indexbracket][0]]
in ('"', "'"))
def is_in_code(self):
"""Is the index given to the HyperParser in normal code?"""
return (not self.isopener[self.indexbracket] or
self.rawtext[self.bracketing[self.indexbracket][0]]
not in ('#', '"', "'"))
def get_surrounding_brackets(self, openers='([{', mustclose=False):
"""Return bracket indexes or None.
If the index given to the HyperParser is surrounded by a
bracket defined in openers (or at least has one before it),
return the indices of the opening bracket and the closing
bracket (or the end of line, whichever comes first).
If it is not surrounded by brackets, or the end of line comes
before the closing bracket and mustclose is True, returns None.
"""
2014-06-16 03:40:24 -03:00
bracketinglevel = self.bracketing[self.indexbracket][1]
before = self.indexbracket
while (not self.isopener[before] or
self.rawtext[self.bracketing[before][0]] not in openers or
self.bracketing[before][1] > bracketinglevel):
before -= 1
if before < 0:
return None
bracketinglevel = min(bracketinglevel, self.bracketing[before][1])
after = self.indexbracket + 1
while (after < len(self.bracketing) and
self.bracketing[after][1] >= bracketinglevel):
after += 1
beforeindex = self.text.index("%s-%dc" %
(self.stopatindex, len(self.rawtext)-self.bracketing[before][0]))
if (after >= len(self.bracketing) or
self.bracketing[after][0] > len(self.rawtext)):
if mustclose:
return None
afterindex = self.stopatindex
else:
# We are after a real char, so it is a ')' and we give the
# index before it.
afterindex = self.text.index(
"%s-%dc" % (self.stopatindex,
len(self.rawtext)-(self.bracketing[after][0]-1)))
return beforeindex, afterindex
# the set of built-in identifiers which are also keywords,
# i.e. keyword.iskeyword() returns True for them
_ID_KEYWORDS = frozenset({"True", "False", "None"})
@classmethod
def _eat_identifier(cls, str, limit, pos):
"""Given a string and pos, return the number of chars in the
identifier which ends at pos, or 0 if there is no such one.
This ignores non-identifier eywords are not identifiers.
"""
is_ascii_id_char = _IS_ASCII_ID_CHAR
# Start at the end (pos) and work backwards.
i = pos
# Go backwards as long as the characters are valid ASCII
# identifier characters. This is an optimization, since it
# is faster in the common case where most of the characters
# are ASCII.
while i > limit and (
ord(str[i - 1]) < 128 and
is_ascii_id_char[ord(str[i - 1])]
):
i -= 1
# If the above loop ended due to reaching a non-ASCII
# character, continue going backwards using the most generic
# test for whether a string contains only valid identifier
# characters.
if i > limit and ord(str[i - 1]) >= 128:
while i - 4 >= limit and ('a' + str[i - 4:pos]).isidentifier():
i -= 4
if i - 2 >= limit and ('a' + str[i - 2:pos]).isidentifier():
i -= 2
if i - 1 >= limit and ('a' + str[i - 1:pos]).isidentifier():
i -= 1
# The identifier candidate starts here. If it isn't a valid
# identifier, don't eat anything. At this point that is only
# possible if the first character isn't a valid first
# character for an identifier.
if not str[i:pos].isidentifier():
return 0
elif i < pos:
# All characters in str[i:pos] are valid ASCII identifier
# characters, so it is enough to check that the first is
# valid as the first character of an identifier.
if not _IS_ASCII_ID_FIRST_CHAR[ord(str[i])]:
return 0
# All keywords are valid identifiers, but should not be
# considered identifiers here, except for True, False and None.
if i < pos and (
iskeyword(str[i:pos]) and
str[i:pos] not in cls._ID_KEYWORDS
):
return 0
return pos - i
# This string includes all chars that may be in a white space
_whitespace_chars = " \t\n\\"
def get_expression(self):
"""Return a string with the Python expression which ends at the
given index, which is empty if there is no real one.
"""
if not self.is_in_code():
raise ValueError("get_expression should only be called"
"if index is inside a code.")
rawtext = self.rawtext
bracketing = self.bracketing
brck_index = self.indexbracket
brck_limit = bracketing[brck_index][0]
pos = self.indexinrawtext
last_identifier_pos = pos
postdot_phase = True
while 1:
# Eat whitespaces, comments, and if postdot_phase is False - a dot
while 1:
if pos>brck_limit and rawtext[pos-1] in self._whitespace_chars:
# Eat a whitespace
pos -= 1
elif (not postdot_phase and
pos > brck_limit and rawtext[pos-1] == '.'):
# Eat a dot
pos -= 1
postdot_phase = True
# The next line will fail if we are *inside* a comment,
# but we shouldn't be.
elif (pos == brck_limit and brck_index > 0 and
rawtext[bracketing[brck_index-1][0]] == '#'):
# Eat a comment
brck_index -= 2
brck_limit = bracketing[brck_index][0]
pos = bracketing[brck_index+1][0]
else:
# If we didn't eat anything, quit.
break
if not postdot_phase:
# We didn't find a dot, so the expression end at the
# last identifier pos.
break
ret = self._eat_identifier(rawtext, brck_limit, pos)
if ret:
# There is an identifier to eat
pos = pos - ret
last_identifier_pos = pos
# Now, to continue the search, we must find a dot.
postdot_phase = False
# (the loop continues now)
elif pos == brck_limit:
# We are at a bracketing limit. If it is a closing
# bracket, eat the bracket, otherwise, stop the search.
level = bracketing[brck_index][1]
while brck_index > 0 and bracketing[brck_index-1][1] > level:
brck_index -= 1
if bracketing[brck_index][0] == brck_limit:
# We were not at the end of a closing bracket
break
pos = bracketing[brck_index][0]
brck_index -= 1
brck_limit = bracketing[brck_index][0]
last_identifier_pos = pos
if rawtext[pos] in "([":
# [] and () may be used after an identifier, so we
# continue. postdot_phase is True, so we don't allow a dot.
pass
else:
# We can't continue after other types of brackets
if rawtext[pos] in "'\"":
# Scan a string prefix
while pos > 0 and rawtext[pos - 1] in "rRbBuU":
pos -= 1
last_identifier_pos = pos
break
else:
# We've found an operator or something.
break
return rawtext[last_identifier_pos:self.indexinrawtext]
if __name__ == '__main__':
import unittest
unittest.main('idlelib.idle_test.test_hyperparser', verbosity=2)