diff --git a/Tools/idle/PyParse.py b/Tools/idle/PyParse.py new file mode 100644 index 00000000000..e3b6e1f66d3 --- /dev/null +++ b/Tools/idle/PyParse.py @@ -0,0 +1,470 @@ +import string +import re +import sys + +# Reason last stmt is continued (or C_NONE if it's not). +C_NONE, C_BACKSLASH, C_STRING, C_BRACKET = range(4) + +if 0: # for throwaway debugging output + def dump(*stuff): + import sys + sys.__stdout__.write(string.join(map(str, stuff), " ") + "\n") + +# find a def or class stmt +_defclassre = re.compile(r""" + ^ + [ \t]* + (?: + def [ \t]+ [a-zA-Z_]\w* [ \t]* \( + | class [ \t]+ [a-zA-Z_]\w* [ \t]* + (?: \( .* \) )? + [ \t]* : + ) +""", re.VERBOSE | re.MULTILINE).search + +# match blank line or non-indenting comment line +_junkre = re.compile(r""" + [ \t]* + (?: \# [^ \t\n] .* )? + \n +""", re.VERBOSE).match + +# match any flavor of string; the terminating quote is optional +# so that we're robust in the face of incomplete program text +_match_stringre = re.compile(r""" + \""" [^"\\]* (?: + (?: \\. | "(?!"") ) + [^"\\]* + )* + (?: \""" )? + +| " [^"\\\n]* (?: \\. [^"\\\n]* )* "? + +| ''' [^'\\]* (?: + (?: \\. | '(?!'') ) + [^'\\]* + )* + (?: ''' )? + +| ' [^'\\\n]* (?: \\. [^'\\\n]* )* '? +""", re.VERBOSE | re.DOTALL).match + +# match a line that doesn't start with something interesting; +# used to skip junk lines when searching for the first element +# of a bracket structure +_not_itemre = re.compile(r""" + [ \t]* + [#\n\\] +""", re.VERBOSE).match + +# match start of stmts that should be followed by a dedent +_closere = re.compile(r""" + \s* + (?: return + | break + | continue + | raise + | pass + ) + \b +""", re.VERBOSE).match + +# Build translation table to map uninteresting chars to "x", open +# brackets to "(", and close brackets to ")". + +_tran = ['x'] * 256 +for ch in "({[": + _tran[ord(ch)] = '(' +for ch in ")}]": + _tran[ord(ch)] = ')' +for ch in "\"'\\\n#": + _tran[ord(ch)] = ch +_tran = string.join(_tran, '') +del ch + +class Parser: + + def __init__(self, indentwidth, tabwidth): + self.indentwidth = indentwidth + self.tabwidth = tabwidth + + def set_str(self, str): + assert len(str) == 0 or str[-1] == '\n' + self.str = str + self.study_level = 0 + + # Return index of start of last (probable!) def or class stmt, or + # None if none found. It's only probable because we can't know + # whether we're in a string without reparsing from the start of + # the file -- and that's too slow to bear. + # + # Ack, hack: in the shell window this kills us, because there's + # no way to tell the differences between output, >>> etc and + # user input. Indeed, IDLE's first output line makes the rest + # look like it's in an unclosed paren!: + # Python 1.5.2 (#0, Apr 13 1999, ... + + def find_last_def_or_class(self, _defclassre=_defclassre): + str, pos = self.str, None + i = 0 + while 1: + m = _defclassre(str, i) + if m: + pos, i = m.span() + else: + break + if pos is None: + # hack for shell window + ps1 = '\n' + sys.ps1 + i = string.rfind(str, ps1) + if i >= 0: + pos = i + len(ps1) + self.str = str[:pos-1] + '\n' + str[pos:] + return pos + + # Throw away the start of the string. Intended to be called with + # find_last_def_or_class's result. + + def set_lo(self, lo): + assert lo == 0 or self.str[lo-1] == '\n' + if lo > 0: + self.str = self.str[lo:] + + # As quickly as humanly possible , find the line numbers (0- + # based) of the non-continuation lines. + # Creates self.{stmts, continuation}. + + def _study1(self, _replace=string.replace, _find=string.find): + if self.study_level >= 1: + return + self.study_level = 1 + + # Map all uninteresting characters to "x", all open brackets + # to "(", all close brackets to ")", then collapse runs of + # uninteresting characters. This can cut the number of chars + # by a factor of 10-40, and so greatly speed the following loop. + str = self.str + str = string.translate(str, _tran) + str = _replace(str, 'xxxxxxxx', 'x') + str = _replace(str, 'xxxx', 'x') + str = _replace(str, 'xx', 'x') + str = _replace(str, 'xx', 'x') + str = _replace(str, '\nx', '\n') + # note that replacing x\n with \n would be incorrect, because + # x may be preceded by a backslash + + # March over the squashed version of the program, accumulating + # the line numbers of non-continued stmts, and determining + # whether & why the last stmt is a continuation. + continuation = C_NONE + level = lno = 0 # level is nesting level; lno is line number + self.stmts = stmts = [0] + push_stmt = stmts.append + i, n = 0, len(str) + while i < n: + ch = str[i] + # cases are checked in decreasing order of frequency + + if ch == 'x': + i = i+1 + continue + + if ch == '\n': + lno = lno + 1 + if level == 0: + push_stmt(lno) + # else we're in an unclosed bracket structure + i = i+1 + continue + + if ch == '(': + level = level + 1 + i = i+1 + continue + + if ch == ')': + if level: + level = level - 1 + # else the program is invalid, but we can't complain + i = i+1 + continue + + if ch == '"' or ch == "'": + # consume the string + quote = ch + if str[i:i+3] == quote * 3: + quote = quote * 3 + w = len(quote) + i = i+w + while i < n: + ch = str[i] + if ch == 'x': + i = i+1 + continue + + if str[i:i+w] == quote: + i = i+w + break + + if ch == '\n': + lno = lno + 1 + i = i+1 + if w == 1: + # unterminated single-quoted string + if level == 0: + push_stmt(lno) + break + continue + + if ch == '\\': + assert i+1 < n + if str[i+1] == '\n': + lno = lno + 1 + i = i+2 + continue + + # else comment char or paren inside string + i = i+1 + + else: + # didn't break out of the loop, so it's an + # unterminated triple-quoted string + assert w == 3 + continuation = C_STRING + continue + + if ch == '#': + # consume the comment + i = _find(str, '\n', i) + assert i >= 0 + continue + + assert ch == '\\' + assert i+1 < n + if str[i+1] == '\n': + lno = lno + 1 + if i+2 == n: + continuation = C_BACKSLASH + i = i+2 + + # Push the final line number as a sentinel value, regardless of + # whether it's continued. + if stmts[-1] != lno: + push_stmt(lno) + + # The last stmt may be continued for all 3 reasons. + # String continuation takes precedence over bracket + # continuation, which beats backslash continuation. + if continuation != C_STRING and level > 0: + continuation = C_BRACKET + self.continuation = continuation + + def get_continuation_type(self): + self._study1() + return self.continuation + + # study1 was sufficient to determine the continuation status, + # but doing more requires looking at every character. study2 + # does this for the last interesting statement in the block. + # Creates: + # self.stmt_start, stmt_end + # slice indices of last interesting stmt + # self.lastch + # last non-whitespace character before optional trailing + # comment + # self.lastopenbracketpos + # if continuation is C_BRACKET, index of last open bracket + + def _study2(self, _rfind=string.rfind, _find=string.find, + _ws=string.whitespace): + if self.study_level >= 2: + return + self._study1() + self.study_level = 2 + + self.lastch = "" + + # Set p and q to slice indices of last interesting stmt. + str, stmts = self.str, self.stmts + i = len(stmts) - 1 + p = len(str) # index of newest line + found = 0 + while i: + assert p + # p is the index of the stmt at line number stmts[i]. + # Move p back to the stmt at line number stmts[i-1]. + q = p + for nothing in range(stmts[i-1], stmts[i]): + # tricky: sets p to 0 if no preceding newline + p = _rfind(str, '\n', 0, p-1) + 1 + # The stmt str[p:q] isn't a continuation, but may be blank + # or a non-indenting comment line. + if _junkre(str, p): + i = i-1 + else: + found = 1 + break + self.stmt_start, self.stmt_end = p, q + + # Analyze this stmt, to find the last open bracket (if any) + # and last interesting character (if any). + stack = [] # stack of open bracket indices + push_stack = stack.append + while p < q: + ch = str[p] + if ch == '"' or ch == "'": + # consume string + # Note that study1 did this with a Python loop, but + # we use a regexp here; the reason is speed in both + # cases; the string may be huge, but study1 pre-squashed + # strings to a couple of characters per line. study1 + # also needed to keep track of newlines, and we don't + # have to. + self.lastch = ch + p = _match_stringre(str, p, q).end() + continue + + if ch == '#': + # consume comment and trailing newline + p = _find(str, '\n', p, q) + 1 + assert p > 0 + continue + + if ch == '\\': + assert p+1 < q + if str[p+1] != '\n': + # the program is invalid, but can't complain + self.lastch = str[p:p+2] + p = p+2 + continue + + if ch not in _ws: + self.lastch = ch + if ch in "([{": + push_stack(p) + elif ch in ")]}" and stack: + del stack[-1] + p = p+1 + + # end while p < q: + + if stack: + self.lastopenbracketpos = stack[-1] + + # Assuming continuation is C_BRACKET, return the number + # of spaces the next line should be indented. + + def compute_bracket_indent(self, _find=string.find): + self._study2() + assert self.continuation == C_BRACKET + j = self.lastopenbracketpos + str = self.str + n = len(str) + origi = i = string.rfind(str, '\n', 0, j) + 1 + j = j+1 + # find first list item + while _not_itemre(str, j): + # this line is junk; advance to the next line + i = _find(str, '\n', j) + if i < 0: + break + j = i = i+1 + if i < 0 or j >= n: + # nothing interesting follows the bracket; + # reproduce the bracket line's indentation + a level + j = i = origi + extra = self.indentwidth + else: + # the first list item begins on this line; line up with + # the first interesting character + extra = 0 + while str[j] in " \t": + j = j+1 + return len(string.expandtabs(str[i:j], + self.tabwidth)) + extra + + # Return number of physical lines in last stmt (whether or not + # it's an interesting stmt! this is intended to be called when + # continuation is C_BACKSLASH). + + def get_num_lines_in_stmt(self): + self._study1() + stmts = self.stmts + return stmts[-1] - stmts[-2] + + # Assuming continuation is C_BACKSLASH, return the number of spaces + # the next line should be indented. Also assuming the new line is + # the first one following the initial line of the stmt. + + def compute_backslash_indent(self): + self._study2() + assert self.continuation == C_BACKSLASH + str = self.str + i = self.stmt_start + while str[i] in " \t": + i = i+1 + startpos = i + endpos = string.find(str, '\n', startpos) + 1 + found = level = 0 + while i < endpos: + ch = str[i] + if ch in "([{": + level = level + 1 + i = i+1 + elif ch in ")]}": + if level: + level = level - 1 + i = i+1 + elif ch == '"' or ch == "'": + i = _match_stringre(str, i, endpos).end() + elif ch == '#': + break + elif level == 0 and ch == '=' and \ + (i == 0 or str[i-1] not in "=<>!") and \ + str[i+1] != '=': + found = 1 + break + else: + i = i+1 + + if found: + # found a legit =, but it may be the last interesting + # thing on the line + i = i+1 # move beyond the = + found = re.match(r"\s*\\", str[i:endpos]) is None + + if not found: + # oh well ... settle for moving beyond the first chunk + # of non-whitespace chars + i = startpos + while str[i] not in " \t\n": + i = i+1 + + return len(string.expandtabs(str[self.stmt_start : + i], + self.tabwidth)) + 1 + + # Return the leading whitespace on the initial line of the last + # interesting stmt. + + def get_base_indent_string(self): + self._study2() + i, n = self.stmt_start, self.stmt_end + assert i is not None + j = i + str = self.str + while j < n and str[j] in " \t": + j = j + 1 + return str[i:j] + + # Did the last interesting stmt open a block? + + def is_block_opener(self): + self._study2() + return self.lastch == ':' + + # Did the last interesting stmt close a block? + + def is_block_closer(self): + self._study2() + return _closere(self.str, self.stmt_start) is not None