Switch inputtransformer2 back to stdlib tokenize module

IPython/core/inputtransformer2.py

@@ -1,534 +1,553 b''
 """Input transformer machinery to support IPython special syntax.
 
 This includes the machinery to recognise and transform ``%magic`` commands,
 ``!system`` commands, ``help?`` querying, prompt stripping, and so forth.
 """
 
 # Copyright (c) IPython Development Team.
 # Distributed under the terms of the Modified BSD License.
 
 from codeop import compile_command
 import re
+import tokenize
 from typing import List, Tuple
-from IPython.utils import tokenize2
-from IPython.utils.tokenutil import generate_tokens
 
 _indent_re = re.compile(r'^[ \t]+')
 
 def leading_indent(lines):
     """Remove leading indentation.
     
     If the first line starts with a spaces or tabs, the same whitespace will be
     removed from each following line.
     """
     m = _indent_re.match(lines[0])
     if not m:
         return lines
     space = m.group(0)
     n = len(space)
     return [l[n:] if l.startswith(space) else l
             for l in lines]
 
 class PromptStripper:
     """Remove matching input prompts from a block of input.
     
     Parameters
     ----------
     prompt_re : regular expression
         A regular expression matching any input prompt (including continuation)
     initial_re : regular expression, optional
         A regular expression matching only the initial prompt, but not continuation.
         If no initial expression is given, prompt_re will be used everywhere.
         Used mainly for plain Python prompts, where the continuation prompt
         ``...`` is a valid Python expression in Python 3, so shouldn't be stripped.
     
     If initial_re and prompt_re differ,
     only initial_re will be tested against the first line.
     If any prompt is found on the first two lines,
     prompts will be stripped from the rest of the block.
     """
     def __init__(self, prompt_re, initial_re=None):
         self.prompt_re = prompt_re
         self.initial_re = initial_re or prompt_re
 
     def _strip(self, lines):
         return [self.prompt_re.sub('', l, count=1) for l in lines]
 
     def __call__(self, lines):
         if self.initial_re.match(lines[0]) or \
                 (len(lines) > 1 and self.prompt_re.match(lines[1])):
             return self._strip(lines)
         return lines
 
 classic_prompt = PromptStripper(
     prompt_re=re.compile(r'^(>>>|\.\.\.)( |$)'),
     initial_re=re.compile(r'^>>>( |$)')
 )
 
 ipython_prompt = PromptStripper(re.compile(r'^(In \[\d+\]: |\s*\.{3,}: ?)'))
 
 def cell_magic(lines):
     if not lines[0].startswith('%%'):
         return lines
     if re.match('%%\w+\?', lines[0]):
         # This case will be handled by help_end
         return lines
     magic_name, _, first_line = lines[0][2:-1].partition(' ')
     body = ''.join(lines[1:])
     return ['get_ipython().run_cell_magic(%r, %r, %r)\n'
             % (magic_name, first_line, body)]
 
 # -----
 
 def _find_assign_op(token_line):
     # Find the first assignment in the line ('=' not inside brackets)
     # We don't try to support multiple special assignment (a = b = %foo)
     paren_level = 0
     for i, ti in enumerate(token_line):
         s = ti.string
         if s == '=' and paren_level == 0:
             return i 
         if s in '([{':
             paren_level += 1
         elif s in ')]}':
             paren_level -= 1
 
 def find_end_of_continued_line(lines, start_line: int):
     """Find the last line of a line explicitly extended using backslashes.
 
     Uses 0-indexed line numbers.
     """
     end_line = start_line
     while lines[end_line].endswith('\\\n'):
         end_line += 1
         if end_line >= len(lines):
             break
     return end_line
 
 def assemble_continued_line(lines, start: Tuple[int, int], end_line: int):
     """Assemble pieces of a continued line into a single line.
 
     Uses 0-indexed line numbers. *start* is (lineno, colno).
     """
     parts = [lines[start[0]][start[1]:]] + lines[start[0]+1:end_line+1]
     return ' '.join([p[:-2] for p in parts[:-1]]  # Strip backslash+newline
                     + [parts[-1][:-1]])         # Strip newline from last line
 
 class TokenTransformBase:
     # Lower numbers -> higher priority (for matches in the same location)
     priority = 10
 
     def sortby(self):
         return self.start_line, self.start_col, self.priority
 
     def __init__(self, start):
         self.start_line = start[0] - 1   # Shift from 1-index to 0-index
         self.start_col = start[1]
 
     def transform(self, lines: List[str]):
         raise NotImplementedError
 
 class MagicAssign(TokenTransformBase):
     @classmethod
     def find(cls, tokens_by_line):
         """Find the first magic assignment (a = %foo) in the cell.
         
         Returns (line, column) of the % if found, or None. *line* is 1-indexed.
         """
         for line in tokens_by_line:
             assign_ix = _find_assign_op(line)
             if (assign_ix is not None) \
                     and (len(line) >= assign_ix + 2) \
                     and (line[assign_ix+1].string == '%') \
-                    and (line[assign_ix+2].type == tokenize2.NAME):
+                    and (line[assign_ix+2].type == tokenize.NAME):
                 return cls(line[assign_ix+1].start)
     
     def transform(self, lines: List[str]):
         """Transform a magic assignment found by find
         """
         start_line, start_col = self.start_line, self.start_col
         lhs = lines[start_line][:start_col]
         end_line = find_end_of_continued_line(lines, start_line)
         rhs = assemble_continued_line(lines, (start_line, start_col), end_line)
         assert rhs.startswith('%'), rhs
         magic_name, _, args = rhs[1:].partition(' ')
         
         lines_before = lines[:start_line]
         call = "get_ipython().run_line_magic({!r}, {!r})".format(magic_name, args)
         new_line = lhs + call + '\n'
         lines_after = lines[end_line+1:]
         
         return lines_before + [new_line] + lines_after
 
 
 class SystemAssign(TokenTransformBase):
     @classmethod
     def find(cls, tokens_by_line):
         """Find the first system assignment (a = !foo) in the cell.
 
         Returns (line, column) of the ! if found, or None. *line* is 1-indexed.
         """
         for line in tokens_by_line:
             assign_ix = _find_assign_op(line)
             if (assign_ix is not None) \
                     and (len(line) >= assign_ix + 2) \
-                    and (line[assign_ix + 1].type == tokenize2.ERRORTOKEN):
+                    and (line[assign_ix + 1].type == tokenize.ERRORTOKEN):
                 ix = assign_ix + 1
 
-                while ix < len(line) and line[ix].type == tokenize2.ERRORTOKEN:
+                while ix < len(line) and line[ix].type == tokenize.ERRORTOKEN:
                     if line[ix].string == '!':
                         return cls(line[ix].start)
                     elif not line[ix].string.isspace():
                         break
                     ix += 1
 
     def transform(self, lines: List[str]):
         """Transform a system assignment found by find
         """
         start_line, start_col = self.start_line, self.start_col
 
         lhs = lines[start_line][:start_col]
         end_line = find_end_of_continued_line(lines, start_line)
         rhs = assemble_continued_line(lines, (start_line, start_col), end_line)
         assert rhs.startswith('!'), rhs
         cmd = rhs[1:]
 
         lines_before = lines[:start_line]
         call = "get_ipython().getoutput({!r})".format(cmd)
         new_line = lhs + call + '\n'
         lines_after = lines[end_line + 1:]
 
         return lines_before + [new_line] + lines_after
 
 # The escape sequences that define the syntax transformations IPython will
 # apply to user input.  These can NOT be just changed here: many regular
 # expressions and other parts of the code may use their hardcoded values, and
 # for all intents and purposes they constitute the 'IPython syntax', so they
 # should be considered fixed.
 
 ESC_SHELL  = '!'     # Send line to underlying system shell
 ESC_SH_CAP = '!!'    # Send line to system shell and capture output
 ESC_HELP   = '?'     # Find information about object
 ESC_HELP2  = '??'    # Find extra-detailed information about object
 ESC_MAGIC  = '%'     # Call magic function
 ESC_MAGIC2 = '%%'    # Call cell-magic function
 ESC_QUOTE  = ','     # Split args on whitespace, quote each as string and call
 ESC_QUOTE2 = ';'     # Quote all args as a single string, call
 ESC_PAREN  = '/'     # Call first argument with rest of line as arguments
 
 ESCAPE_SINGLES = {'!', '?', '%', ',', ';', '/'}
 ESCAPE_DOUBLES = {'!!', '??'}  # %% (cell magic) is handled separately
 
 def _make_help_call(target, esc, next_input=None):
     """Prepares a pinfo(2)/psearch call from a target name and the escape
     (i.e. ? or ??)"""
     method  = 'pinfo2' if esc == '??' \
                 else 'psearch' if '*' in target \
                 else 'pinfo'
     arg = " ".join([method, target])
     #Prepare arguments for get_ipython().run_line_magic(magic_name, magic_args)
     t_magic_name, _, t_magic_arg_s = arg.partition(' ')
     t_magic_name = t_magic_name.lstrip(ESC_MAGIC)
     if next_input is None:
         return 'get_ipython().run_line_magic(%r, %r)' % (t_magic_name, t_magic_arg_s)
     else:
         return 'get_ipython().set_next_input(%r);get_ipython().run_line_magic(%r, %r)' % \
            (next_input, t_magic_name, t_magic_arg_s)
 
 def _tr_help(content):
     "Translate lines escaped with: ?"
     # A naked help line should just fire the intro help screen
     if not content:
         return 'get_ipython().show_usage()'
 
     return _make_help_call(content, '?')
 
 def _tr_help2(content):
     "Translate lines escaped with: ??"
     # A naked help line should just fire the intro help screen
     if not content:
         return 'get_ipython().show_usage()'
 
     return _make_help_call(content, '??')
 
 def _tr_magic(content):
     "Translate lines escaped with: %"
     name, _, args = content.partition(' ')
     return 'get_ipython().run_line_magic(%r, %r)' % (name, args)
 
 def _tr_quote(content):
     "Translate lines escaped with: ,"
     name, _, args = content.partition(' ')
     return '%s("%s")' % (name, '", "'.join(args.split()) )
 
 def _tr_quote2(content):
     "Translate lines escaped with: ;"
     name, _, args = content.partition(' ')
     return '%s("%s")' % (name, args)
 
 def _tr_paren(content):
     "Translate lines escaped with: /"
     name, _, args = content.partition(' ')
     return '%s(%s)' % (name, ", ".join(args.split()))
 
 tr = { ESC_SHELL  : 'get_ipython().system({!r})'.format,
        ESC_SH_CAP : 'get_ipython().getoutput({!r})'.format,
        ESC_HELP   : _tr_help,
        ESC_HELP2  : _tr_help2,
        ESC_MAGIC  : _tr_magic,
        ESC_QUOTE  : _tr_quote,
        ESC_QUOTE2 : _tr_quote2,
        ESC_PAREN  : _tr_paren }
 
 class EscapedCommand(TokenTransformBase):
     @classmethod
     def find(cls, tokens_by_line):
         """Find the first escaped command (%foo, !foo, etc.) in the cell.
 
         Returns (line, column) of the escape if found, or None. *line* is 1-indexed.
         """
         for line in tokens_by_line:
             ix = 0
-            while line[ix].type in {tokenize2.INDENT, tokenize2.DEDENT}:
+            while line[ix].type in {tokenize.INDENT, tokenize.DEDENT}:
                 ix += 1
             if line[ix].string in ESCAPE_SINGLES:
                 return cls(line[ix].start)
 
     def transform(self, lines):
         start_line, start_col = self.start_line, self.start_col
 
         indent = lines[start_line][:start_col]
         end_line = find_end_of_continued_line(lines, start_line)
         line = assemble_continued_line(lines, (start_line, start_col), end_line)
 
         if line[:2] in ESCAPE_DOUBLES:
             escape, content = line[:2], line[2:]
         else:
             escape, content = line[:1], line[1:]
         call = tr[escape](content)
 
         lines_before = lines[:start_line]
         new_line = indent + call + '\n'
         lines_after = lines[end_line + 1:]
 
         return lines_before + [new_line] + lines_after
 
 _help_end_re = re.compile(r"""(%{0,2}
                               [a-zA-Z_*][\w*]*        # Variable name
                               (\.[a-zA-Z_*][\w*]*)*   # .etc.etc
                               )
                               (\?\??)$                # ? or ??
                               """,
                               re.VERBOSE)
 
 class HelpEnd(TokenTransformBase):
     # This needs to be higher priority (lower number) than EscapedCommand so
     # that inspecting magics (%foo?) works.
     priority = 5
 
     def __init__(self, start, q_locn):
         super().__init__(start)
         self.q_line = q_locn[0] - 1  # Shift from 1-indexed to 0-indexed
         self.q_col = q_locn[1]
 
     @classmethod
     def find(cls, tokens_by_line):
         for line in tokens_by_line:
             # Last token is NEWLINE; look at last but one
             if len(line) > 2 and line[-2].string == '?':
                 # Find the first token that's not INDENT/DEDENT
                 ix = 0
-                while line[ix].type in {tokenize2.INDENT, tokenize2.DEDENT}:
+                while line[ix].type in {tokenize.INDENT, tokenize.DEDENT}:
                     ix += 1
                 return cls(line[ix].start, line[-2].start)
 
     def transform(self, lines):
         piece = ''.join(lines[self.start_line:self.q_line+1])
         indent, content = piece[:self.start_col], piece[self.start_col:]
         lines_before = lines[:self.start_line]
         lines_after = lines[self.q_line + 1:]
 
         m = _help_end_re.search(content)
         assert m is not None, content
         target = m.group(1)
         esc = m.group(3)
 
         # If we're mid-command, put it back on the next prompt for the user.
         next_input = None
         if (not lines_before) and (not lines_after) \
                 and content.strip() != m.group(0):
             next_input = content.rstrip('?\n')
 
         call = _make_help_call(target, esc, next_input=next_input)
         new_line = indent + call + '\n'
 
         return lines_before + [new_line] + lines_after
 
 def make_tokens_by_line(lines):
+    """Tokenize a series of lines and group tokens by line.
+
+    The tokens for a multiline Python string or expression are
+    grouped as one line.
+    """
+    # NL tokens are used inside multiline expressions, but also after blank
+    # lines or comments. This is intentional - see https://bugs.python.org/issue17061
+    # We want to group the former case together but split the latter, so we
+    # track parentheses level, similar to the internals of tokenize.
+    NEWLINE, NL = tokenize.NEWLINE, tokenize.NL
     tokens_by_line = [[]]
-    for token in generate_tokens(iter(lines).__next__):
-        tokens_by_line[-1].append(token)
-        if token.type == tokenize2.NEWLINE:
-            tokens_by_line.append([])
+    parenlev = 0
+    try:
+        for token in tokenize.generate_tokens(iter(lines).__next__):
+            tokens_by_line[-1].append(token)
+            if (token.type == NEWLINE) \
+                    or ((token.type == NL) and (parenlev <= 0)):
+                tokens_by_line.append([])
+            elif token.string in {'(', '[', '{'}:
+                parenlev += 1
+            elif token.string in {')', ']', '}'}:
+                parenlev -= 1
+    except tokenize.TokenError:
+        # Input ended in a multiline string or expression. That's OK for us.
+        pass
     
     return tokens_by_line
 
 def show_linewise_tokens(s: str):
     """For investigation"""
     if not s.endswith('\n'):
         s += '\n'
     lines = s.splitlines(keepends=True)
     for line in make_tokens_by_line(lines):
         print("Line -------")
         for tokinfo in line:
             print(" ", tokinfo)
 
 class TransformerManager:
     def __init__(self):
         self.cleanup_transforms = [
             leading_indent,
             classic_prompt,
             ipython_prompt,
         ]
         self.line_transforms = [
             cell_magic,
         ]
         self.token_transformers = [
             MagicAssign,
             SystemAssign,
             EscapedCommand,
             HelpEnd,
         ]
     
     def do_one_token_transform(self, lines):
         """Find and run the transform earliest in the code.
         
         Returns (changed, lines).
         
         This method is called repeatedly until changed is False, indicating
         that all available transformations are complete.
 
         The tokens following IPython special syntax might not be valid, so
         the transformed code is retokenised every time to identify the next
         piece of special syntax. Hopefully long code cells are mostly valid
         Python, not using lots of IPython special syntax, so this shouldn't be
         a performance issue.
         """
         tokens_by_line = make_tokens_by_line(lines)
         candidates = []
         for transformer_cls in self.token_transformers:
             transformer = transformer_cls.find(tokens_by_line)
             if transformer:
                 candidates.append(transformer)
 
         if not candidates:
             # Nothing to transform
             return False, lines
 
         transformer = min(candidates, key=TokenTransformBase.sortby)
         return True, transformer.transform(lines)
 
     def do_token_transforms(self, lines):
         while True:
             changed, lines = self.do_one_token_transform(lines)
             if not changed:
                 return lines
 
     def transform_cell(self, cell: str):
         if not cell.endswith('\n'):
             cell += '\n'  # Ensure the cell has a trailing newline
         lines = cell.splitlines(keepends=True)
         for transform in self.cleanup_transforms + self.line_transforms:
             #print(transform, lines)
             lines = transform(lines)
 
         lines = self.do_token_transforms(lines)
         return ''.join(lines)
 
     def check_complete(self, cell: str):
         """Return whether a block of code is ready to execute, or should be continued
 
         Parameters
         ----------
         source : string
           Python input code, which can be multiline.
 
         Returns
         -------
         status : str
           One of 'complete', 'incomplete', or 'invalid' if source is not a
           prefix of valid code.
         indent_spaces : int or None
           The number of spaces by which to indent the next line of code. If
           status is not 'incomplete', this is None.
         """
         if not cell.endswith('\n'):
             cell += '\n'  # Ensure the cell has a trailing newline
         lines = cell.splitlines(keepends=True)
         if lines[-1][:-1].endswith('\\'):
             # Explicit backslash continuation
             return 'incomplete', find_last_indent(lines)
 
         try:
             for transform in self.cleanup_transforms:
                 lines = transform(lines)
         except SyntaxError:
             return 'invalid', None
 
         if lines[0].startswith('%%'):
             # Special case for cell magics - completion marked by blank line
             if lines[-1].strip():
                 return 'incomplete', find_last_indent(lines)
             else:
                 return 'complete', None
 
         try:
             for transform in self.line_transforms:
                 lines = transform(lines)
             lines = self.do_token_transforms(lines)
         except SyntaxError:
             return 'invalid', None
 
         tokens_by_line = make_tokens_by_line(lines)
-        if tokens_by_line[-1][-1].type != tokenize2.ENDMARKER:
+        if tokens_by_line[-1][-1].type != tokenize.ENDMARKER:
             # We're in a multiline string or expression
             return 'incomplete', find_last_indent(lines)
 
         # Find the last token on the previous line that's not NEWLINE or COMMENT
         toks_last_line = tokens_by_line[-2]
         ix = len(toks_last_line) - 1
-        while ix >= 0 and toks_last_line[ix].type in {tokenize2.NEWLINE,
-                                                      tokenize2.COMMENT}:
+        while ix >= 0 and toks_last_line[ix].type in {tokenize.NEWLINE,
+                                                      tokenize.COMMENT}:
             ix -= 1
 
         if toks_last_line[ix].string == ':':
             # The last line starts a block (e.g. 'if foo:')
             ix = 0
-            while toks_last_line[ix].type in {tokenize2.INDENT, tokenize2.DEDENT}:
+            while toks_last_line[ix].type in {tokenize.INDENT, tokenize.DEDENT}:
                 ix += 1
             indent = toks_last_line[ix].start[1]
             return 'incomplete', indent + 4
 
         # If there's a blank line at the end, assume we're ready to execute.
         if not lines[-1].strip():
             return 'complete', None
 
         # At this point, our checks think the code is complete (or invalid).
         # We'll use codeop.compile_command to check this with the real parser.
 
         try:
             res = compile_command(''.join(lines), symbol='exec')
         except (SyntaxError, OverflowError, ValueError, TypeError,
                 MemoryError, SyntaxWarning):
             return 'invalid', None
         else:
             if res is None:
                 return 'incomplete', find_last_indent(lines)
         return 'complete', None
 
 
 def find_last_indent(lines):
     m = _indent_re.match(lines[-1])
     if not m:
         return 0
     return len(m.group(0).replace('\t', ' '*4))