##// END OF EJS Templates
CLN: core/display: reformat per darker
CLN: core/display: reformat per darker

File last commit:

r28327:124787ee
r28917:44421d9a
Show More
inputtransformer2.py
827 lines | 28.7 KiB | text/x-python | PythonLexer
"""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.
Added: IPython 7.0. Replaces inputsplitter and inputtransformer which were
deprecated in 7.0.
"""
# Copyright (c) IPython Development Team.
# Distributed under the terms of the Modified BSD License.
import ast
from codeop import CommandCompiler, Compile
import re
import sys
import tokenize
from typing import List, Tuple, Optional, Any
import warnings
from IPython.utils import tokenutil
_indent_re = re.compile(r'^[ \t]+')
def leading_empty_lines(lines):
"""Remove leading empty lines
If the leading lines are empty or contain only whitespace, they will be
removed.
"""
if not lines:
return lines
for i, line in enumerate(lines):
if line and not line.isspace():
return lines[i:]
return lines
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 in the cell.
"""
if not lines:
return lines
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,
e.g. ``...``)
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.
Notes
-----
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 not lines:
return 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"""
^( # Match from the beginning of a line, either:
# 1. First-line prompt:
((\[nav\]|\[ins\])?\ )? # Vi editing mode prompt, if it's there
In\ # The 'In' of the prompt, with a space
\[\d+\]: # Command index, as displayed in the prompt
\ # With a mandatory trailing space
| # ... or ...
# 2. The three dots of the multiline prompt
\s* # All leading whitespace characters
\.{3,}: # The three (or more) dots
\ ? # With an optional trailing space
)
""",
re.VERBOSE,
)
)
def cell_magic(lines):
if not lines or not lines[0].startswith('%%'):
return lines
if re.match(r'%%\w+\?', lines[0]):
# This case will be handled by help_end
return lines
magic_name, _, first_line = lines[0][2:].rstrip().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) -> Optional[int]:
"""Get the index of the first assignment in the line ('=' not inside brackets)
Note: 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 {')', ']', '}'}:
if paren_level > 0:
paren_level -= 1
return None
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):
r"""Assemble a single line from multiple continued line pieces
Continued lines are lines ending in ``\``, and the line following the last
``\`` in the block.
For example, this code continues over multiple lines::
if (assign_ix is not None) \
and (len(line) >= assign_ix + 2) \
and (line[assign_ix+1].string == '%') \
and (line[assign_ix+2].type == tokenize.NAME):
This statement contains four continued line pieces.
Assembling these pieces into a single line would give::
if (assign_ix is not None) and (len(line) >= assign_ix + 2) and (line[...
This uses 0-indexed line numbers. *start* is (lineno, colno).
Used to allow ``%magic`` and ``!system`` commands to be continued over
multiple lines.
"""
parts = [lines[start[0]][start[1]:]] + lines[start[0]+1:end_line+1]
return ' '.join([p.rstrip()[:-1] for p in parts[:-1]] # Strip backslash+newline
+ [parts[-1].rstrip()]) # Strip newline from last line
class TokenTransformBase:
"""Base class for transformations which examine tokens.
Special syntax should not be transformed when it occurs inside strings or
comments. This is hard to reliably avoid with regexes. The solution is to
tokenise the code as Python, and recognise the special syntax in the tokens.
IPython's special syntax is not valid Python syntax, so tokenising may go
wrong after the special syntax starts. These classes therefore find and
transform *one* instance of special syntax at a time into regular Python
syntax. After each transformation, tokens are regenerated to find the next
piece of special syntax.
Subclasses need to implement one class method (find)
and one regular method (transform).
The priority attribute can select which transformation to apply if multiple
transformers match in the same place. Lower numbers have higher priority.
This allows "%magic?" to be turned into a help call rather than a magic call.
"""
# 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]
@classmethod
def find(cls, tokens_by_line):
"""Find one instance of special syntax in the provided tokens.
Tokens are grouped into logical lines for convenience,
so it is easy to e.g. look at the first token of each line.
*tokens_by_line* is a list of lists of tokenize.TokenInfo objects.
This should return an instance of its class, pointing to the start
position it has found, or None if it found no match.
"""
raise NotImplementedError
def transform(self, lines: List[str]):
"""Transform one instance of special syntax found by ``find()``
Takes a list of strings representing physical lines,
returns a similar list of transformed lines.
"""
raise NotImplementedError
class MagicAssign(TokenTransformBase):
"""Transformer for assignments from magics (a = %foo)"""
@classmethod
def find(cls, tokens_by_line):
"""Find the first magic assignment (a = %foo) in the cell.
"""
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 == tokenize.NAME):
return cls(line[assign_ix+1].start)
def transform(self, lines: List[str]):
"""Transform a magic assignment found by the ``find()`` classmethod.
"""
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):
"""Transformer for assignments from system commands (a = !foo)"""
@classmethod
def find_pre_312(cls, tokens_by_line):
for line in tokens_by_line:
assign_ix = _find_assign_op(line)
if (assign_ix is not None) \
and not line[assign_ix].line.strip().startswith('=') \
and (len(line) >= assign_ix + 2) \
and (line[assign_ix + 1].type == tokenize.ERRORTOKEN):
ix = assign_ix + 1
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
@classmethod
def find_post_312(cls, tokens_by_line):
for line in tokens_by_line:
assign_ix = _find_assign_op(line)
if (
(assign_ix is not None)
and not line[assign_ix].line.strip().startswith("=")
and (len(line) >= assign_ix + 2)
and (line[assign_ix + 1].type == tokenize.OP)
and (line[assign_ix + 1].string == "!")
):
return cls(line[assign_ix + 1].start)
@classmethod
def find(cls, tokens_by_line):
"""Find the first system assignment (a = !foo) in the cell."""
if sys.version_info < (3, 12):
return cls.find_pre_312(tokens_by_line)
return cls.find_post_312(tokens_by_line)
def transform(self, lines: List[str]):
"""Transform a system assignment found by the ``find()`` classmethod.
"""
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):
"""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)
return "get_ipython().run_line_magic(%r, %r)" % (t_magic_name, t_magic_arg_s)
def _tr_help(content):
"""Translate lines escaped with: ?
A naked help line should fire the intro help screen (shell.show_usage())
"""
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 fire the intro help screen (shell.show_usage())
"""
if not content:
return 'get_ipython().show_usage()'
return _make_help_call(content, '??')
def _tr_magic(content):
"Translate lines escaped with a percent sign: %"
name, _, args = content.partition(' ')
return 'get_ipython().run_line_magic(%r, %r)' % (name, args)
def _tr_quote(content):
"Translate lines escaped with a comma: ,"
name, _, args = content.partition(' ')
return '%s("%s")' % (name, '", "'.join(args.split()) )
def _tr_quote2(content):
"Translate lines escaped with a semicolon: ;"
name, _, args = content.partition(' ')
return '%s("%s")' % (name, args)
def _tr_paren(content):
"Translate lines escaped with a slash: /"
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):
"""Transformer for escaped commands like %foo, !foo, or /foo"""
@classmethod
def find(cls, tokens_by_line):
"""Find the first escaped command (%foo, !foo, etc.) in the cell.
"""
for line in tokens_by_line:
if not line:
continue
ix = 0
ll = len(line)
while ll > ix and line[ix].type in {tokenize.INDENT, tokenize.DEDENT}:
ix += 1
if ix >= ll:
continue
if line[ix].string in ESCAPE_SINGLES:
return cls(line[ix].start)
def transform(self, lines):
"""Transform an escaped line found by the ``find()`` classmethod.
"""
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 len(line) > 1 and line[:2] in ESCAPE_DOUBLES:
escape, content = line[:2], line[2:]
else:
escape, content = line[:1], line[1:]
if escape in tr:
call = tr[escape](content)
else:
call = ''
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}
(?!\d)[\w*]+ # Variable name
(\.(?!\d)[\w*]+|\[-?[0-9]+\])* # .etc.etc or [0], we only support literal integers.
)
(\?\??)$ # ? or ??
""",
re.VERBOSE,
)
class HelpEnd(TokenTransformBase):
"""Transformer for help syntax: obj? and obj??"""
# 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):
"""Find the first help command (foo?) in the cell.
"""
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 {tokenize.INDENT, tokenize.DEDENT}:
ix += 1
return cls(line[ix].start, line[-2].start)
def transform(self, lines):
"""Transform a help command found by the ``find()`` classmethod.
"""
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)
if not m:
raise SyntaxError(content)
assert m is not None, content
target = m.group(1)
esc = m.group(3)
call = _make_help_call(target, esc)
new_line = indent + call + '\n'
return lines_before + [new_line] + lines_after
def make_tokens_by_line(lines:List[str]):
"""Tokenize a series of lines and group tokens by line.
The tokens for a multiline Python string or expression are grouped as one
line. All lines except the last lines should keep their line ending ('\\n',
'\\r\\n') for this to properly work. Use `.splitlines(keeplineending=True)`
for example when passing block of text to this function.
"""
# 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.
# reexported from token on 3.7+
NEWLINE, NL = tokenize.NEWLINE, tokenize.NL # type: ignore
tokens_by_line: List[List[Any]] = [[]]
if len(lines) > 1 and not lines[0].endswith(("\n", "\r", "\r\n", "\x0b", "\x0c")):
warnings.warn(
"`make_tokens_by_line` received a list of lines which do not have lineending markers ('\\n', '\\r', '\\r\\n', '\\x0b', '\\x0c'), behavior will be unspecified",
stacklevel=2,
)
parenlev = 0
try:
for token in tokenutil.generate_tokens_catch_errors(
iter(lines).__next__, extra_errors_to_catch=["expected EOF"]
):
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 {')', ']', '}'}:
if parenlev > 0:
parenlev -= 1
except tokenize.TokenError:
# Input ended in a multiline string or expression. That's OK for us.
pass
if not tokens_by_line[-1]:
tokens_by_line.pop()
return tokens_by_line
def has_sunken_brackets(tokens: List[tokenize.TokenInfo]):
"""Check if the depth of brackets in the list of tokens drops below 0"""
parenlev = 0
for token in tokens:
if token.string in {"(", "[", "{"}:
parenlev += 1
elif token.string in {")", "]", "}"}:
parenlev -= 1
if parenlev < 0:
return True
return False
def show_linewise_tokens(s: str):
"""For investigation and debugging"""
warnings.warn(
"show_linewise_tokens is deprecated since IPython 8.6",
DeprecationWarning,
stacklevel=2,
)
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)
# Arbitrary limit to prevent getting stuck in infinite loops
TRANSFORM_LOOP_LIMIT = 500
class TransformerManager:
"""Applies various transformations to a cell or code block.
The key methods for external use are ``transform_cell()``
and ``check_complete()``.
"""
def __init__(self):
self.cleanup_transforms = [
leading_empty_lines,
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
ordered_transformers = sorted(candidates, key=TokenTransformBase.sortby)
for transformer in ordered_transformers:
try:
return True, transformer.transform(lines)
except SyntaxError:
pass
return False, lines
def do_token_transforms(self, lines):
for _ in range(TRANSFORM_LOOP_LIMIT):
changed, lines = self.do_one_token_transform(lines)
if not changed:
return lines
raise RuntimeError("Input transformation still changing after "
"%d iterations. Aborting." % TRANSFORM_LOOP_LIMIT)
def transform_cell(self, cell: str) -> str:
"""Transforms a cell of input code"""
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:
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
----------
cell : 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.
"""
# Remember if the lines ends in a new line.
ends_with_newline = False
for character in reversed(cell):
if character == '\n':
ends_with_newline = True
break
elif character.strip():
break
else:
continue
if not ends_with_newline:
# Append an newline for consistent tokenization
# See https://bugs.python.org/issue33899
cell += '\n'
lines = cell.splitlines(keepends=True)
if not lines:
return 'complete', None
for line in reversed(lines):
if not line.strip():
continue
elif line.strip("\n").endswith("\\"):
return "incomplete", find_last_indent(lines)
else:
break
try:
for transform in self.cleanup_transforms:
if not getattr(transform, 'has_side_effects', False):
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:
if not getattr(transform, 'has_side_effects', False):
lines = transform(lines)
lines = self.do_token_transforms(lines)
except SyntaxError:
return 'invalid', None
tokens_by_line = make_tokens_by_line(lines)
# Bail if we got one line and there are more closing parentheses than
# the opening ones
if (
len(lines) == 1
and tokens_by_line
and has_sunken_brackets(tokens_by_line[0])
):
return "invalid", None
if not tokens_by_line:
return 'incomplete', find_last_indent(lines)
if (
tokens_by_line[-1][-1].type != tokenize.ENDMARKER
and tokens_by_line[-1][-1].type != tokenize.ERRORTOKEN
):
# We're in a multiline string or expression
return 'incomplete', find_last_indent(lines)
newline_types = {tokenize.NEWLINE, tokenize.COMMENT, tokenize.ENDMARKER} # type: ignore
# Pop the last line which only contains DEDENTs and ENDMARKER
last_token_line = None
if {t.type for t in tokens_by_line[-1]} in [
{tokenize.DEDENT, tokenize.ENDMARKER},
{tokenize.ENDMARKER}
] and len(tokens_by_line) > 1:
last_token_line = tokens_by_line.pop()
while tokens_by_line[-1] and tokens_by_line[-1][-1].type in newline_types:
tokens_by_line[-1].pop()
if not tokens_by_line[-1]:
return 'incomplete', find_last_indent(lines)
if tokens_by_line[-1][-1].string == ':':
# The last line starts a block (e.g. 'if foo:')
ix = 0
while tokens_by_line[-1][ix].type in {tokenize.INDENT, tokenize.DEDENT}:
ix += 1
indent = tokens_by_line[-1][ix].start[1]
return 'incomplete', indent + 4
if tokens_by_line[-1][0].line.endswith('\\'):
return 'incomplete', 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:
with warnings.catch_warnings():
warnings.simplefilter('error', SyntaxWarning)
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)
if last_token_line and last_token_line[0].type == tokenize.DEDENT:
if ends_with_newline:
return 'complete', None
return 'incomplete', find_last_indent(lines)
# If there's a blank line at the end, assume we're ready to execute
if not lines[-1].strip():
return 'complete', None
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))
class MaybeAsyncCompile(Compile):
def __init__(self, extra_flags=0):
super().__init__()
self.flags |= extra_flags
class MaybeAsyncCommandCompiler(CommandCompiler):
def __init__(self, extra_flags=0):
self.compiler = MaybeAsyncCompile(extra_flags=extra_flags)
_extra_flags = ast.PyCF_ALLOW_TOP_LEVEL_AWAIT
compile_command = MaybeAsyncCommandCompiler(extra_flags=_extra_flags)