Failure on py3.9: SystemError: Negative size passed to PyUnicode_New

[sobolevn]

Hi! My CI crashed with something rather interesting:

Link: https://github.com/wemake-services/wemake-python-styleguide/runs/4175506726?check_suite_focus=true

 =================================== FAILURES ===================================
______________________________ test_no_exceptions ______________________________

self = <hypothesmith.syntactic.GrammarStrategy object at 0x7f4930fc3e80>
data = ConjectureData(VALID, 28 bytes, frozen)
symbol = NonTerminal('simple_stmt')
draw_state = DrawState(result=['global', 'A', '\\ \n', '#\n'])

    def draw_symbol(self, data, symbol, draw_state):  # type: ignore
        count = len(draw_state.result)
        super().draw_symbol(data, symbol, draw_state)
        if symbol.name in COMPILE_MODES:
            try:
>               compile(
                    source="".join(draw_state.result[count:]),
                    filename="<string>",
                    mode=COMPILE_MODES[symbol.name],
                )
E               SystemError: Negative size passed to PyUnicode_New

.venv/lib/python3.9/site-packages/hypothesmith/syntactic.py:110: SystemError

The above exception was the direct cause of the following exception:

default_options = options(min_name_length=2, max_name_length=45, i_control_code=True, i_dont_control_code=True, max_noqa_comments=10, ne...gnitive_average=8, max_call_level=3, max_annotation_complexity=3, max_import_from_members=8, max_tuple_unpack_length=4)
parse_ast_tree = <function parse_ast_tree.<locals>.factory at 0x7f479cd01430>
parse_tokens = <function parse_tokens.<locals>.factory at 0x7f479ce863a0>

>   ???

tests/test_checker/test_hypothesis.py:40: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
.venv/lib/python3.9/site-packages/hypothesmith/syntactic.py:89: in do_draw
    result = super().do_draw(data)
.venv/lib/python3.9/site-packages/hypothesis/extra/lark.py:153: in do_draw
    self.draw_symbol(data, start, state)
.venv/lib/python3.9/site-packages/hypothesmith/syntactic.py:107: in draw_symbol
    super().draw_symbol(data, symbol, draw_state)
.venv/lib/python3.9/site-packages/hypothesis/extra/lark.py:181: in draw_symbol
    self.draw_symbol(data, e, draw_state)
.venv/lib/python3.9/site-packages/hypothesmith/syntactic.py:107: in draw_symbol
    super().draw_symbol(data, symbol, draw_state)
.venv/lib/python3.9/site-packages/hypothesis/extra/lark.py:181: in draw_symbol
    self.draw_symbol(data, e, draw_state)
.venv/lib/python3.9/site-packages/hypothesmith/syntactic.py:107: in draw_symbol
    super().draw_symbol(data, symbol, draw_state)
.venv/lib/python3.9/site-packages/hypothesis/extra/lark.py:181: in draw_symbol
    self.draw_symbol(data, e, draw_state)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <hypothesmith.syntactic.GrammarStrategy object at 0x7f4930fc3e80>
data = ConjectureData(VALID, 28 bytes, frozen)
symbol = NonTerminal('simple_stmt')
draw_state = DrawState(result=['global', 'A', '\\ \n', '#\n'])

    def draw_symbol(self, data, symbol, draw_state):  # type: ignore
        count = len(draw_state.result)
        super().draw_symbol(data, symbol, draw_state)
        if symbol.name in COMPILE_MODES:
            try:
                compile(
                    source="".join(draw_state.result[count:]),
                    filename="<string>",
                    mode=COMPILE_MODES[symbol.name],
                )
            except SyntaxError:
                # Python's grammar doesn't actually fully describe the behaviour of the
                # CPython parser and AST-post-processor, so we just filter out errors.
                assume(False)
            except Exception as err:  # pragma: no cover
                # Attempting to compile almost-valid strings has triggered a wide range
                # of bizzare errors in CPython, especially with the new PEG parser,
                # and so we maintain this extra clause to ensure that we get a decent
                # error message out of it.
                if isinstance(err, SystemError) and sys.version_info[:3] == (3, 9, 0):
                    # We've triggered https://bugs.python.org/issue42218 - it's been
                    # fixed upstream, so we'll treat it as if it were a SyntaxError.
                    assume(False)
                source_code = ascii("".join(draw_state.result[count:]))
>               raise type(err)(
                    f"compile({source_code}, '<string>', "
                    f"{COMPILE_MODES[symbol.name]!r}) "
                    f"raised {type(err).__name__}: {str(err)}"
                ) from err
E               SystemError: compile('globalA\\ \n#\n', '<string>', 'single') raised SystemError: Negative size passed to PyUnicode_New

.venv/lib/python3.9/site-packages/hypothesmith/syntactic.py:129: SystemError
---------------------------------- Hypothesis ----------------------------------
Highest target scores:
               6  (label='(hypothesmith) number of unique ast node types')
               9  (label='(hypothesmith) instructions in bytecode')
              13  (label='(hypothesmith) total number of ast nodes')


You can reproduce this example by temporarily adding @reproduce_failure('6.24.2', b'AXicY2RkYGBlABIgwMwIYgAJCB8AAUAAEQ==') as a decorator on your test case

[sobolevn]

Test itself: https://github.com/wemake-services/wemake-python-styleguide/blob/master/tests/test_checker/test_hypothesis.py#L39

[Zac-HD]

Huh. Can you reproduce this with the compile('globalA\\ \n#\n', '<string>', 'single') call? In my Python 3.9 env this just raises a SyntaxError: unexpected character after line continuation character, which is much less interesting.

If you can reproduce, we should report this upstream to CPython core.

[sobolevn]

I am trying to reproduce this on 3.9.1 (originally happened on 3.9.8):

>   ???
E   hypothesis.errors.DidNotReproduce: The shape of the test data has changed in some way from where this blob was defined. Are you sure you're running the same test?

tests/test_checker/test_hypothesis.py:40: DidNotReproduce

Right now 3.9.8 is not available in pyenv: pyenv/pyenv#2152

[sobolevn]

@Zac-HD I just got another one!

https://github.com/wemake-services/wemake-python-styleguide/runs/4175737661?check_suite_focus=true#step:7:1041

This one is different, it also has warnings:

 =================================== FAILURES ===================================
______________________________ test_no_exceptions ______________________________

self = <hypothesmith.syntactic.GrammarStrategy object at 0x7ffa6e39ab80>
data = ConjectureData(VALID, 26 bytes, frozen)
symbol = NonTerminal('simple_stmt')
draw_state = DrawState(result=['global', 'A', '\\ \n', ';', '\n'])

    def draw_symbol(self, data, symbol, draw_state):  # type: ignore
        count = len(draw_state.result)
        super().draw_symbol(data, symbol, draw_state)
        if symbol.name in COMPILE_MODES:
            try:
>               compile(
                    source="".join(draw_state.result[count:]),
                    filename="<string>",
                    mode=COMPILE_MODES[symbol.name],
                )
E               SystemError: Negative size passed to PyUnicode_New

.venv/lib/python3.9/site-packages/hypothesmith/syntactic.py:110: SystemError

The above exception was the direct cause of the following exception:

default_options = options(min_name_length=2, max_name_length=45, i_control_code=True, i_dont_control_code=True, max_noqa_comments=10, ne...gnitive_average=8, max_call_level=3, max_annotation_complexity=3, max_import_from_members=8, max_tuple_unpack_length=4)
parse_ast_tree = <function parse_ast_tree.<locals>.factory at 0x7ff8dfa10c10>
parse_tokens = <function parse_tokens.<locals>.factory at 0x7ff8df980160>

>   ???

tests/test_checker/test_hypothesis.py:40: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
.venv/lib/python3.9/site-packages/hypothesmith/syntactic.py:89: in do_draw
    result = super().do_draw(data)
.venv/lib/python3.9/site-packages/hypothesis/extra/lark.py:153: in do_draw
    self.draw_symbol(data, start, state)
.venv/lib/python3.9/site-packages/hypothesmith/syntactic.py:107: in draw_symbol
    super().draw_symbol(data, symbol, draw_state)
.venv/lib/python3.9/site-packages/hypothesis/extra/lark.py:181: in draw_symbol
    self.draw_symbol(data, e, draw_state)
.venv/lib/python3.9/site-packages/hypothesmith/syntactic.py:107: in draw_symbol
    super().draw_symbol(data, symbol, draw_state)
.venv/lib/python3.9/site-packages/hypothesis/extra/lark.py:181: in draw_symbol
    self.draw_symbol(data, e, draw_state)
.venv/lib/python3.9/site-packages/hypothesmith/syntactic.py:107: in draw_symbol
    super().draw_symbol(data, symbol, draw_state)
.venv/lib/python3.9/site-packages/hypothesis/extra/lark.py:181: in draw_symbol
    self.draw_symbol(data, e, draw_state)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <hypothesmith.syntactic.GrammarStrategy object at 0x7ffa6e39ab80>
data = ConjectureData(VALID, 26 bytes, frozen)
symbol = NonTerminal('simple_stmt')
draw_state = DrawState(result=['global', 'A', '\\ \n', ';', '\n'])

    def draw_symbol(self, data, symbol, draw_state):  # type: ignore
        count = len(draw_state.result)
        super().draw_symbol(data, symbol, draw_state)
        if symbol.name in COMPILE_MODES:
            try:
                compile(
                    source="".join(draw_state.result[count:]),
                    filename="<string>",
                    mode=COMPILE_MODES[symbol.name],
                )
            except SyntaxError:
                # Python's grammar doesn't actually fully describe the behaviour of the
                # CPython parser and AST-post-processor, so we just filter out errors.
                assume(False)
            except Exception as err:  # pragma: no cover
                # Attempting to compile almost-valid strings has triggered a wide range
                # of bizzare errors in CPython, especially with the new PEG parser,
                # and so we maintain this extra clause to ensure that we get a decent
                # error message out of it.
                if isinstance(err, SystemError) and sys.version_info[:3] == (3, 9, 0):
                    # We've triggered https://bugs.python.org/issue42218 - it's been
                    # fixed upstream, so we'll treat it as if it were a SyntaxError.
                    assume(False)
                source_code = ascii("".join(draw_state.result[count:]))
>               raise type(err)(
                    f"compile({source_code}, '<string>', "
                    f"{COMPILE_MODES[symbol.name]!r}) "
                    f"raised {type(err).__name__}: {str(err)}"
                ) from err
E               SystemError: compile('globalA\\ \n;\n', '<string>', 'single') raised SystemError: Negative size passed to PyUnicode_New

.venv/lib/python3.9/site-packages/hypothesmith/syntactic.py:129: SystemError
---------------------------------- Hypothesis ----------------------------------
Highest target scores:
              19  (label='(hypothesmith) number of unique ast node types')
              46  (label='(hypothesmith) instructions in bytecode')
              74  (label='(hypothesmith) total number of ast nodes')


You can reproduce this example by temporarily adding @reproduce_failure('6.24.2', b'AXicY2RkYGJlYGQAAWZGCA0hAQFGABE=') as a decorator on your test case
=============================== warnings summary ===============================
tests/test_checker/test_hypothesis.py: 933 warnings
  <string>:1: SyntaxWarning: 'int' object is not callable; perhaps you missed a comma?

tests/test_checker/test_hypothesis.py: 916 warnings
  <string>:3: SyntaxWarning: 'int' object is not callable; perhaps you missed a comma?

-- Docs: https://docs.pytest.org/en/stable/warnings.html

----------- coverage: platform linux, python 3.9.8-final-0 -----------
Name    Stmts   Miss Branch BrPart  Cover   Missing
---------------------------------------------------
---------------------------------------------------
TOTAL    6290      0   2813      0   100%

141 files skipped due to complete coverage.
Coverage HTML written to dir htmlcov
Coverage XML written to file coverage.xml

Required test coverage of 100% reached. Total coverage: 100.00%
============================= SnapshotTest summary =============================
11 snapshots passed.
=========================== short test summary info ============================
FAILED tests/test_checker/test_hypothesis.py::test_no_exceptions - SystemErro...
==== 1 failed, 32026 passed, 12 skipped, 1849 warnings in 330.95s (0:05:30) ====
make: *** [Makefile:14: unit] Error 1

[Zac-HD]

The pyenv PR you linked to has been merged four days ago, so I think it should work?

This does seem like a real bug if we can trigger it with multiple inputs like this; I'd suggest one more attempt to reproduce and then reporting upstream with links to https://bugs.python.org/issue42218 https://bugs.python.org/issue40661 and https://bugs.python.org/issue42109

[Zac-HD]

Ah, https://bugs.python.org/issue45738 has already been opened.

[Zac-HD]

Fixed in 67fe545, version 0.2.0 released 🎉