crash while formatting a delete statement with given line length

[15r10nk]

Describe the bug

The following code can not be parsed/formatted by black:

del ([], name_2), [(), [], name_4, name_3], name_1[[name_2 for name_1 in name_0]]

black reported the following error:

> black -l 80 -C bug.py
error: cannot format bug.py: INTERNAL ERROR: Black 25.1.1.dev21+g254e829.d20250317 on Python (CPython) 3
.12.6 produced code that is not equivalent to the source.  Please report a bug on https://github.com/psf
/black/issues.  This diff might be helpful: /tmp/blk_qtuwnx6z.log

Oh no! 💥 💔 💥
1 file failed to reformat.

but it can be parsed by cpython:

from ast import parse
parse('''
del ([], name_2), [(), [], name_4, name_3], name_1[[name_2 for name_1 in name_0]]
''')

result:

returncode: 0

black transforms the code with --fast into

del (
    ([], name_2),
    [(), [], name_4, name_3],
    name_1[[name_2 for name_1 in name_0]],
)

This has technical the same meaning but a different ast representation (tuple vs targets) https://docs.python.org/3/library/ast.html#ast.Delete.

❯ echo 'del (a,b); del a,b' | python -m ast
Module(
   body=[
      Delete(
         targets=[
            Tuple(
               elts=[
                  Name(id='a', ctx=Del()),
                  Name(id='b', ctx=Del())],
               ctx=Del())]),
      Delete(
         targets=[
            Name(id='a', ctx=Del()),
            Name(id='b', ctx=Del())])],
   type_ignores=[])

Environment

• Black's version: current main (5342d2e
  )
• OS and Python version: Linux/Python 3.12.6 (main, Sep 9 2024, 22:11:19) [Clang 18.1.8 ]

Additional context

The bug was found by pysource-codegen (see #3908)

[JelleZijlstra]

Interesting, we have code in the AST safety check specifically allowing this:

black/src/black/parsing.py

Line 211 in 5342d2e

if (

. Perhaps it's buggy.

[15r10nk]

I think the problem is that there is a tuple inside the tuple in the formatted code.

[tusharsadhwani]

@15r10nk great seeing your work being used to catch bugs in black. I'll pick up fixing this over the week, assuming you aren't working on a fix already.

[15r10nk]

@tusharsadhwani Thank you, you can fix it if you want. I wanted to fix it, but I haven't found the time yet.

[tusharsadhwani]

Minimized the case to:

del (),

$ black -l1 -v x.py 
[...]
Traceback (most recent call last):
  File "/Users/tusharsadhwani/code/black/src/black/__init__.py", line 911, in reformat_one
    if changed is not Changed.CACHED and format_file_in_place(
                                         ^^^^^^^^^^^^^^^^^^^^^
  File "/Users/tusharsadhwani/code/black/src/black/__init__.py", line 953, in format_file_in_place
    dst_contents = format_file_contents(
                   ^^^^^^^^^^^^^^^^^^^^^
  File "/Users/tusharsadhwani/code/black/src/black/__init__.py", line 1085, in format_file_contents
    check_stability_and_equivalence(
  File "/Users/tusharsadhwani/code/black/src/black/__init__.py", line 1059, in check_stability_and_equivalence
    assert_equivalent(src_contents, dst_contents)
  File "/Users/tusharsadhwani/code/black/src/black/__init__.py", line 1540, in assert_equivalent
    raise ASTSafetyError(
black.parsing.ASTSafetyError: INTERNAL ERROR: Black 22.1.1.dev715+g97a730d.d20250219 on Python (CPython) 3.12.3 produced code that is not equivalent to the source.  Please report a bug on https://github.com/psf/black/issues.

[tusharsadhwani]

It was indeed because we end up with a tuple inside a tuple after formatting.

The AST check code was only unwrapping a top level tuple if present, causing the inner tuple after formatting to not get unwrapped before the AST comparison.