internal error while formatting with-statement with tuple and none tuple argument

[15r10nk]

Describe the bug

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

with (name_5, name_4), name_5:
    pass

(playground)

black reported the following error:

> black -l 100 -C -t py313 bug.py
error: cannot format bug.py: INTERNAL ERROR: Black 25.1.1.dev21+g944a38e.d20250317 on Python (CPython) 3.13.1 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_sxhcobwt.log

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

the reported diff in /tmp/blk_sxhcobwt.log is:

--- src
+++ dst
@@ -5,30 +5,29 @@
         Pass(
         )  # /Pass
         items=
         withitem(
             context_expr=
-            Tuple(
+            Name(
                 ctx=
                 Load(
                 )  # /Load
-                elts=
-                Name(
-                    ctx=
-                    Load(
-                    )  # /Load
-                    id=
-                    'name_5',  # str
-                )  # /Name
-                Name(
-                    ctx=
-                    Load(
-                    )  # /Load
-                    id=
-                    'name_4',  # str
-                )  # /Name
-            )  # /Tuple
+                id=
+                'name_5',  # str
+            )  # /Name
+            optional_vars=
+            None,  # NoneType
+        )  # /withitem
+        withitem(
+            context_expr=
+            Name(
+                ctx=
+                Load(
+                )  # /Load
+                id=
+                'name_4',  # str
+            )  # /Name
             optional_vars=
             None,  # NoneType
         )  # /withitem
         withitem(
             context_expr=

but it can be parsed by cpython:

from ast import parse
parse(
    'with (name_5, name_4), name_5:\n'
    '    pass\n'
)

The code can be formatted with black -l 100 -C -t py313 bug.py --fast:

with name_5, name_4, name_5:
    pass

Environment

• Black's version: current main (950ec38)
• OS and Python version: Linux/Python 3.13.1 (main, Jan 14 2025, 22:47:38) [Clang 19.1.6 ]

Additional context

The bug was found by pysource-codegen (see #3908)
The above problem description was created from a script,
let me know if you think it can be improved.

[Pedro-Muller29]

I believe this doesn't need a fix since this code will always fail at runtime, as tuples don't support the context manager protocol.

In fact, running a code such as:

with (open("a.txt", "r"), open("b.txt", "r")), open("c.txt", "r"):
    pass

Will throw the following error:

Traceback (most recent call last):
  File "/anon_path/with_tuple.py", line 1, in <module>
    with (open("a.txt", "r"), open("b.txt", "r")), open("c.txt", "r"):
         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: 'tuple' object does not support the context manager protocol

[15r10nk]

To be honest, I don't like to create these kind of pedantic issues. But this are the things which are found by pysource-codegen.

I can understand when you do not want to fix it, but I have a example where this code can be useful.

def test_tuple_as_contextmanager():
    from contextlib import nullcontext

    try:
        with (nullcontext(),nullcontext()),nullcontext():
            pass
    except TypeError: 
        # test passed
        pass
    else:
        # this should be a type error
        assert False

Code like this could be used in tests of an python implementation. I have no real world example, but this is the only use case I can think of where this syntax is "useful".

[Pedro-Muller29]

Fair point, I guess there is some niche where this would be useful. Working on a PR.

[JelleZijlstra]

I do think we should fix this sort of thing even if the code always fails—if nothing else so that we can continue fuzzing, because fuzzing is a good way to find bugs!

Thanks @15r10nk for doing this fuzzing and thanks @Pedro-Muller29 for the fix.

[15r10nk]

@Pedro-Muller29 I think your fix solves only 50% of the cases.

This is fixed:

with (a, b), c:
    pass

This not:

with c,(a, b):
    pass

I have looked at your merge request, but I can't see the problem.

[Pedro-Muller29]

@Pedro-Muller29 I think your fix solves only 50% of the cases.

This is fixed:

with (a, b), c:
    pass

This not:

with c,(a, b):
    pass

I have looked at your merge request, but I can't see the problem.

Nice catch! Yeah, it's weird that my PR didn't fix that since it looks for commas both ways. I can try to fix it this weekend.