Pyright now incorrectly declaring "Invalid conditional operand"

[dlew]

Describe the bug

The latest release of pandas-stubs seems to have added some typing that is overly restrictive. I'm seeing many instances of Invalid conditional operand of type "Unknown | Series[_bool]" where before it did not appear.

To Reproduce
Running pandas 2.2.3 + pandas-stubs 2.2.3.250308 + pyright 1.1.396

df = pd.DataFrame(
    data={"foo": "bar"},
    index=["one"],
)
if pd.isna(df.loc["one"]["foo"]):
    pass
assert df.loc["one"]["foo"] == "bar"

Pyright 1.1.396 returns these errors:

error: Invalid conditional operand of type "Series[bool]"
    Method __bool__ for type "Series[bool]" returns type "NoReturn" rather than "bool" (reportGeneralTypeIssues)

error: Invalid conditional operand of type "Unknown | Series[_bool]"
    Method __bool__ for type "Series[_bool]" returns type "NoReturn" rather than "bool" (reportGeneralTypeIssues)

The first error is for the pd.isna(), the second is for the equality assertion. But in both cases, df.loc["one"]["foo"] is a str, so the type checking ought to work out.

Please complete the following information:

• OS: MacOS
• OS Version: Sequoia 15.3.1
• Python: 3.12
• Pyright: 1.1.396
• pandas-stubs: 2.2.3.250308

Additional context
Did not see this on pandas-stubs 2.2.3.241126.

[Dr-Irv]

Instead of:

if pd.isna(df.loc["one"]["foo"]):
    pass
assert df.loc["one"]["foo"] == "bar"

use

if pd.isna(df.loc["one", "foo"]):
    pass
assert df.loc["one", "foo"] == "bar"

This change was to handle some cases that were ambiguous and using .loc as shown above is the preferred method.

[dlew]

Alright, makes sense, thanks!