read_sql_*() methods have incorrect typing for dtype argument.

[Dr-Irv]

The read_sql() docs say for the new dtype argument:

dtype Type name or dict of columns
Data type for data or columns. E.g. np.float64 or {‘a’: np.float64, ‘b’: np.int32, ‘c’: ‘Int64’}. The argument is ignored if a table is passed instead of a query.

But in the typing in the source code, the argument is DtypeArg | None which doesn't allow something like {"c": "str"} . The implementation just passes dtype to DataFrame.astype().

See pandas-dev/pandas-stubs#676

NOTE: In pandas-stubs we did work to create all the possible strings and dtypes for astype(), so maybe that should be copied over to the pandas source??

[Dr-Irv]

For pandas, I think the solution is to redefine

pandas/pandas/_typing.py

Line 169 in 042ebab

DtypeArg = Union[Dtype, Dict[Hashable, Dtype]]

to use Mapping instead of Dict

[teacherc]

take

[twoertwein]

For pandas, I think the solution is to redefine

pandas/pandas/_typing.py

Line 169 in 042ebab

DtypeArg = Union[Dtype, Dict[Hashable, Dtype]]

to use Mapping instead of Dict

Closing this issue as the suggested change has been made:

pandas/pandas/_typing.py

Line 221 in de1131f

DtypeArg = Union[Dtype, Mapping[Hashable, Dtype]]