DecimalTuple can have string exponent

[sobolevn]

Source code: https://github.com/python/cpython/blob/db115682bd639a2642c617f0b7d5b30cd7d7f472/Lib/_pydecimal.py#L1049-L1053

Closes #9191

[github-actions]

Diff from mypy_primer, showing the effect of this PR on open source code:

mongo-python-driver (https://github.com/mongodb/mongo-python-driver)
+ bson/decimal128.py: note: In function "_decimal_to_128":
+ bson/decimal128.py:103: error: No overload variant of "__add__" of "str" matches argument type "int"  [operator]
+ bson/decimal128.py:103: note: Possible overload variants:
+ bson/decimal128.py:103: note:     def __add__(self, str, /) -> str
+ bson/decimal128.py:103: note: Left operand is of type "Union[int, Literal['n', 'N', 'F']]"

spark (https://github.com/apache/spark)
+ python/pyspark/sql/connect/column.py:114: error: Argument 1 to "abs" has incompatible type "Union[int, Literal['n', 'N', 'F']]"; expected "SupportsAbs[int]"  [arg-type]
+ python/pyspark/sql/connect/column.py:115: error: Argument 1 to "abs" has incompatible type "Union[int, Literal['n', 'N', 'F']]"; expected "SupportsAbs[int]"  [arg-type]

pydantic (https://github.com/samuelcolvin/pydantic)
+ pydantic/_internal/_validators.py:160: error: Unsupported operand types for >= ("Literal['n']" and "int")  [operator]
+ pydantic/_internal/_validators.py:160: error: Unsupported operand types for >= ("Literal['N']" and "int")  [operator]
+ pydantic/_internal/_validators.py:160: error: Unsupported operand types for >= ("Literal['F']" and "int")  [operator]
+ pydantic/_internal/_validators.py:160: note: Left operand is of type "Union[int, Literal['n', 'N', 'F']]"
+ pydantic/_internal/_validators.py:162: error: Unsupported operand types for + ("int" and "Literal['n']")  [operator]
+ pydantic/_internal/_validators.py:162: error: Unsupported operand types for + ("int" and "Literal['N']")  [operator]
+ pydantic/_internal/_validators.py:162: error: Unsupported operand types for + ("int" and "Literal['F']")  [operator]
+ pydantic/_internal/_validators.py:162: note: Right operand is of type "Union[int, Literal['n', 'N', 'F']]"
+ pydantic/_internal/_validators.py:170: error: Argument 1 to "abs" has incompatible type "Union[int, Literal['n', 'N', 'F']]"; expected "SupportsAbs[int]"  [arg-type]
+ pydantic/_internal/_validators.py:171: error: Argument 1 to "abs" has incompatible type "Union[int, Literal['n', 'N', 'F']]"; expected "SupportsAbs[int]"  [arg-type]
+ pydantic/_internal/_validators.py:174: error: Argument 1 to "abs" has incompatible type "Union[int, Literal['n', 'N', 'F']]"; expected "SupportsAbs[int]"  [arg-type]
+ pydantic/json.py:38: error: Unsupported operand types for >= ("Literal['n']" and "int")  [operator]
+ pydantic/json.py:38: error: Unsupported operand types for >= ("Literal['N']" and "int")  [operator]
+ pydantic/json.py:38: error: Unsupported operand types for >= ("Literal['F']" and "int")  [operator]
+ pydantic/json.py:38: note: Left operand is of type "Union[int, Literal['n', 'N', 'F']]"

psycopg (https://github.com/psycopg/psycopg)
+ psycopg/psycopg/types/numeric.py:394: error: Unused "type: ignore" comment
+ psycopg/psycopg/types/numeric.py:396: error: Unused "type: ignore" comment

[sobolevn]

This failure is not related: https://github.com/python/typeshed/actions/runs/3460013243/jobs/5776061994 I cannot restart it.

Looking at the primer results ⌛

[sobolevn]

mongo's code is correct. But, we cannot narrow exponent based on other calls: https://github.com/mongodb/mongo-python-driver/blob/fcb11514506acddbd50b8ec13e76a7c34d336aac/bson/decimal128.py#L81-L83

    sign, digits, exponent = value.as_tuple()

    if value.is_nan():

pyspark's code is broken: https://github.com/apache/spark/blob/632784dec8a0c682a85a888c66e83f709e402e18/python/pyspark/sql/connect/column.py#L114-L115

It is calling abs() on a possibly string. They even state that raise ValueError("Python Decimal not supported.")

pydantic has correct runtime code: https://github.com/pydantic/pydantic/blob/13ad034e4400baf5a875767e363aba879501065f/pydantic/_internal/_validators.py

            _, digit_tuple, exponent = value.as_tuple()
            if not self.allow_inf_nan and exponent in {'F', 'n', 'N'}:
                raise PydanticKnownError('finite_number')

            if self.check_digits:
                if exponent >= 0:

But, looks like we cannot narrow exponent to be int in this case :(

Code in psycopg is actually correct (it has type: ignore that will be removed): https://github.com/psycopg/psycopg/blob/a8168869b762c821e7f07d1fc52d3642f899711e/psycopg/psycopg/types/numeric.py#L393-L397