Allow setting non-string typed values in set_properties

[kevinjqliu]

Requires #503 to be merged first, extension of #502.

This PR modifies Transaction's set_properties API to accept table properties as type Dict[str, Any], either as dictionary or kwargs.

For example,

with table.transaction() as transaction:
    transaction.set_properties(abc=123)
    transaction.set_properties({"abc": 123})

This is done by reusing the field validator defined in #469. Setting None property value is disallowed since the transformation will change None to str "None" which is unintuitive.

Note

The Iceberg spec stores table properties as strings. That is why when setting a property value as int, reading it back produces a string.

table = table.transaction().set_properties({"abc": 123}).commit_transaction()
assert table.properties.get("abc") == "123"

Use PropertyUtil.property_as_int to return property value as int

table = table.transaction().set_properties({"abc": 123}).commit_transaction()
assert PropertyUtil.property_as_int("abc") == 123

[HonahX]

This looks great! @kevinjqliu . It adds lots of convenience when setting properties through transaction.

[HonahX]

Suggested change

	transform_properties_dict_value_to_str = field_validator('updates', mode='before')(transform_dict_value_to_str)
	@field_validator('updates', mode='before')
	def transform_properties_dict_value_to_str(cls, properties: Properties) -> Dict[str, str]:
	return transform_dict_value_to_str(properties)

Shall we reformat the validator to use decorator? This aligns with the way we define model validators in metadata.py and also the example in Pydantic doc. WDYT?

[kevinjqliu]

looks much cleaner, thank you!

[kevinjqliu]

rebased after #503

[HonahX]

Sorry for being late. Overall LGTM. Just have one small comment regarding the type.

BTW, could you please update the same field validator in metadata to use the decorator @field_validator?
https://github.com/apache/iceberg-python/blob/main/pyiceberg/table/metadata.py#L224 Thanks!

[HonahX]

I think this one should be Dict[str, str] because at the time when the model is constructed, the before validator has converted all values to str

[Fokko]

Thanks @kevinjqliu for working on this, and @HonahX for the review!