fix: Validate SetStatisticsUpdate correctly

[ragnard]

Closes #2865

Previously the pydantic @model_validator on SetStatisticsUpdate would fail because it assumed statistics was a model instance. In a "before"" validator that is not necessarily case.

Check type and handle both models and dicts instead.

Before

>>> import pyiceberg.table.update
>>> pyiceberg.table.update.SetStatisticsUpdate.model_validate({'statistics': {'snapshot-id': 1234, 'file-size-in-bytes': 0, 'statistics-path': '', 'file-footer-size-in-bytes': 0, 'blob-metadata': []}})
Traceback (most recent call last):
  File "<python-input-1>", line 1, in <module>
    pyiceberg.table.update.SetStatisticsUpdate.model_validate({'statistics': {'snapshot-id': 1234, 'file-size-in-bytes': 0, 'statistics-path': '', 'file-footer-size-in-bytes': 0, 'blob-metadata': []}})
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/ragge/projects/github.com/ragnard/iceberg-python/.venv/lib/python3.14/site-packages/pydantic/main.py", line 716, in model_validate
    return cls.__pydantic_validator__.validate_python(
           ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
        obj,
        ^^^^
    ...<5 lines>...
        by_name=by_name,
        ^^^^^^^^^^^^^^^^
    )
    ^
  File "/home/ragge/projects/github.com/ragnard/iceberg-python/pyiceberg/table/update/__init__.py", line 191, in validate_snapshot_id
    data["snapshot_id"] = stats.snapshot_id
                          ^^^^^^^^^^^^^^^^^
AttributeError: 'dict' object has no attribute 'snapshot_id'

After

>>> import pyiceberg.table.update
>>> pyiceberg.table.update.SetStatisticsUpdate.model_validate({'statistics': {'snapshot-id': 1234, 'file-size-in-bytes': 0, 'statistics-path': '', 'file-footer-size-in-bytes': 0, 'blob-metadata': []}})
SetStatisticsUpdate(action='set-statistics', statistics=StatisticsFile(snapshot_id=1234, statistics_path='', file_size_in_bytes=0, file_footer_size_in_bytes=0, key_metadata=None, blob_metadata=[]), snapshot_id=1234)

Rationale for this change

Are these changes tested?

Yes, but only using the two-liners above.

Are there any user-facing changes?

No.

[kevinjqliu]

Thanks for the PR. I think we should just deprecate the top-level snapshot_id entirely.
For context, the "before model_validator` was added in 3b53edc#diff-769b43e1d8beaa86141f694679de2bbea3604a65f987a9acd7d9e9efca193b7eR181-R193 to maintain backwards compatibility and prep for deprecation

[kevinjqliu]

oops didnt mean to approve

[ragnard]

@kevinjqliu Ok, do you want me to change the fix so that snapshot_id is still there, but just not automatically populated?

[kevinjqliu]

Looks like the java implementation has not deprecated the top-level snapshot_id. lets proceed with this change since it improves the current validation logic.

Thanks for the PR! Lets add a test and it should be good to go!

[kevinjqliu]

Suggested change

	@model_validator(mode="after")
	def validate_snapshot_id(self) -> Self:
	return self.model_copy(update={"snapshot_id": self.statistics.snapshot_id})
	@model_validator(mode="after")
	def validate_snapshot_id(self) -> Self:
	self.snapshot_id = self.statistics.snapshot_id
	return self

nit: use direct assignment

[ragnard]

Looks like the java implementation has not deprecated the top-level snapshot_id. lets proceed with this change since it improves the current validation logic.

Thanks for the PR! Lets add a test and it should be good to go!

@kevinjqliu Thanks for the (quick!) review. I've changed the fix a bit:

1. Since the model is frozen it is not possible to use direct assignment, but it is also not possible to use model_copy like I did because an "after" validator needs to return the same model instance. I've reverted to a "before" validator, but handle the dict case properly.

2. For testing, the issue is not really about tables with statistics, but that it was not possible to instantiate the SetStatisticsUpdate model from a dict. I added an model_roundtrips helper that is now used to check that any model roundtrips (model -> dict -> model) correctly.

Please let me know if you want further changes.

[kevinjqliu]

Thanks! Looks good, i found a small bug. Would be great to add a test for this case

[kevinjqliu]

Suggested change

	elif isinstance(stats, dict):
	snapshot_id = stats.get("snapshot_id")
	elif isinstance(stats, dict):
	snapshot_id = stats.get("snapshot_id")
	else:
	snapshot_id = None

nit: i think we can inline the else here

[kevinjqliu]

Suggested change

	snapshot_id = stats.get("snapshot_id")
	snapshot_id = stats.get("snapshot-id")

i think this should be snapshot-id since before validator takes in json as input

iceberg-python/pyiceberg/table/statistics.py

Lines 32 to 40 in fa03e08

	class StatisticsCommonFields(IcebergBaseModel):
	"""Common fields between table and partition statistics structs found on metadata."""
	snapshot_id: int = Field(alias="snapshot-id")
	statistics_path: str = Field(alias="statistics-path")
	file_size_in_bytes: int = Field(alias="file-size-in-bytes")
	class StatisticsFile(StatisticsCommonFields):

[kevinjqliu]

could you add a test case for this (and possibly one for the else case too)?

The current test only test the StatisticsFile instance branch

iceberg-python/tests/table/test_init.py

Lines 1370 to 1381 in fa03e08

	def test_set_statistics_update(table_v2_with_statistics: Table) -> None:
	snapshot_id = table_v2_with_statistics.metadata.current_snapshot_id
	blob_metadata = BlobMetadata(
	type="apache-datasketches-theta-v1",
	snapshot_id=snapshot_id,
	sequence_number=2,
	fields=[1],
	properties={"prop-key": "prop-value"},
	)
	statistics_file = StatisticsFile(

[ragnard]

good catch with the snapshot-id. I've fixed that, and added a test specifically for the snapshot_id handling, creating a model both from a model instance, dict and json. the roundtrip testing also exercises the previous bug.

[kevinjqliu]

LGTM!

[geruh]

LGTM!

[kevinjqliu]

Thank you @ragnard for the PR and thanks @geruh for the review!