From 419f05976a22b83a1b3af9317edf94573fa6f65c Mon Sep 17 00:00:00 2001 From: Mohammad Sohail Shaikh <122166523+sohailshk@users.noreply.github.com> Date: Sun, 8 Nov 2026 22:02:48 +0530 Subject: [PATCH] Allow adding a partition field with the name of a removed one update_spec().remove_field(name).add_field(..., name) raised "Cannot add duplicate partition field name" because add_field checked whether the new field's id was being removed instead of the existing field's id. Replacing a partition field while keeping its name, for example to change the number of buckets, works in Java. Check the existing field instead, and on v1 tables rename the removed field, which stays in the spec as a void transform, to name_ as Java does. --- pyiceberg/table/update/spec.py | 7 ++++++- tests/table/test_init.py | 20 ++++++++++++++++++++ 2 files changed, 26 insertions(+), 1 deletion(-) diff --git a/pyiceberg/table/update/spec.py b/pyiceberg/table/update/spec.py index 5b55d96b19..f04fe69038 100644 --- a/pyiceberg/table/update/spec.py +++ b/pyiceberg/table/update/spec.py @@ -111,13 +111,18 @@ def add_field( self._transform_to_added_field[transform_key] = new_field existing_partition_field = self._name_to_field.get(new_field.name) - if existing_partition_field and new_field.field_id not in self._deletes: + if existing_partition_field and existing_partition_field.field_id not in self._deletes: if isinstance(existing_partition_field.transform, VoidTransform): self.rename_field( existing_partition_field.name, existing_partition_field.name + "_" + str(existing_partition_field.field_id) ) else: raise ValueError(f"Cannot add duplicate partition field name: {existing_partition_field.name}") + elif existing_partition_field: + # The removed field becomes a void field on v1 tables, so it gives up its name + self._renames[existing_partition_field.name] = ( + existing_partition_field.name + "_" + str(existing_partition_field.field_id) + ) self._name_to_added_field[new_field.name] = new_field self._adds.append(new_field) diff --git a/tests/table/test_init.py b/tests/table/test_init.py index db6dc00e39..7bfaca7311 100644 --- a/tests/table/test_init.py +++ b/tests/table/test_init.py @@ -97,6 +97,7 @@ from pyiceberg.transforms import ( BucketTransform, IdentityTransform, + VoidTransform, ) from pyiceberg.typedef import Record from pyiceberg.types import ( @@ -1546,6 +1547,25 @@ def test_update_spec_remove_and_add_same_field_reuses_id(table: Table) -> None: assert txn.table_metadata.last_partition_id == 1000 +def test_update_spec_remove_and_add_field_with_same_name(table_v1: Table, table_v2: Table) -> None: + for table in [table_v1, table_v2]: + txn = table.transaction() + txn.update_spec().add_field("y", BucketTransform(16), "y_shard").commit() + txn.update_spec().remove_field("y_shard").add_field("y", BucketTransform(32), "y_shard").commit() + + if table.format_version == 1: + assert txn.table_metadata.spec().fields == ( + PartitionField(source_id=1, field_id=1000, transform=IdentityTransform(), name="x"), + PartitionField(source_id=2, field_id=1001, transform=VoidTransform(), name="y_shard_1001"), + PartitionField(source_id=2, field_id=1002, transform=BucketTransform(32), name="y_shard"), + ) + else: + assert txn.table_metadata.spec().fields == ( + PartitionField(source_id=1, field_id=1000, transform=IdentityTransform(), name="x"), + PartitionField(source_id=2, field_id=1002, transform=BucketTransform(32), name="y_shard"), + ) + + def test_assert_default_spec_id(table_v2: Table) -> None: base_metadata = table_v2.metadata AssertDefaultSpecId(default_spec_id=base_metadata.default_spec_id).validate(base_metadata)