piccolo migrations new --auto crashes when an altered column's previous class was a custom Column subclass #1428

Summary

SchemaDiffer.alter_columns looks up UniqueGlobalNames.COLUMN_<NAME> for both the new and the old column class. The lookup for the new class passes a None default; the one for the old class doesn't. Altering a column whose previous class was a user-defined Column subclass therefore raises AttributeError, and piccolo migrations new --auto dies before writing anything.

Reproduction

No database needed:

from piccolo.apps.migrations.auto.diffable_table import DiffableTable
from piccolo.apps.migrations.auto.schema_differ import SchemaDiffer
from piccolo.columns.column_types import Varchar


class CustomVarchar(Varchar):
    pass


old = CustomVarchar(length=10)
old._meta.name = "name"

new = Varchar(length=20)
new._meta.name = "name"

differ = SchemaDiffer(
    schema=[DiffableTable(class_name="Band", tablename="band", columns=[new])],
    schema_snapshot=[
        DiffableTable(class_name="Band", tablename="band", columns=[old])
    ],
    auto_input="y",
)
print(differ.alter_columns.statements)

Expected: one manager.alter_column(...) statement.

Actual:

Traceback (most recent call last):
  File "<stdin>", line 16, in <module>
  File ".../piccolo/apps/migrations/auto/schema_differ.py", line 516, in alter_columns
    expect_conflict_with_global_name=getattr(
                                     ^^^^^^^^
AttributeError: type object 'UniqueGlobalNames' has no attribute 'COLUMN_CUSTOMVARCHAR'. Did you mean: 'COLUMN_VARCHAR'?

Cause

schema_differ.py makes six getattr(UniqueGlobalNames, f"COLUMN_{...}") lookups. Five pass None as the default (lines 503, 594, 635, 725, 767 on master). The sixth - the old_column_class branch at line 516, twelve lines below its column_class twin - does not:

if alter_column.column_class is not None:
    ...
        expect_conflict_with_global_name=getattr(
            UniqueGlobalNames,
            f"COLUMN_{alter_column.column_class.__name__.upper()}",
            None,          # <- present
        ),

if alter_column.old_column_class is not None:
    ...
        expect_conflict_with_global_name=getattr(
            UniqueGlobalNames,
            f"COLUMN_{alter_column.old_column_class.__name__.upper()}",
        ),                 # <- missing

UniqueGlobalNames only carries the built-in column classes, so a miss is the expected outcome for any subclass a user defines - which the five other call sites already treat as normal.

Impact

Subclassing a column type is a supported and fairly common thing to do - wrapping Varchar to fix a length, or Timestamptz to add a server-side default, so the constraint lives in one place rather than at every declaration. Any project that does this loses --auto entirely the first time such a column changes, with a traceback that points at global-name bookkeeping rather than at the column involved. The workaround is to hand-write the migration.

Environment

• Piccolo master @ 6ef37514, and the 1.36.0 release
• Python 3.12
• Database-independent: SchemaDiffer raises before any engine is used

Suggested fix

Add the None default so both branches behave identically. I have a branch with that one-line change plus a regression test in tests/apps/migrations/auto/test_schema_differ.py (styled on the neighbouring test_alter_column_precision); tests/apps/migrations and tests/columns pass, as do isort, black and flake8.
See #1429