Possible regression since 1.12.0 on BIGSERIAL handling

[hugo-ricateau]

Describe the bug
Since alembic==1.12.0, auto-generation now detects a type change for the BIGSERIAL receipt

INFO  [alembic.autogenerate.compare] Detected type change from BIGINT() to BIGSERIAL() on '<table>.<column>'

Expected behavior
Not sure if it is a regression or a new feature; if not a bug, what is the expected way to handle it?

To Reproduce
This model

class BIGSERIAL(BigInteger):
    pass

@compiles(BIGSERIAL, postgresql.dialect.name)
def _serial(element, compiler, **kw):
    return 'BIGSERIAL'

class Table(Base):
    column = Column(BIGSERIAL, index=True, nullable=False, comment='Not a primary key')
    ...

were originally created as

op.create_table(
    'table',
    sa.Column('column', path.to.BIGSERIAL(), nullable=False),
    ...,
)

and since alembic==1.12.0, auto-generation now produces a

op.alter_column('table', 'column', existing_type=sa.BIGINT(), type_=path.to.BIGSERIAL(), existing_nullable=False)

So I assume it is an issue when detecting the existing type?

Versions

• OS: Cross-platform
• Python: 3.11.6
• Alembic: >=1.12.0
• SQLAlchemy: 1.4.53
• Database: PostgreSQL
• DBAPI: Psycopg

[zzzeek]

so BIGSERIAL is a pseudo-type in PostgreSQL which will come back in reflection as BIGINT. This is a custom type you're adding, so to have alembic ignore this you would need to write a custom rule to compare types

def my_compare_type(context, inspected_column,
            metadata_column, inspected_type, metadata_type):
    if isinstance(inspected_type, BIGINT):
        return False
    return None

context.configure(
    # ...
    compare_type = my_compare_type
)

[hugo-ricateau]

Indeed, I missed it:

Changed in version 1.12.0: The default value of EnvironmentContext.configure.compare_type has been changed to True.

Many thanks for your quick and very complete answer.