`compare_server_default` doesn't work with a string column that contains a backslash in its default value

[marcinbarczynski]

Describe the bug
If server_default of a string column contains a backslash, alembic generates op.alter_column migration despite the fact that the database and model are consistent.

Expected behavior
If the database and model are consistent, alembic revision --autogenerate should create an empty migration.

To Reproduce

1. Create mymodel.py:

from sqlalchemy.ext.declarative import declarative_base
from sqlalchemy import String, Column, Integer

Base = declarative_base()

class Demo(Base):
    __tablename__ = 'demo'
    key = Column(Integer, primary_key=True)
    regex = Column(String, nullable=False, default='abc\\..*', server_default='abc\\..*')

2. Run alembic init alembic
3. Set compare_server_default to True in env.py
4. Set up mymodel in env.py:

from mymodel import Base
target_metadata = Base.metadata

5. Run alembic revision --autogenerate. It created the following migration:

"""empty message

Revision ID: f5adea3b05e6
Revises: 
Create Date: 2022-12-22 16:15:14.111949

"""
from alembic import op
import sqlalchemy as sa


# revision identifiers, used by Alembic.
revision = 'f5adea3b05e6'
down_revision = None
branch_labels = None
depends_on = None


def upgrade() -> None:
    # ### commands auto generated by Alembic - please adjust! ###
    op.create_table('demo',
    sa.Column('key', sa.Integer(), nullable=False),
    sa.Column('regex', sa.String(), server_default='abc\\..*', nullable=False),
    sa.PrimaryKeyConstraint('key')
    )
    # ### end Alembic commands ###


def downgrade() -> None:
    # ### commands auto generated by Alembic - please adjust! ###
    op.drop_table('demo')
    # ### end Alembic commands ###

6. Upgrade database with alembic upgrade head
7. Run alembic revision --autogenerate again.

Error

The last alembic revision --autogenerate creates a non-empty migration file:
Due to a bug it generates a non-empty migration:

"""empty message

Revision ID: 198e4eb04d5c
Revises: f5adea3b05e6
Create Date: 2022-12-22 16:15:26.936522

"""
from alembic import op
import sqlalchemy as sa


# revision identifiers, used by Alembic.
revision = '198e4eb04d5c'
down_revision = 'f5adea3b05e6'
branch_labels = None
depends_on = None


def upgrade() -> None:
    # ### commands auto generated by Alembic - please adjust! ###
    op.alter_column('demo', 'regex',
               existing_type=sa.VARCHAR(),
               server_default='abc\\..*',
               existing_nullable=False)
    # ### end Alembic commands ###


def downgrade() -> None:
    # ### commands auto generated by Alembic - please adjust! ###
    op.alter_column('demo', 'regex',
               existing_type=sa.VARCHAR(),
               server_default=sa.text("'abc\\..*'::character varying"),
               existing_nullable=False)
    # ### end Alembic commands ###

The problem happens only if there is \ in the default string.

Versions.

• OS: Ubuntu 22.04
• Python: 3.10
• Alembic: 1.8.1
• SQLAlchemy: 1.4.42
• Database: PostgreSQL 13.9
• DBAPI: psycopg2==2.9.3

Additional context

[zzzeek]

the patch is approximately:

diff --git a/alembic/autogenerate/compare.py b/alembic/autogenerate/compare.py
index c9971ea..8301e34 100644
--- a/alembic/autogenerate/compare.py
+++ b/alembic/autogenerate/compare.py
@@ -1022,7 +1022,7 @@ def _render_server_default_for_compare(
     if isinstance(metadata_default, str):
         if metadata_col.type._type_affinity is sqltypes.String:
             metadata_default = re.sub(r"^'|'$", "", metadata_default)
-            return repr(metadata_default)
+            return f"'{metadata_default}'"
         else:
             return metadata_default
     else:

however there seem to be no case-oriented tests for many server default cases in test_autogen_diffs, sort of surprising as there was a lot of work done on this at some point , it looks like tests were added to dialect-specific like test_postgresql for Identity, things like that.

string tests should include different patterns of backslashes as well as single quotes that require SQL-escaping, with a round trip to create the DDL in the DB, get the value back, compare it to rendered from the metadata.

[sqla-tester]

Mike Bayer has proposed a fix for this issue in the main branch:

dont use repr to quote string in compare_server_default https://gerrit.sqlalchemy.org/c/sqlalchemy/alembic/+/4324

[marcinbarczynski]

Thanks! That was quick!