Repeated generation migration

[panla]

Describe the bug

Repeated generation migration, remove id and then create id

To Reproduce

model

class Project(Model):
    id = fields.BigIntField(primary_key=True, source_field='id')

    name = fields.CharField(max_length=64, null=False, description='名称')
    desc = fields.CharField(max_length=1024, null=False, default='', description='描述')

    class Meta:
        app = 'default_db'
        table = 'projects'
        table_description = '项目表'

        indexes = (('name',),)

tortoise makemigrations

get 0001_initial.py

from tortoise import migrations
from tortoise.migrations import operations as ops
from orjson import loads
from tortoise.fields.base import OnDelete
from tortoise.fields.data import JSON_DUMPS
from tortoise import fields
from tortoise.indexes import Index

class Migration(migrations.Migration):
    initial = True

    operations = [
        
        ops.CreateModel(
            name='Project',
            fields=[
                ('name', fields.CharField(description='名称', max_length=64)),
                ('desc', fields.CharField(default='', description='描述', max_length=1024)),
            ],
            options={'table': 'projects', 'app': 'default_db', 'indexes': [Index(fields=['name'])], 'pk_attr': 'id', 'table_description': '项目表'},
            bases=['Model'],
        ),
        
    ]

run

tortoise migrate

create tables

no change model define code and run makemigrations

tortoise makemigrations

get 0002_auto_20260408_1716.py

from tortoise import migrations
from tortoise.migrations import operations as ops
from tortoise import fields
from tortoise.indexes import Index

class Migration(migrations.Migration):
    dependencies = [('default_db', '0001_initial')]

    initial = False

    operations = [
        ops.RemoveField(model_name='Project', name='id'),
        ops.AddField(
            model_name='Project',
            name='id',
            field=fields.BigIntField(source_field='id', generated=True, primary_key=True, unique=True, db_index=True),
        ),
        ops.AddIndex(
            model_name='Project',
            index=Index(fields=['id']),
        ),
    ]

remove pk, id and create pk, id and index

and.

if

id = fields.BigIntField(primary_key=True)

gen normal

ops.CreateModel(
            name='Project',
            fields=[
                ('id', fields.BigIntField(generated=True, primary_key=True, unique=True, db_index=True)),
                ('name', fields.CharField(description='名称', max_length=64)),
                ('desc', fields.CharField(default='', description='描述', max_length=1024)),
            ],
            options={'table': 'projects', 'app': 'default_db', 'indexes': [Index(fields=['name'])], 'pk_attr': 'id', 'table_description': '项目表'},
            bases=['Model'],
        ),

will normal

Expected behavior

gen normal migration

Additional context

tortoise-orm 1.1.7
mysql: 5.7

[Karmenzind]

Same issue here.
Tortoise removed id fields from many tables and added them back.

tortoise-orm 1.1.7
postgresql 16.6

[seladb]

@panla @Karmenzind this is because you defined source_field="id". If you remove it or change source_field=<NAME_DIFFERENT_FROM_FIELD_NAME> it'd work, for example:

class Project(Model):
    id = fields.BigIntField(primary_key=True, source_field='my_id')

    name = fields.CharField(max_length=64, null=False, description='名称')
    desc = fields.CharField(max_length=1024, null=False, default='', description='描述')

    class Meta:
        app = 'default_db'
        table = 'projects'
        table_description = '项目表'

        indexes = (('name',),)

Or:

class Project(Model):
    id = fields.BigIntField(primary_key=True)

    name = fields.CharField(max_length=64, null=False, description='名称')
    desc = fields.CharField(max_length=1024, null=False, default='', description='描述')

    class Meta:
        app = 'default_db'
        table = 'projects'
        table_description = '项目表'

        indexes = (('name',),)

I'm still trying to understand the bug when defining source_field="id" (which is the field name), but source_field is redundant in this case.

[Karmenzind]

@seladb Thanks. But I didn’t define source_field explicitly.

In my case, the issue appears when I set generated=False, or when I define a default factory function, for example:

id = fields.BigIntField(primary_key=True, generated=False) 

# or:

id = fields.BigIntField(primary_key=True, default=some_factory) 

FYI.

[panla]

Even if the table definition remains unchanged, this situation can still occur.
@Karmenzind

[Karmenzind]

Even if the table definition remains unchanged, this situation can still occur.
@Karmenzind

yah. same here

[seladb]

@seladb Thanks. But I didn’t define source_field explicitly.

In my case, the issue appears when I set generated=False, or when I define a default factory function, for example:

id = fields.BigIntField(primary_key=True, generated=False) 

# or:

id = fields.BigIntField(primary_key=True, default=some_factory) 

FYI.

@Karmenzind I managed to reproduce it with generated=False but not with default=some_factory. I assume it's related to the fact that this is a primary index field...

Even if the table definition remains unchanged, this situation can still occur.
@Karmenzind

@panla I'm not sure what you mean 🤔

[Karmenzind]

I guess @panla means that even after the initial migration is applied for models containing those ID fields, makemigration keeps generating migrations for those fields despite no actual changes being made. @seladb

[seladb]

I think I found the bug! I opened this PR with the 2 fixes: #2195

[panla]

我猜@panla这意味着即使在对包含这些 ID 字段的模型应用了初始迁移之后，makemigration尽管没有进行任何实际更改，它仍然会继续为这些字段生成迁移。@seladb

My situation is that I will keep the model definition unchanged,
If source_field is used, then after the migration is executed,
The bug of repeatedly generating migration files will occur during the second generation of migration files,
If the source_field is not set, it is normal