simple-history 2.6.0 generates a migration while 2.5.1 doesn't

[atodorov]

Describe the bug
Kiwi TCMS uses simple-history as dependency. Until recently we had version 2.5.1 and we've upgraded to 2.6.0. The problem is that with 2.5.1 ./manage.py migrate/makemigrations doesn't detect any new changes but with 2.6.0 it does.

Here's how the initial migration looks like:

from django.db import migrations, models

        migrations.AddField(
            model_name='testplan',
            name='parent',
            field=models.ForeignKey(blank=True, null=True, on_delete=models.deletion.CASCADE,
                                    related_name='child_set', to='testplans.TestPlan'),
        ),

        migrations.CreateModel(
            name='HistoricalTestPlan',
            fields=[
                ('parent', models.ForeignKey(blank=True, db_constraint=False, null=True,
                                             on_delete=models.deletion.DO_NOTHING, related_name='+',
                                             to='testplans.TestPlan')),
            ]
        )

Here's what makemigrations produces with 2.6.0:

import django.db.models.deletion

        migrations.AlterField(
            model_name='historicaltestplan',
            name='parent',
            field=models.ForeignKey(blank=True, db_constraint=False, null=True, on_delete=django.db.models.deletion.DO_NOTHING, related_name='+', to='testplans.HistoricalTestPlan'),
        ),

The only difference I see is how we reference the on_delete constraint.

To Reproduce
Steps to reproduce the behavior:

1. git clone https://github.com/kiwitcms/Kiwi/
2. pip install -r requirements/devel.txt
3. ./manage.py migrate
4. ./manage.py makemigrations <--- produces a new migration

Repeat the above with 2.5.1 instead and the last step tells you there are no changes detected

Environment (please complete the following information):

• OS: Linux, RHEL 7
• Django Simple History Version: 2.5.1 and 2.6
• Django Version: 2.1.5
• Database Version: SQLite

[rossmechanic]

Hey @atodorov thanks for this. There seems to be a difference in the to field. My guess is that the issue stems from #462 and I'm looking into it now. Will let you know when I have an update

[rossmechanic]

Found the issue @atodorov ! It's this line here https://github.com/treyhunner/django-simple-history/pull/462/files#diff-db0d12c32869d48141bfca2c94659322L203. In adding the deconstruct method to that code, we accidentally got rid of that line. DSH still supports foreign keys to the same object, but we accidentally got rid of support for using 'self' as the ForeignKey (as you do here https://github.com/kiwitcms/Kiwi/blob/master/tcms/testplans/models.py#L49). I'll get a fix together for this, but if you change 'self' to 'TestPlan' in line 49, that will also fix the issue on your end. Let me know if that helps!

[atodorov]

Found the issue @atodorov !

@rossmechanic thank you for the lightning fast reaction!

I'll get a fix together for this, but if you change 'self' to 'TestPlan' in line 49, that will also fix the issue on your end. Let me know if that helps!

This WA works for me and doesn't produce any more migrations.

Unfortunately we do have a live version of Kiwi TCMS out there which customers have deployed and are hitting this issue. As we are shipping with pinned dependencies in our Docker image even if you do release a fix there's not much to be done on our end except publishing a new docker image. Sigh.

[rossmechanic]

@atodorov sorry about that! That should have been caught in automated tests, but evidently whoever wrote that feature originally didn't test it... I do have a PR up that should be merged by tomorrow, and I've been planning to have a new release by end of week. Sorry for the additional work 😞