Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fixed #373 -- Added CompositePrimaryKey-based Meta.primary_key. #18056

Open
wants to merge 71 commits into
base: main
Choose a base branch
from

Conversation

csirmazbendeguz
Copy link

@csirmazbendeguz csirmazbendeguz commented Apr 7, 2024

Trac ticket number

ticket-373

Branch description

Proposal
Previous PR

class Tenant(models.Model):
    pass


class User(models.Model):
    tenant = models.ForeignKey(Tenant, on_delete=models.CASCADE)
    id = models.SmallIntegerField()

    class Meta:
        primary_key = ("tenant_id", "id")


class Comment(models.Model):
    tenant = models.ForeignKey(Tenant, on_delete=models.CASCADE)
    id = models.SmallIntegerField()
    user_id = models.SmallIntegerField()
    user = models.ForeignObject(
        User,
        on_delete=models.CASCADE,
        from_fields=("tenant_id", "user_id"),
        to_fields=("tenant_id", "id"),
        related_name="+",
    )

    class Meta:
        primary_key = ("tenant_id", "id")

Checklist

  • This PR targets the main branch.
  • The commit message is written in past tense, mentions the ticket number, and ends with a period.
  • I have checked the "Has patch" ticket flag in the Trac system.
  • I have added or updated relevant tests.
  • I have added or updated relevant docs, including release notes if applicable.
  • For UI changes, I have attached screenshots in both light and dark modes.

@grjones
Copy link

grjones commented Apr 17, 2024

I was trying out this exciting branch and ran into this error when running a test:

<...>/lib/python3.12/site-packages/django/db/models/lookups.py:30: in __init__
    self.rhs = self.get_prep_lookup()
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

self = TupleIn(<django.db.models.fields.composite.Cols object at 0x107560980>, <django.db.models.sql.query.Query object at 0x1074e23f0>)

    def get_prep_lookup(self):
        if not isinstance(self.lhs, Cols):
            raise ValueError(
                "The left-hand side of the 'in' lookup must be an instance of Cols"
            )
        if not isinstance(self.rhs, Iterable):
>           raise ValueError(
                "The right-hand side of the 'in' lookup must be an iterable"
            )
E           ValueError: The right-hand side of the 'in' lookup must be an iterable

The issue stems from the use of isnull like so:

MyModel.objects.filter(
    type_override__severity__isnull=False
).update(severity="high")

Curious if anyone ran into this as well.

Edited for traceback:

<...>
lib/python3.12/site-packages/django/db/models/sql/compiler.py:2080: in pre_sql_setup
    self.query.add_filter("pk__in", query)
lib/python3.12/site-packages/django/db/models/sql/query.py:1601: in add_filter
    self.add_q(Q((filter_lhs, filter_rhs)))
lib/python3.12/site-packages/django/db/models/sql/query.py:1617: in add_q
    clause, _ = self._add_q(q_object, self.used_aliases)
lib/python3.12/site-packages/django/db/models/sql/query.py:1649: in _add_q
    child_clause, needed_inner = self.build_filter(
lib/python3.12/site-packages/django/db/models/sql/query.py:1563: in build_filter
    condition = self.build_lookup(lookups, col, value)
lib/python3.12/site-packages/django/db/models/sql/query.py:1393: in build_lookup
    lookup = lookup_class(lhs, rhs)
lib/python3.12/site-packages/django/db/models/lookups.py:30: in __init__
    self.rhs = self.get_prep_lookup()

So, this is part of SQLUpdateCompiler and is coming from the update code path.

@csirmazbendeguz
Copy link
Author

csirmazbendeguz commented Apr 18, 2024

Thanks for testing and reporting the issue @grjones! Indeed, I forgot to handle this use case. I'll look into it this week.

@csirmazbendeguz csirmazbendeguz force-pushed the ticket_373 branch 2 times, most recently from 6a26b19 to c75dcdd Compare April 19, 2024 12:22
@csirmazbendeguz
Copy link
Author

@grjones, FYI I pushed the fix

@grjones
Copy link

grjones commented Apr 20, 2024

@grjones, FYI I pushed the fix

Nice! I hope this gets merged in soon. Your branch has been working great for me.

@grjones
Copy link

grjones commented Apr 22, 2024

I may have found one other small issue. When adding a regular primary_key=True on a single field, a unique constraint is added. But when using this branch, it becomes an IntegrityError instead. Adding a UniqueConstraint on the composite fields is a work-a-round but ideally would be captured in this PR. Imo, this PR is sooooo close. I'm excited for it to be merged in.

@csirmazbendeguz
Copy link
Author

@grjones , thanks, I appreciate the feedback, I'll look into it. If a model defines Meta.primary_key, defining primary_key=True on a field should not be possible - could you give me a code example so I know how to reproduce the issue? I didn't know Django added unique constraints to primary keys, I'll check, but isn't that redundant?

@grjones
Copy link

grjones commented Apr 23, 2024

@grjones , thanks, I appreciate the feedback, I'll look into it. If a model defines Meta.primary_key, defining primary_key=True on a field should not be possible - could you give me a code example so I know how to reproduce the issue? I didn't know Django added unique constraints to primary keys, I'll check, but isn't that redundant?

I'll see if I can give you a solid failing test. My "unique constraint" phrasing might not be exactly right. But ultimately, I believe Django queries the DB first to see if the new object's PK already exists and throws a validation error. The composite key logic doesn't seem to be doing that and so an unhandled IntegrityError is raised instead.

@csirmazbendeguz
Copy link
Author

csirmazbendeguz commented May 1, 2024

@grjones , sorry for the late reply, I've been busy last week. Could you give me more specifics? What's the error message you expect?

@grjones
Copy link

grjones commented May 2, 2024

@grjones , sorry for the late reply, I've been busy last week. Could you give me more specifics? What's the error message you expect?

Actually, I think it's mostly ok. I was using Django Spanner and it's just not quite working with composite keys and will need to be fixed there. I wrote this and it passed. It probably shouldn't say Id though?

from django.core.exceptions import ValidationError
from django.test import TestCase

from .models import Tenant, User


class CompositePKCleanTests(TestCase):
    """
    Test the .clean() method of composite_pk models.
    """

    @classmethod
    def setUpTestData(cls):
        cls.tenant = Tenant.objects.create()

    def test_validation_error_is_raised_when_pk_already_exists(self):
        test_cases = [
            {"tenant": self.tenant, "id": 2412, "email": "user2412@example.com"},
            {"tenant_id": self.tenant.id, "id": 5316, "email": "user5316@example.com"},
            {"pk": (self.tenant.id, 7424), "email": "user7424@example.com"},
        ]
        expected = "{'id': ['User with this Id already exists.']}"
        for fields in test_cases:
            User.objects.create(**fields)
            with self.assertRaisesMessage(ValidationError, expected):
                User(**fields).clean()

Copy link
Contributor

@LilyFoote LilyFoote left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is a great start!

I've left a bunch of ideas for improvement. Feel free to push back if you think I'm wrong about anything.

django/db/models/base.py Outdated Show resolved Hide resolved
django/db/models/fields/composite.py Outdated Show resolved Hide resolved
django/db/models/fields/composite.py Outdated Show resolved Hide resolved
django/db/models/fields/composite.py Outdated Show resolved Hide resolved
django/db/models/fields/composite.py Outdated Show resolved Hide resolved
tests/composite_pk/test_get.py Outdated Show resolved Hide resolved
tests/composite_pk/test_get.py Outdated Show resolved Hide resolved
tests/composite_pk/test_update.py Outdated Show resolved Hide resolved
tests/composite_pk/tests.py Outdated Show resolved Hide resolved
tests/composite_pk/tests.py Outdated Show resolved Hide resolved
@csirmazbendeguz
Copy link
Author

Thank you so much for taking the time to review my changes @LilyFoote !
I have two questions:

  1. If Meta.primary_key is defined, this PR will automatically add a composite field called primary_key to the model. What do you think about this approach? I felt like it was easier to handle the composite primary keys this way as we can run checks against the meta class instead of traversing the model's fields for a composite field.
  2. I wrote a lot of tests testing the underlying queries made by the ORM. It makes a lot of sense to me, but I haven't seen this type of tests that much in the Django source code - do these tests look okay to you?

@LilyFoote
Copy link
Contributor

If Meta.primary_key is defined, this PR will automatically add a composite field called primary_key to the model. What do you think about this approach?

I don't feel strongly that this is better or worse than another option here, so happy to go with what you think is best.

I wrote a lot of tests testing the underlying queries made by the ORM. It makes a lot of sense to me, but I haven't seen this type of tests that much in the Django source code - do these tests look okay to you?

I like your tests quite a bit - they're pretty readable and comprehensive. The main issue I have with them is that they're written for specific databases instead of for generic database features. Where possible Django strongly prefers to test based on features because then the tests apply to as many databases as possible (including third party database libraries). I think the asserts of the actual SQL might be a bit tricky to adapt though, so we might need a different way to check what they're checking.

Also, after I reviewed yesterday, I thought of some more things:

  • We should add migrations tests to make sure that adding/removing Meta.primary_key works correctly and that removing a field that's part of a primary key also does something appropriate.
  • We might want tests for composite keys in forms and the admin. Maybe there's other areas too that we need to check the interactions.

Comment on lines 26 to 32
user = models.ForeignObject(
User,
on_delete=models.CASCADE,
from_fields=("tenant_id", "user_id"),
to_fields=("tenant_id", "id"),
related_name="+",
)
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

yeah I think this should be a stretch goal to get it working. See the comment above about MultiColSource.

return compiler.compile(WhereNode(exprs, connector=OR))


class Cols(Expression):
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I wonder if there is an opportunity to merge this TuplesIn, Cols, and friends logic with MultiColSource so it's less of an 👽. They both do very similar thing.

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks @charettes , I'll need to look into this, I wasn't aware.

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I merged Cols with MultiColSource (ad51da4) however, I'm not sure this is correct.

As far as I understand, MultiColSource was meant to represent columns in a JOIN, and as such, it has a sources field. Cols, on the other hand, was meant to represent a list of columns and it doesn't need a sources field. WDYT?

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I reverted back to Cols. Please resolve if you agree.

Copy link
Contributor

@LilyFoote LilyFoote left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I have two minor points about error message wording. They're not urgent to address.

Comment on lines 87 to 89
ValueError, "'primary_key' must be a tuple or a list"
):
Comment.objects.get(pk=1)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It would be nice if this could be "'pk' must be a tuple or a list".

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't think this is possible unfortunately, pk is translated to the field

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

How about this?

'exact' lookup of 'primary_key' field must be a tuple or a list

)
if not all(len(self.lhs) == len(vals) for vals in self.rhs):
raise ValueError(
"The left-hand side and right-hand side of the 'in' lookup must "
"have the same number of elements"
f"'{self.lhs.field.name}' must have {len(self.lhs)} elements"
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't think this is quite clear. If the pk has two component fields, but the lookup is pk__in=([1, 2], [3]) we will raise an error 'pk' must have 2 elements but the tuple does have two elements. I'm not sure what to suggest instead though - it's tricky...

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I added some changes, let me know what you think

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

How about this?

'exact' lookup of 'primary_key' field must have 2 elements each

@csirmazbendeguz
Copy link
Author

csirmazbendeguz commented May 20, 2024

I'm not sure what you mean by that, I don't think we can, because Options.pk refers to the field, while Options.primary_key is the list of field names.

You're completely right. In this case is pk set to CompositePrimaryKey when Meta.primary_key is defined and is primary_key set when a non-composite primary is used as well?

It would not be set, if it's a regular primary key, Meta.primary_key is None.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
5 participants