Orphan polymorphic relations get assigned to random DataObject class

[maxime-rainville]

Module version(s) affected

5.x-dev ... but probably every version of 5

Description

Orphan Polymorphic has_one get attached to the first DataObject class available.

How to reproduce

1. Run the scrip below.

Expected results: PolyHasOne comes back null
Actual results: An unsaved Bar object is returned. (Can be a different DataObject ... dependes what the first DataObject to be discovered is).

<?php

use SilverStripe\ORM\DataObject;
use SilverStripe\Dev\BuildTask;

class Bar extends DataObject {

    private static array $has_one = [
        'PolyHasOne' => DataObject::class,
        'RegularHasOne' => Page::class
    ];
}

class MultiRelation extends BuildTask
{
    public function run($request)
    {
        echo "Creating a Foo object with a Bar object\n";
        $bar = Bar::create();
        $bar->write();
        $bar = Bar::get()->byID($bar->ID);
        echo $bar->debug();
        echo $bar->PolyHasOne->debug();
        echo $bar->RegularHasOne->debug();
    }
}

Possible Solution

No response

Additional Context

I think this happens because the Class column for polymorphic relations are enum and default to whatever the first DataObject class in the enum is.

I'm not actually sure to what degree this is a problem. It's definitely confusing. But what the expected behaviour should be in this situation is also not 100% clear to me.

Related card

• RFC: Improve behaviour of unsaved has_one / belongs_to relations #7818

[GuySartorelli]

Expected results: PolyHasOne comes back null

That's not how has_one relations work usually - regular (non-polymorphic) has_one relations give you a singleton of the relevant class. There is a lot of code out there that assumes a dummy record like this will be provided when requesting a non-existent has_one relation so we can't just return null all of a sudden.

i.e. there's a lot of if ($this->MyHasOne()->exists()) {/* do something */} in the wild which would result in exceptions being thrown if we made that change.

[GuySartorelli]

One solution is to always return null for any missing has_one relation in a future major release, which I'd be in favour of.

[maxime-rainville]

I agree that's not clear what the correct outcome should be here. The very confusing thing here is that you'll get a completely random DataObject class on your Polymoprhic has one.

On the related question of what should you get when trying to access an has_one without a value, I think returning null would make more sense. I guess it's a bit better that you are at least getting an instance of the relation's Class.

I'm inclined to call this a CMS6 problem for now.