Cannot dynamically add fields alongside removal.

[mjsir911]

Hi,

per the dynamically modifying fields, I would like to modify the source attribute of a field at runtime via __init__

class TestSerializer(ModelSerializer):
    generic_test = IntegerField()
    def __init__(self, *args, direction, **kwargs):
        super().__init__(*args, **kwargs)
        self.fields['something'].source = f"{direction}_test"
    class Meta:
         fields = [ "generic_test" ]

Obviously I can't modify a class attribute for a single instance I want modified, so I move the attribute outside of class initialization:

class TestSerializer(ModelSerializer):
    def __init__(self, *args, direction, **kwargs):
        super().__init__(*args, **kwargs)
        self.fields['generic_test'] = IntegerField(source=f"{direction}_test")
    class Meta:
         fields = [ "generic_test" ]

but this errors out due to the field not being a class attribute at class-definition-time.

The current workaround:

class TestSerializer(ModelSerializer):
    generic_test = IntegerField()
    def __init__(self, *args, direction, **kwargs):
        super().__init__(*args, **kwargs)
        self.fields['generic_test'] = IntegerField(source=f"{direction}_test")
    class Meta:
         fields = [ "generic_test" ]

this repeats unnecessary code and doesn't signal properly that the initial class attribute is going to always be overwritten.

Is the documentation's "Dynamically modifying fields" only for filtering out subsets of fields? There's no real way to modify or add fields in __init__ currently.

[tomchristie]

I don't think I would trust that setting .source = ... would necessarily work as expected.
Perhaps it does, perhaps it doesn't.

I haven't looked into it, but it's certainly not intentional usage.

Personally if I really needed to do this I'd probably just change my outlook to be okay with this as a constraint.

class TestSerializer(ModelSerializer):
    generic_test = IntegerField()  # Note that we dynamically update this field in `__init__`

    def __init__(self, *args, direction, **kwargs):
        super().__init__(*args, **kwargs)
        self.fields['generic_test'] = IntegerField(source=f"{direction}_test")

    class Meta:
         fields = [ "generic_test" ]

this "repeats unnecessary code" but does "signal properly that the initial class attribute is going to always be overwritten."

Not perfect. But perfectly okay. 😌

[mjsir911]

Yeah, this is basically what I ended up doing with a DummyField() for better signalling, but I would consider it a bug that I can't assign new self.field[]s without causing errors.

The documentation makes it sound like more can be done with self.fields than just filtering, even if that's the only example provided.

[mjsir911]

Even something like an emergency hatch _declared_fields = [ 'generic_test' ] on the class would be helpful. This is a super fringe edge case and just providing a sane intended way to do this would be appreciated.