Cache lookup_field result to avoid repeated evaluation in inlines

[alexxxxey]

Added simple caching for lookup_field() in UnfoldAdminReadonlyField.
Without it, custom computed fields in inlines are re-evaluated multiple times per render, causing performance issues or side effects.

[lukasvinclav]

Thanks for PR. I'm just wiring you that I will take a look at this.

[lukasvinclav]

Can you provide more information about the side effects or performance issues? I'm already checking the implementation and I'm trying to find edge cases which I have to test. For example if you have an issue with duplicated SQL queries, provide instructions how to can I replicate because on the demo I'm not able to find anything.

[lukasvinclav]

btw I drafted another PR which will probably fix your issue: #1651

[alexxxxey]

For example, i have this inline

class AdRawFieldInline(TabularInline):
    model = AdRawField
    extra = 0
    fields = ['key_cleaned', 'value', 'value_cleaned', 'get_correlation_human']
    readonly_fields = ['get_correlation_human', 'key_cleaned', 'value_cleaned']
    def get_correlation_human(self, obj):
        res = []
        ans = ObjectType.objects.all()
        print('>>>>>>> DB HIT <<<<<<<<')
        return ans

I add prints in

class AdminReadonlyField:
    def __init__(self, form, field, is_first, model_admin=None):
        # Make self.field look a little bit like a field. This means that
        # {{ field.name }} must be a useful class name to identify the field.
        # For convenience, store other field-related data here too.
        print(f'=========== {field} ===============================')
...

and in each is_json, is_image etc.. Also add prints

...
        print('is_json', field, obj,)
        try:
            f, attr, value = lookup_field(field, obj, model_admin)
        except (AttributeError, ValueError, ObjectDoesNotExist):
            return False
....

And in console output:

=========== get_correlation_human ===============================
is_json get_correlation_human attributes: Windows - Tinted rear
>>>>>>> DB HIT <<<<<<<<
is_image get_correlation_human attributes: Windows - Tinted rear
>>>>>>> DB HIT <<<<<<<<
contents get_correlation_human attributes: Windows - Tinted rear
>>>>>>> DB HIT <<<<<<<<
is_file get_correlation_human attributes: Windows - Tinted rear
>>>>>>> DB HIT <<<<<<<<
contents get_correlation_human attributes: Windows - Tinted rear
>>>>>>> DB HIT <<<<<<<<
=========== url ===============================
....

I noticed that lookup_field is called multiple times for the same readonly field (e.g. get_correlation_human), which triggers repeated DB hits within one inline rendering cycle.
In my PR I added a minimal caching mechanism around lookup_field to prevent redundant calls.

I intentionally kept the change very localized — without refactoring the overall logic — just to highlight the root cause and avoid breaking anything unintentionally. :))

Your proposed PR looks cleaner and more correct architecturally, but I haven’t had a chance to test it yet.