ChoiceField metadata does not support grouped choices.

[maxpeterson]

If a serialiser has a ChoiceField with grouped choices:

TYPE_CHOICES = (
        ('Default', (
            (1, 'Option 1'),
            (2, 'Option 2'),
            (3, 'Option 3'))),
        (4, 'Option 4'))

The OPTIONS metadata for the field includes the choices:

[
    {
        "value": "Default",
        "display_name": "((1, 'Option 1'), (2, 'Option 2'), (3, 'Option 3'))"
    },
    {
        "value": 4,
        "display_name": "Option 4"
    }
]

I am not certain what the behaviour should be, but something including the nested choices would be useful:

[
    {
        "choices": [
            {
                "value": 1,
                "display_name": "Option 1"
            },
            {
                "value": 2,
                "display_name": "Option 2"
            },
            {
                "value": 3,
                "display_name": "Option 3"
            }
        ],
        "display_name": "Default"
    },
    {
        "value": 4,
        "display_name": "Option 4"
    }
]

If this is not appropriate then a flat list of choices might be appropriate.

[
    {
        "value": 1,
        "display_name": "Option 1"
    },
    {
        "value": 2,
        "display_name": "Option 2"
    },
    {
        "value": 3,
        "display_name": "Option 3"
    },
    {
        "value": 4,
        "display_name": "Option 4"
    }
]

[tomchristie]

The flat list of choices would be my preferred option here.

[maxpeterson]

@tomchristie I was hoping you would go for the hierarchical one as I need the hierarchy in an application consuming the api.

If the choices are flat, then how would you recommend we get the hierarchy from the api. My first idea would be to extend SimpleMetadata globally (or on my view) to include an extra attribute for the field.

[tomchristie]

My first idea would be to extend SimpleMetadata globally (or on my view) to include an extra attribute for the field.

Exactly so, yes.

[maxpeterson]

Related to #1533

[tomchristie]

Rolling this into #1533.

[wernerhp]

@maxpeterson @tomchristie I also need the nested/grouped representation on to be returned by the meta for an app. Can you guys perhaps link me to an example of hot to do this?

[maxpeterson]

@wernerhp I have achieved this by adding an additional grouped_choices attribute to the meta. I am afraid the code is in a private repo so hopefully these "snippets" will help:

def get_choice(choice_value, choice_name):
    if isinstance(choice_name, (list, tuple)):
        return {
            'choices': [get_choice(*choice) for choice in choice_name],
            'display_name': force_text(choice_value, strings_only=True)
        }
    return {
        'value': choice_value,
        'display_name': force_text(choice_name, strings_only=True)
    }


def grouped_choices(choices_list):
    return [get_choice(*choice) for choice in choices_list]


class GroupedChoiceMetadata(SimpleMetadata):
    def get_field_info(self, field):
        field_info = super().get_field_info(field)

        if not field_info.get('read_only') and hasattr(field, 'choices_list'):
            field_info['grouped_choices'] = grouped_choices(field.choices_list)

        return field_info

[wernerhp]

@maxpeterson Thanks for the snippet. I'm still battling a bit.

choices_list doesn't seem to be a standard attribute of serializers.ChoiceField.
Would you minding including a snippet from your Serializer, Model and your nested/grouped MEDIA_CHOICES data structure.

Working with the Django example

REST_FRAMEWORK = {
...
    'DEFAULT_METADATA_CLASS': 'custom.metadata.GroupedChoiceMetadata',
...
}

class Media(models.Model):
    MEDIA_CHOICES = (
        ('Audio', (
                ('vinyl', 'Vinyl'),
                ('cd', 'CD'),
            )
        ),
        ('Video', (
                ('vhs', 'VHS Tape'),
                ('dvd', 'DVD'),
            )
        ),
        ('unknown', 'Unknown'),
    )

    type = models.CharField(max_length=255, choices=MEDIA_CHOICES)

class MediaSerializer(serializers.ModelSerializer):
    class Meta:
        fields = [
            'type',
        ]
        model = Media

printing out field.choices in GroupedChoiceMetadata.get_field_info yields an OrderedDict
OrderedDict([('vinyl', 'Vinyl'), ('cd', 'CD'), ('vhs', 'VHS Tape'), ('dvd', 'DVD'), ('unknown', 'Unknown')]) so the structure has already been lost/flattened.
DRF 3.4.0
Thanks in advance

[maxpeterson]

@wernerhp try using field.grouped_choices - Sorry about that choices_list was based on an earlier version I used

[maxpeterson]

Here is the PR #3225 grouped_choices was added and my original one #3108 (where I suggested choices_list)

[maxpeterson]

To clarify try ...

def get_choice(choice_value, choice_name):
    if isinstance(choice_name, (list, tuple)):
        return {
            'choices': [get_choice(*choice) for choice in choice_name],
            'display_name': force_text(choice_value, strings_only=True)
        }
    return {
        'value': choice_value,
        'display_name': force_text(choice_name, strings_only=True)
    }


def grouped_choices(choices_list):
    return [get_choice(*choice) for choice in choices_list]


class GroupedChoiceMetadata(SimpleMetadata):
    def get_field_info(self, field):
        field_info = super().get_field_info(field)

        if not field_info.get('read_only') and hasattr(field, 'grouped_choices'):
            field_info['grouped_choices'] = grouped_choices(field.grouped_choices)

        return field_info

[wernerhp]

After some effort I found that group_choices is actually a property on ChoiceField
This seems to do the trick

class GroupedChoiceMetadata(SimpleMetadata):
    def get_field_info(self, field):
        field_info = super(GroupedChoiceMetadata, self).get_field_info(field)

        if field_info.get('type') is 'choice':
            field_info['grouped_choices']= field.grouped_choices

        return field_info

I wonder why grouped_choices aren't returned by default?

@tomchristie Do you perhaps have some insights for me?

[tomchristie]

@wernerhp - Mostly likely simply historical. Expect they didn't exist when we added the metadata API.