Caution:
The code below was written for an older version of Django (before Custom
User Models were introduced). It contains a race condition, and
should only be used with a Transaction Isolation Level of SERIALIZABLE
and request-scoped transactions.
Your code won't work, as the attributes of field instances are read-only. I fear it might be a wee bit more complicated than you're thinking.
If you'll only ever create User instances with a form, you can define a custom ModelForm that enforces this behavior:
from django import forms
from django.contrib.auth.models import User
class UserForm(forms.ModelForm):
class Meta:
model = User
def clean_email(self):
email = self.cleaned_data.get('email')
username = self.cleaned_data.get('username')
if email and User.objects.filter(email=email).exclude(username=username).exists():
raise forms.ValidationError(u'Email addresses must be unique.')
return email
Then just use this form wherever you need to create a new user.
BTW, you can use Model._meta.get_field('field_name') to get fields by name, rather than by position. So for example:
# The following lines are equivalent
User._meta.fields[4]
User._meta.get_field('email')
UPDATE
The Django documentation recommends you use the clean method for all validation that spans multiple form fields, because it's called after all the <FIELD>.clean and <FIELD>_clean methods. This means that you can (mostly) rely on the field's value being present in cleaned_data from within clean.
Since the form fields are validated in the order they're declared, I think it's okay to occasionally place multi-field validation in a <FIELD>_clean method, so long as the field in question appears after all other fields it depends on. I do this so any validation errors are associated with the field itself, rather than with the form.
