What\'s a djangonautic way of handling default settings in an app if one isn\'t defined in settings.py
?
I\'ve currently placed a default_settings<
How about this?
In myapp/settings.py:
from django.conf import settings
FOO = 'bar'
BAR = 'baz'
_g = globals()
for key, value in _g.items():
_g[key] = getattr(settings, key, value)
In myapp/other.py:
import myapp.settings
print myapp.settings.FOO
Given this answer by ncoghlan, I feel ok using globals() this way.
Number 3 is best because it is the most simple one. And very consistent look.
Number 1: it is easy to overlook it. If I'll open your code and I won't scroll to the bottom I'll miss it and I will think that settings can't be overridden in my own module.
Number 2: is not only repetitive, it is harder to read because it is too long, also default values will be defined multiple time and scattered all over your code.
Number 4: non-consistent-look, repetitive calls.
Number 5: Non consistent, we expect settings to be defined in a module not in a class. Well at least I do expect to find to be defined as module because I've seen many apps using method 3, and I use it my self so I might be biased.
It seems that every solution I see there tends to create an internal copy of application settings, proxy, wrap or whatever. This is confusing and creates problems when settings are modified in run time like they do in tests.
To me all settings belong in django.conf.settings
and only there. You should not read them from somewhere else nor copy it for later use (as they may change). You should set them once and don't bother about defaults later on.
I understand the impulse to drop the app prefix when app setting is used internally, but this also is IMHO a bad idea. When in trouble looking for SOME_APP_FOO
will not yield results, as it's used just as FOO
internally. Confusing right? And for what, few letters? Remember that explicit is better?
IMHO the best way is to just set those defaults in Django's own settings, and why don't use piping that is already there? No module import hooks or hijacking models.py
being always imported to initialize some extra and complicated meta class piping.
Why not use AppConfig.ready for setting defaults?
class FooBarConfig(AppConfig):
name = 'foo_bar'
def ready(self):
from django.conf import settings
settings = settings._wrapped.__dict__
settings.setdefault('FOO_BAR_SETTING', 'whatever')
Or better yet define them in clean simple way in a separate module and import them as (or close to how) Settings class does it:
class FooBarConfig(AppConfig):
name = 'foo_bar'
def ready(self):
from . import app_settings as defaults
from django.conf import settings
for name in dir(defaults):
if name.isupper() and not hasattr(settings, name):
setattr(settings, name, getattr(defaults, name))
I'm not sure use of __dict__
is the best solution, but you get the idea, you can always user hasattr
/setattr
combo to get the efect.
This way your app settings are:
django.conf.settings
— you can implement some transposition of names if you want toPS. There is a warning about not modifying settings in run time but it does not explain why. So I think this one time, during initialization may be a reasonable exception ;)
PS2. Don't name the separate module just settings
as this may get confusing when you import settings
from django.conf
.
I think it's quite common to create a settings.py
in your app's package, where you define your settings like this:
from django.conf import settings
FOO = getattr(settings, 'FOO', "default_value")
In your app you can import them from your app's settings
module:
from myapp.settings import *
def print_foo():
print FOO
But I think everybody agrees that Django is lacking a better generic architecture for this! If you're looking for a more sophisticated way to handle this, there are some third party apps for this like django-appconf, but it's your decision if you want to introduce one more dependency for your app or not!
In settings.py
, put settings.*
before the property.
from django.conf import settings
settings.FOO = getattr(settings, 'FOO', "default_value")
In response to Phil Gyford's comment, exposing the problem of settings not overwritten in tests (since already imported in modules), what I did was to define an AppSettings
class in __init__.py
with:
__init__
method to initialize each setting to None
load
method to load every settings from gettersThen in the code:
from . import AppSettings
def in_some_function():
some_setting = AppSettings.get_some_setting()
Or if you want to load them all in once (but overriding settings in tests won't work for the impacted module):
from . import AppSettings
app_settings = AppSettings()
app_settings.load()
def in_some_function():
print(app_settings.some_setting)
You can then use the override_settings
decorator in your tests, and still have some DRY and clear way of using app settings, at the cost of more instructions executed each time you want to get a setting (just for tests...).