How to use decouple with a regular python library?

[fccoelho]

I am using decouple with a library I am developing.
It works just fine when I am testing it from the development directory. However after I install the library and try to import it from a python process started on a arbitrary directory, it cannot find the settings.ini even if I copy it to the directory I am starting python from.

In such cases, where should the settings.ini be placed? this should be made clear in the documentation.

[henriquebastos]

When you from decouple import config you're using the default AutoConfig instance.

It has a magic that searches for the settings.ini file from the caller's module directory. If it can't find there, it keeps searching up the tree.

If you want to have full controle and point directly where your settings.ini is:

from decouple import RepositoryIni, Config
config = Config(RepositoryIni('path/to/settings.ini'))

DEBUG = config('DEBUG')

This API has room for improvement. You're the 2nd to request something like this.

[nielsonsantana]

I have interest in this feature too. I solve it by override _caller_path

from decouple import AutoConfig

def _caller_path(self):
    return os.getcwd()

AutoConfig._caller_path = _caller_path
config = AutoConfig()

[henriquebastos]

@nielsonsantana can you provide a patch so we can invert the dependency on the path?

[nielsonsantana]

I trying figure out how change your code. Are you suggesting to override "_caller_path"?
Replace:

def _caller_path(self):
    # MAGIC! Get the caller's module path.
    frame = sys._getframe()
    path = os.path.dirname(frame.f_back.f_back.f_code.co_filename)
    return path 

for

def _caller_path(self):
    return os.getcwd()

In my tests, it's work based on where it is called from. Per example, if it is called from a terminal, it will use the local .env based on the path of the terminal.

[osantana]

I'd like to make a design suggestion: avoid configuration handling in libraries. The "right place" to handle configuration is your project (that requires/uses the library).

It's better make your library configuration-agnostic, so, the users of your library can use any configuration system. And you can also reduce the number of depedencies.

"Wrong" way

# mylib.py

class Client:
    def __init__(self, url=None):
        self.url = config("URL", default="http://localhost/")

# myproject.py

from mylib import Client

client = Client()

"Right" way

# mylib.py

class Client:
    def __init__(self, url):
        self.url = url

# myproject.py

from mylib import Client

url = config("URL", "http://localhost")
client = Client(url)

[rmax]

I faced this issue when trying to use decouple.config in a jupyter notebook.

This PR #28 allows the users to instantiate AutoConfig with a custom initial search path. For example, in a jupyter notebook:

config = AutoConfig(os.getcwd())

[henriquebastos]

Fixed tks to @rmax