Zirconium

Zirconium is a powerful configuration tool for loading and using configuration in your application.

Use Case

Zirconium abstracts away the process of loading and type-coercing configuration so that it Just Works for your application. For example

Key Features

Features

• Support for libraries to provide their own default configuration and/or configuration file locations
• Applications specify their own configuration with @zirconium.configure decorator
• Automatic replacement of ${ENVIRONMENT_VARIABLES} in strings
• Consistent type coercion for common data types: paths, ints, floats, decimals, bytes, lists, dicts, sets, dates, timedeltas, and datetimes
• Where dictionary-style declarations are not supported, instead use the dot syntax (e.g. "foo.bar")
• Supports multiple file encodings
• Extensible to other formats as needed
• Configuration is dict-like for ease-of-use in existing locations (e.g. Flask)
• Multiple files can be specified with different weights to control loading order
• Supports default vs. normal configuration file (defaults always loaded first)
• Supports thread-safe injection of the configuration into your application via autoinject
• Supports specifying default configuration for libraries in entry points zirconium.config and for parsers in zirconium.parsers, as well as using the @zirconium.configure decorator.

Supported configuration methods

• Database tables (with SQLAlchemy installed)
• YAML (with pyyaml installed)
• TOML (with toml installed or Python >= 3.11)
• JSON
• Setuptools-like CFG files
• INI files (following the defaults of the configparser module)
• Environment variables

Priority Order

Later items in this list will override previous items

1. Files registered with register_default_file(), in ascending order by weight (or order called)
2. Files registered with register_file(), in ascending order by weight
3. Files from environment variables registered with register_file_from_environ(), in ascending order by weight
4. Values from environment variables registered with register_environ_var()

Example Usage

import pathlib
import zirconium
from autoinject import injector


@zirconium.configure
def add_config(config):
  
    # Direct load configuration from dict:
    config.load_from_dict({
        "version": "0.0.1",
        "database": {
            # Load these from environment variables
            "username": "${MYAPP_DATABASE_USERNAME}",
            "password": "${MYAPP_DATABASE_PASSWORD}",
        },
        "escaped_environment_example": "$${NOT_AN_ENVIRONMENT VARIABLE",
        "preceding_dollar_sign": "$$${STOCK_PRICE_ENV_VARIABLE}",
    })
    
    # Default configuration, relative to this file, will override the above dict
    base_file = pathlib.Path(__file__).parent / ".myapp.defaults.toml"
    config.register_default_file(base_file) 
    
    # File in user home directory, overrides the defaults
    config.register_file("~/.myapp.toml")
    
    # File in CWD, will override whatever is in home
    config.register_file("./.myapp.toml")
    
    # Load a file path from environment variable, will override ALL registered files
    config.register_file_from_environ("MYAPP_CONFIG_FILE")
    
    # Load values direct from the environment, will override ALL files including those specific in environment variables
    # sets config["database"]["password"]
    config.register_environ_var("MYAPP_DATABASE_PASSWORD", "database", "password")
    # sets config["database"]["username"]
    config.register_environ_var("MYAPP_DATABASE_USERNAME", "database", "username")
    
  
# Injection example
class NeedsConfiguration:

    config: zirconium.ApplicationConfig = None

    @injector.construct
    def __init__(self):
        # you have self.config available as of here
        pass
    
    
# Method example

@injector.inject 
def with_config(config: zirconium.ApplicationConfig = None):
    print(f"Hello world, my name is {config.as_str('myapp', 'welcome_name')}")
    print(f"Database user: {config.as_str('database', 'username')}")

Type Coercion Examples

import zirconium
from autoinject import injector, auto

@zirconium.configure 
def add_config(config):
    config.load_from_dict({
        "bytes_example": "5K",
        "timedelta_example": "5m",
        "date_example": "2023-05-05",
        "datetime_example": "2023-05-05T17:05:05",
        "int_example": "5",
        "float_example": "5.55",
        "decimal_example": "5.55",
        "str_example": "5.55",
        "bool_false_example": 0,
        "bool_true_example": 1,
        "path_example": "~/user/file",
        "set_example": ["one", "one", "two"],
        "list_example": ["one", "one", "two"],
        "dict_example": {
          "one": 1,
          "two": 2,
        }
    })
    

@injector.inject 
def show_examples(config: zirconium.ApplicationConfig = auto()):
    config.as_bytes("bytes_example")                # 5120 (int)
    config.as_timedelta("timedelta_example")         # datetime.timedelta(minutes=5)
    config.as_date("date_example")                  # datetime.date(2023, 5, 5)
    config.as_datetime("datetime_example")          # datetime.datetime(2023, 5, 5, 17, 5, 5)
    config.as_int("int_example")                    # 5 (int)
    config.as_float("float_example")                # 5.55 (float)
    config.as_decimal("decimal_example")            # decimal.Decimal("5.55")
    config.as_str("str_example")                    # "5.55"
    config.as_bool("bool_false_example")            # False (bool)
    config.as_bool("bool_true_example")             # True (bool)
    config.as_path("path_example")                  # pathlib.Path("~/user/file")
    config.as_set("set_example")                    # {"one", "two"}
    config.as_list("list_example")                  # ["one", "one", "two"]
    config.as_dict("dict_example")                  # {"one": 1, "two": 2}
    
    # Raw dicts can still be used as sub-keys, for example
    config.as_int(("dict_example", "one"))          # 1 (int)  
 

Config References

In certain cases, your application might want to let the configuration be reloaded. This is possible via the reload_config() method which will reset your configuration to its base and reload all the values from the original files. However, where a value has already been used in your program, that value will need to be updated. This leads us to the ConfigRef() pattern which lets applications obtain a value and keep it current with the latest value loaded. If you do not plan on reloading your configuration on-the-fly, you can skip this section.

When using the methods that end in _ref(), you will obtain an instance of _ConfigRef(). This object has a few special properties but will mostly behave as the underlying configuration value with a few exceptions:

• isinstance will not work with it
• is None will not return True even if the configuration value is actually None (use .is_none() instead)

To get a raw value to work with, use raw_value().

The value is cached within the _ConfigRef() object but this cache is invalidated whenever reload_config() is called. This should reduce the work you have to do when reloading your configuration (though you may still need to call certain methods when the configuration is reloaded).

To call a method on reload, you can add it via config.on_load(callable). If callable needs to interact with a different thread or process than the one where reload_config() is called, it is your responsibility to manage this communication (e.g. use threading.Event to notify the thread that the configuration needs to be reloaded).

Testing classes that use ApplicationConfig

Unit test functions decorated with autoinject.injector.test_case can declare configuration using zirconium.test_with_config(key, val) to declare configuration for testing. For example, this test case should pass:

from autoinject import injector
import zirconium as zr 
import unittest 

class MyTestCase(unittest.TestCase):

  # pass configuration
  @zr.test_with_config({
    "foo": {
      "bar2": "hello world #2"
    }
  })
  def test_something(self):
    
    # As a simple example.
    @injector.inject 
    def do_something(cfg: zr.ApplicationConfig = None):
        self.assertEqual(cfg.as_str(("foo", "bar")), "hello world")
        self.assertEqual(cfg.as_str(("some", "value")), "what")

Note that this pattern replaces all configuration values with the ones declared in decorators, so previously loaded values will not be passed into your test function nor will they be passed between test functions.

Change Log

Version 1.3.0

• Test cases now work with autoinject 2.0.0
• The various as_X() methods should now provide proper type-hinting when a default value is passed.
• Various clean-up items

Version 1.2.1

• Test cases can now use the fixture @zirconium.test_with_config(key: t.Iterable, value: t.Any) to inject test configuration.

Version 1.2.0

• Added as_bytes() which will accept values like 2M and return the value converted into bytes (e.g. 2097152. If you really want to use metric prefixes (e.g. 2MB=2000000), you must pass allow_metric=True and then specify your units as 2MB. Prefixes up to exbibyte (EiB) are handled at the moment. You can also specify B for bytes or bit for a number of bits. If no unit is specified, it uses the default_units parameter, which is B by default. All units are case-insensitive.
• Added as_timedelta() which will accept values like 30m and return datetime.timedelta(minutes=30). Valid units are s, m, h, d, w, us, and ms. If no units are specified, it defaults to the default_units parameter which is s by default. All units are case-insensitive.
• Added a new series of methods as_*_ref() (and get_ref()) which mirror the behaviour of their counterparts not ending in _ref() except these return a _ConfigRef() instance instead of an actual value.
• Added a method print_config() which will print out the configuration to the command line.

Version 1.1.0

• Added as_list() and as_set() which return as expected
• Type-hinting added to the as_X() methods to help with usage in your IDE
• Added support for register_files() which takes a set of directories to use and registers a set of files and default files in each.

Version 1.0.0

• Stable release after extensive testing on my own
• Python 3.11's tomllib now supported for parsing TOML files
• Using pymitter to manage configuration registration was proving problematic when called from a different thread than where the application config object was instatiated. Replaced it with a more robust solution.
• Fixed a bug for registering default files
• Added as_dict() to the configuration object which returns an instance of MutableDeepDict.