Release Notes¶
Dynaconf 3.1.1¶
This release is more about bug fixes and improvements, we included a full diagram of the library to help contributors.
Vendoring¶
Dynaconf doesn't have dependencies because the main libraries used have been vendored. In addition to the vendoring process (which followed the guidelines regarding licensing and structure), all the vendored libraries have been minified to save space on the distribution.
Ideally the vendored libraries will be upgraded on-demand only when a new security or important bug fix is released on the respective vendored repository.
Python-box is an exception since the goal is to remove the dependency and implement its basic features within dynaconf/DynaBox type.
Cleanups¶
We made flake8 more strict and removed some unused imports and variables.
Variable interpolation¶
The mechanism to evaluate Lazy
values has been refactored and now @format
and
@jinja
values can be used within dictionaries and lists in any nesting levels.
Validators¶
Fixed a problem when multiple combined validators are registered and fixed duplication of ValidationError messages.
All names in a must_exist clause are now tested, instead of stopping validation on the first failure.
Fixes¶
- Cli now accepts dotted keys ex:
dynaconf -i config.settings list -k foo.bar.zaz
- boto is now optional for vault authentication
- a
.formatter
attribute lookup was conflicting with Django logging so argument has been renamed. - box_settings made optional to DynaBox
- Included an example of SOPS custom loader
settings.set
is fixed so it now can set dotted attrs directlysettings.FOO = 1
- Fix a problem on object_merge to allow keys with same name
- Added username and password to Redis
- Lots of typos on docs
Dynaconf 3.1.0¶
Providing default or computed values¶
Validators can be used to provide default or computed values.
Default values¶
Validator("FOO", default="A default value for foo")
Then if not able to load the values from files or environment this default value will be set for that key.
Computed values¶
Sometimes you need some values to be computed by calling functions, just pass a callable to the default
argument.
Validator("FOO", default=my_function)
then
def my_function(settings, validator):
return "this is computed during validation time"
If you want it to be lazy evaluated
from dynaconf.utils.parse_conf import empty, Lazy
Validator("FOO", default=Lazy(empty, formatter=my_function))
Dynaconf 3.0.0¶
In Dynaconf 3.0.0 we introduced some improvements and with those improvements it comes some breaking changes.
Some of the changes were discussed on the 1st Dynaconf community meeting video is available and meeting notes #354.
Improvements¶
- Validators now implements
|
and&
operators to allowValidator() &| Validator()
and has moreoperations
available such aslen_eq, len_min, len_max, startswith
#353. - First level variables are now allowed to be
lowercase
it is now possible to accesssettings.foo
orsettings.FOO
#357. - All Dependencies are now vendored, so when installing Dynaconf is not needed to install any dependency.
- Dynaconf configuration options are now aliased so when creating an instance of
LazySettings|FlaskDynaconf|DjangoDynaconf
it is now possible to pass instead ofENVVAR_PREFIX_FOR_DYNACONF
justenvvar_prefix
and this lowercase alias is now accepted. - Fixed bugs in
merge
and deprecated the@reset
token. - Added implementation for
__dir__
to allow auto complete for terminal and IDEs. - Add option to override mount point for vault server.
LazySettings
is now aliased toDynaconf
Dynaconf
class now accepts a parametervalidators=[Validator, ...]
that will be immediately evaluated when passed.
Breaking Changes¶
NOTE 1
from dynaconf import settings
will keep working until version4.0.x
to allow users to migrate gradually without having breaking changes but it will raise some DeprecationWarnings.NOTE 2 If you are using
FLASK
orDJANGO
plugins there will be no breaking change on the framework settings, a.k.a your project will keep working fine!.
dynaconf.settings
global settings instance is now deprecated.¶
Users are now supposed to create their own instance of settings
instead of using the deprecated from dynaconf import settings
.
project/config.py
from dynaconf import Dynaconf
settings = Dynaconf(**options)
and then in your program you do from project.config import settings
instead of from dynaconf import settings
.
The **options
are any of the dynaconf config options
ex:
settings = Dynaconf(
settings_file=["settings.toml", ".secrets.toml"],
envvar_prefix="MYPROJECT"
)
With the above instance your
settings
will load data from those 2.toml
files and also environment variables prefixed withMYPROOJECT
e.g:export MYPROJECT_FOO=1
Dynaconf is now envless by default.¶
Historically Dynaconf worked in a multi layered environments for loading data from files, so you were supposed to have a file like:
[default]
key = 'value'
[production]
key = 'value'
Now starting on 3.0.0 the environments are disabled by default, so the same file can be created as.
key = 'value'
And you can still have the environments but only if you explicit specify it when creating your instance.
To have the previous behavior supporting all environments from the first level of a file.
settings = Dynaconf(
environments=True
)
or to strictly specify your environments.
settings = Dynaconf(
environments=["default", "prodution", "testing", "xyz"],
)
Once it is defined all the other functionalities still works such as.
export ENV_FOR_DYNACONF=production myprogram.py
Automatic load of settings.*
files are disabled by default.¶
Starting on 3.x Dynaconf will now only load files passed explicitly to settings_file option when creating the instance or specified as env var SETTINGS_FILE_FOR_DYNACONF.
When creating the settings instance the file paths or globs should be specified unless you want only env vars and external loaders to be loaded.
Single file
settings = Dynaconf(
settings_file="settings.toml",
)
Or multiple files.
settings = Dynaconf(
settings_file=["settings.toml", "path/*.yaml"],
)
Dynaconf will respect the exact order informed by you on the settings_file
parameter.
NOTE the parameters
preload
,includes
andsecret
continues to work in the same way as before. NOTE the parameterspreload
,includes
andsecret
continues to work in the same way as before.
Load of dotenv .env
is now disabled by default.¶
Dynaconf will NO MORE automatically load the data from .env
file and will do that only
if load_dotenv=True
is provided.
settings = Dynaconf(
load_dotenv=True
)
BY default it keeps searching for files in the current PROJECT_ROOT
named .env
and the dotenv_path
accepts a relative path such as .env
or path/to/.env
DEBUG Logging is now completely removed¶
There is no more logging
or debug
messages being printed, so the variable DEBUG_LEVEL
has no more effect.
Coming in 3.1.x¶
- Support for Pydantic BaseSettings for Validators.
- Support for replacement of
toml
parser on envvars loader.