© 2014-2023 Alice Bevan-McGregor and contributors.
https://github.com/marrow/package
This package is a combination of utilities for handling object lookup, resolving object names, and managing simple to complex plugin architectures. Notably it includes a dependency graph system for extensions and helper for looking up qualified object names.
This library is fully unit tested where possible.
Installing marrow.package
is easy, just execute the following in a terminal:
pip install marrow.package
Note: We strongly recommend always using a container, virtualization, or sandboxing environment of some kind when developing using Python; installing things system-wide is yucky (for a variety of reasons) nine times out of ten. We prefer light-weight virtualenv, others prefer solutions as robust as Vagrant.
If you add marrow.package
to the install_requires
argument of the call to setup()
in your application's
setup.py
file, Marrow Package will be automatically installed and made available when your own application or
library is installed. We recommend using "less than" version numbers to ensure there are no unintentional
side-effects when updating. Use marrow.package<2.2
to get all bugfixes for the current release, and
marrow.package<3.0
to get bugfixes and feature updates while ensuring that large breaking changes are not installed.
Development takes place on GitHub in the marrow.package project. Issue tracking, documentation, and downloads are provided there.
Installing the current development version requires Git, a distributed source code management system. If you have Git you can run the following to download and link the development version into your Python runtime:
git clone https://github.com/marrow/package.git (cd package; python setup.py develop)
You can then upgrade to the latest version at any time:
(cd package; git pull; python setup.py develop)
If you would like to make changes and contribute them back to the project, fork the GitHub project, make your changes, and submit a pull request. This process is beyond the scope of this documentation; for more information see GitHub's documentation.
Object references describe the module and attribute path needed to resolve the object. For example, foo:bar
is a
reference that describes importing "foo" prior to retrieving an object named "bar" from the module. On Python 3.3 a
useful shortcut is provided, __qualname__
which speeds up this lookup.
For example, let's define a class and get a reference to it:
from marrow.package.canonical import name class Example(object): pass assert name(Example) == '__main__:Example'
You can, depending on platform, retrieve a reference to any of the following types of objects:
- Module level:
- class
- class instance
- class method
- class staticmethod
- function
- instance classmethod
- instance method
- instance staticmethod
- nested classes and methods
- closures
The load
utility can optionally be provided a plugin namespace to search. If the target object is found within the
namespace, the name of the plugin entry will be returned. By default, if a named plugin can not be found, a
LookupError
will be raised. If a direct reference is acceptable, the boolean direct
argument (third positional)
can be made truthy to permit direct references.
from marrow.package.canonical import name
assert name(name, 'marrow.package.sample') == 'name'
Two utilities are provided which allow you resolve string path references to objects. The first is quite simple:
from marrow.package.loader import traverse assert traverse({'foo': {'bar': 27}}, 'foo.bar') == 27
This will search the dictionary described for a "foo" element, then "bar" element.
The traverse
function takes some additional optional arguments. If executable
is True
any executable
function encountered will be executed without arguments. Traversal will continue on the result of that call. You can
change the separator as desired, i.e. to a '/' using the separator
argument.
By default attributes (but not array elements) prefixed with an underscore are taboo. They will not resolve, raising
a LookupError. You can allow these by setting protect
to False
.
Certain allowances are made: if a 'path segment' is numerical, it's treated as an array index. If attribute lookup fails, it will re-try on that object using array notation and continue from there. This makes lookup very flexible.
The more complete API for name resolution uses the load
function, which takes the same optional keyword arguments
as traverse
. Additionally, this function accepts an optional namespace
to search for plugins within. For
example:
from marrow.package.loader import load from pip import main # Load class Foo from example.objects load('example.objects:Foo') # Load the result of the class method ``new`` of the Foo object load('example.objects:Foo.new', executable=True) # Load the "pip" command-line interface. assert load('pip', 'console_scripts') is main
Providing a namespace does not prevent explicit object lookup (dot-colon notation) from working.
An attribute-access dictionary is provided that acts as an import cache:
from marrow.package.cache import PackageCache from pip import main cache = PackageCache('console_scripts') assert cache.pip is main assert cache['pip'] is main assert len(cache) == 1 assert 'pip' in cache
You can lazily load and cache an object reference upon dereferencing from an instance using the lazyload
utility
from the marrow.package.lazy
module. Assign the result of calling this function with either an object reference
passed in positionally:
class MyClass: debug = lazyload('logging:debug')
Or the attribute path to traverse (using marrow.package.loader:traverse
) prefixed by a period:
class AnotherClass: target = 'logging:info' log = lazyload('.target')
Any additional arguments are passed to the eventual call to load(). This utility builds on a simpler one that is
also offered for fully-tested re-use, lazy
, a decorator like @property
which will cache the result, with
thread-safe locking to ensure only one call will ever be made to the decorated function, per instance.
This package provides two main methods of dealing with plugins and extensions, the first is simple, the second provides full dependency graphing of the extensions.
The PluginManager
class takes two arguments: the first is the entry point namespace
to search, the second is
an optional iterable of folders to add to the Python search path for installed packages, allowing your application to
have a dedicated plugins folder.
It provides a register
method which take a name and the object to use as the plugin and registers it internally,
supporting both attribute and array-like notation for retrieval, as well as iteration of plugins (includes all entry
point plugins found and any custom registered ones).
At a higher level is a PluginManager
subclass called ExtensionManager
which additionally exposes a sort
method capable of resolving dependency order for extensions which follow a simple protocol: have an attribute or array
element matching the following, all optional:
provides
— declare tags describing the features offered by the pluginneeds
— declare the tags that must be present for this extension to functionuses
— declare the tags that must be evaluated prior to this extension, but aren't hard requirementsfirst
— declare that this extension is a dependency of all other non-first extensionslast
— declare that this extension depends on all other non-last extensionsexcludes
— declare tags that must not be present in other plugins for this one to be usable
- Initial release. Combination of utilities from other Marrow projects.
- Extended decorator support. New code paths and tests added to cover canonicalization of decorated functions.
- Diagnostic information. Removed extraneous diagnostic information.
- Added lazy evaluation. There are two new helpers for caching of
@property
-style attributes and lazy lookup of object references.
- Deprecated Python 2.6 and 3.3. While no particular backwards incompatible change was made; as setuptools no longer supports these versions, these versions are now hard/impossible to test.
- Allow extensions to declare exclusions. Flags that must not be defined for the extension to be usable.
- Updated minimum Python version. Marrow Package now requires Python 3.5 or later.
- Removed Python 2 support and version specific code. The project has been updated to modern Python packaging standards, including modern namespace use. Modern namespaces are wholly incompatible with the previous namespacing mechanism; this project can not be simultaneously installed with any Marrow project that is Python 2 compatible.
- Extensive type annotation and in-development validation. When run without optimizations (-O argument to Python or PYTHONOPTIMIZE environment variable) type annotations will be validated.
- Reduced test fragility. Previously the tests utilized the console_scripts namespace, this was fragile to the presence of other installed libraries, e.g. numpy broke the tests on Travis.
- Migrated from Travis-CI to GitHub Actions for test automation.
- Implement package-relative path lookup. The load utility function can now resolve the path to a file relative to a package. This is particularly useful for looking up the path to template files or on-disk static assets.
- Protected attribute access now fails. Underscore-prefixed attributes are assumed to be "protected", with the technical note that Python adds new internal "double underscore" attributes which must not spontaneously exist, or generate errors other than AttributeError.
- Tests are now independent of third-party plugin registration.
- Update type hinting validation. The
typeguard
package has removed a functional utility; decoration now used. - Canonical plugin name resolution. The
name()
utility can now resolve the plugin name if given a plugin namespace to check.
Marrow Package has been released under the MIT Open Source license.
Copyright © 2014-2023 Alice Bevan-McGregor and contributors.
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.