PluginBase

Ever tried creating a plugin system for a Python application and you discovered fighting against the import system? Me too. This is where PluginBase comes in.

PluginBase is a module for Python which extends the import system for the most common form of plugin usage which is providing a consistent import experience for plugins from a variety of sources. Essentially it allows you to build very flexible plugin based applications which pull in plugins from bundled sources as well as application specific ones without bypassing the Python import system.

How does it work? It’s super simple:

Step 1:

Create a “plugin base” object. It defines a pseudo package under which all your plugins will reside. For instance it could be yourapplication.plugins:

from pluginbase import PluginBase
plugin_base = PluginBase(package='yourapplication.plugins')

Step 2:

Now that you have a plugin base, you can define a plugin source which is the list of all sources which provide plugins:

plugin_source = plugin_base.make_plugin_source(
    searchpath=['./path/to/plugins', './path/to/more/plugins'])

Step 3:

To import a plugin all you need to do is to use the regular import system. The only change is that you need to import the plugin source through the with statement:

with plugin_source:
    from yourapplication.plugins import my_plugin
my_plugin.do_something_cool()

Alternatively you can also import plugins programmatically instead of using the import statement:

my_plugin = plugin_source.load_plugin('my_plugin')

For a more complex example see the one from the git repo: pluginbase-example.

Installation

You can get the library directly from PyPI:

pip install pluginbase

FAQ

Q: Why is there a plugin base and a plugin source class?

This decision was taken so that multiple applications can co-exist together. For instance imagine you have an application that implements a wiki but you want multiple instances of that wiki to exist in the same Python interpreter. The plugin sources split out the load paths of the different applications.

Each instance of the wiki would have its own plugin source and they can work independently of each other.

Q: Do plugins pollute sys.modules?

While a plugin source is alive the plugins do indeed reside in sys.modules. This decision was make conciously so that as little as possible of the Python library ecosystem breaks. However when the plugin source gets garbage collected all loaded plugins will also get garbage collected.

Q: How does PluginBase deal with different versions of the same plugin?

Each plugin source works indepdenently of each other. The way this works is by internally translating the module name. By default that module name is a random number but it can also be forced to a hash of a specific value to make it stable across restarts which allows pickle and similar libraries to work.

This internal module renaming means that yourapplication.module.foo will internally be called pluginbase._internalspace._sp7...be4 for instance. The same plugin loaded from another plugin source will have a different internal name.

Q: What happens if a plugin wants to import other modules?

All fine. Plugins can import from itself as well as other plugins that can be located.

Q: Does PluginBase support pickle?

Yes, pickle works fine for plugins but it does require defining a stable identifier when creating a plugin source. This could for instance be a file system path:

plugin_source = base.make_plugin_source(
    searchpath=[app.plugin_path],
    identifier=app.config_filename)

Q: What happens if I import from the plugin module without the plugin source activated through the with statement?

The import will fail with a descriptive error message explaining that a plugin source needs to be activated.

Q: Can I automatically discover all modules that are available?

Yes you can. Just use the PluginSource.list_plugins() method which returns a list of all plugins that a source can import.

Q: Why would I use this over setuptools based plugins?

PluginBase and setuptools based plugins solve very different problems and are incompatible on an architectural point of view. PluginBase does not solve plugin distribution through PyPI but allows plugins to be virtualized from each other. Setuptools on the other hand is based on PyPI based distribution but piggybacks on top of the regular import system.

There are advantages and disadvantages to both of them. Setuptools based plugins are very useful to extend libraries from other libraries. For instance the Jinja2 template engine hooks into the Babel library for internationalization through setuptools.

On the other hand applications distributed to users can benefit from a PluginBase based system which allows them to take control over how plugins are distributed and full separation from each other.

API

High Level

class pluginbase.PluginBase(package, searchpath=None)

The plugin base acts as a control object around a dummy Python package that acts as a container for plugins. Usually each application creates exactly one base object for all plugins.

Parameters:
  • package – the name of the package that acts as the plugin base. Usually this module does not exist. Unless you know what you are doing you should not create this module on the file system.
  • searchpath – optionally a shared search path for modules that will be used by all plugin sources registered.
make_plugin_source(*args, **kwargs)

Creats a plugin source for this plugin base and returns it. All parameters are forwarded to PluginSource.

package = None

the name of the dummy package.

searchpath = None

the default search path shared by all plugins as list.

class pluginbase.PluginSource(base, identifier=None, searchpath=None, persist=False)

The plugin source is what ultimately decides where plugins are loaded from. Plugin bases can have multiple plugin sources which act as isolation layer. While this is not a security system it generally is not possible for plugins from different sources to accidentally cross talk.

Once a plugin source has been created it can be used in a with statement to change the behavior of the import statement in the block to define which source to load the plugins from:

plugin_source = plugin_base.make_plugin_source(
    searchpath=['./path/to/plugins', './path/to/more/plugins'])

with plugin_source:
    from myapplication.plugins import my_plugin
Parameters:
  • base – the base this plugin source belongs to.
  • identifier – optionally a stable identifier. If it’s not defined a random identifier is picked. It’s useful to set this to a stable value to have consistent tracebacks between restarts and to support pickle.
  • searchpath – a list of paths where plugins are looked for.
  • persist – optionally this can be set to True and the plugins will not be cleaned up when the plugin source gets garbage collected.
base = None

A reference to the plugin base that created this source.

cleanup()

Cleans up all loaded plugins manually. This is necessary to call only if persist is enabled. Otherwise this happens automatically when the source gets garbage collected.

identifier = None

the identifier for this source.

list_plugins()

Returns a sorted list of all plugins that are available in this plugin source. This can be useful to automatically discover plugins that are available and is usually used together with load_plugin().

load_plugin(name)

This automatically loads a plugin by the given name from the current source and returns the module. This is a convenient alternative to the import statement and saves you from invoking __import__ or a similar function yourself.

Parameters:name – the name of the plugin to load.
mod = None

a reference to the module on the internal pluginsource._internalspace.

open_resource(plugin, filename)

This function locates a resource inside the plugin and returns a byte stream to the contents of it. If the resource cannot be loaded an IOError will be raised. Only plugins that are real Python packages can contain resources. Plain old Python modules do not allow this for obvious reasons.

New in version 0.3.

Parameters:
  • plugin – the name of the plugin to open the resource of.
  • filename – the name of the file within the plugin to open.
persist = False

indicates if this plugin source persists or not.

searchpath = None

a list of paths where plugins are searched in.

spaceid = None

The internal module name of the plugin source as it appears in the pluginsource._internalspace.

pluginbase.get_plugin_source(module=None, stacklevel=None)

Returns the PluginSource for the current module or the given module. The module can be provided by name (in which case an import will be attempted) or as a module object.

If no plugin source can be discovered, the return value from this method is None.

This function can be very useful if additional data has been attached to the plugin source. For instance this could allow plugins to get access to a back reference to the application that created them.

Parameters:
  • module – optionally the module to locate the plugin source of.
  • stacklevel – defines how many levels up the module should search for before it discovers the plugin frame. The default is 0. This can be useful for writing wrappers around this function.

Import Hook Control

pluginbase.import_hook.enable(self)

Enables the import hook which drives the plugin base system. This is the default.

pluginbase.import_hook.disable(self)

Disables the import hook and restores the default import system behavior. This effectively breaks pluginbase but can be useful for testing purposes.

pluginbase.import_hook.enabled

Indicates if the import hook is currently active or not.

Internals

pluginbase._internalspace = <module 'pluginbase._internalspace' (built-in)>

This module is where pluginbase keeps track of all loaded plugins. Generally one can completely ignore the existence of it, but in some situations it might be useful to discover currently loaded modules through this when debugging.

Fork me on GitHub