IntenseWebs/freeipa
3
0
Fork 0
mirror of https://salsa.debian.org/freeipa-team/freeipa.git synced 2025-02-25 18:55:28 -06:00
Code Issues Packages Projects Releases Wiki Activity

Finished fist draft of plugin tutorial in ipalib/__init__.py docstring

Browse Source
This commit is contained in:
Jason Gerard DeRose
2008-11-07 02:26:38 -07:00
parent 5bdf860647
commit c26a3c8542
5 changed files with 650 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

646
ipalib/__init__.py
View File

@@ -17,17 +17,622 @@
# along with this program; if not, write to the Free Software
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
"""
'''
Package containing core library.
To learn about the ``ipalib`` library, you should read the code in this order:
=============================
Tutorial for Plugin Authors
=============================
1. Get the big picture from some actual plugins, like `plugins.f_user`.
This tutorial gives a broad learn-by-doing introduction to writing plugins
for freeIPA v2. As not to overwhelm the reader, it does not cover every
detail, but it does provides enough to get one started and is heavily
cross-referenced with further documentation that (hopefully) fills in the
missing details.
2. Learn about the base classes for frontend plugins in `frontend`.
Where the documentation has left the reader confused, the many built-in
plugins in `ipalib.plugins` and `ipa_server.plugins` provide real-life
examples of how to write good plugins.
3. Learn about the core plugin framework in `plugable`.
"""
*Note:*
This tutorial, along with all the Python docstrings in freeIPA v2,
uses the *reStructuredText* markup language. For documentation on
reStructuredText, see:
http://docutils.sourceforge.net/rst.html
For documentation on using reStructuredText markup with epydoc, see:
http://epydoc.sourceforge.net/manual-othermarkup.html
------------------------------------
First steps: A simple command plugin
------------------------------------
Our first example will create the most basic command plugin possible. A
command plugin simultaneously adds a new command that can be called through
the command-line ``ipa`` script *and* adds a new XML-RPC method... the two are
one in the same, simply invoked in different ways.
All plugins must subclass from `plugable.Plugin`, and furthermore, must
subclass from one of the base classes allowed by the `plugable.API` instance
returned by the `get_standard_api()` function.
To be a command plugin, your plugin must subclass from `frontend.Command`.
Creating a basic plugin involves two steps, defining the class and then
registering the class:
>>> from ipalib import Command, get_standard_api
>>> api = get_standard_api()
>>> class my_command(Command): # Step 1, define class
... """My example plugin."""
...
>>> api.register(my_command) # Step 2, register class
Notice that we are registering the ``my_command`` class itself and not an
instance thereof.
Until `plugable.API.finalize()` is called, your plugin class has not been
instantiated nor the does the ``Command`` namespace yet exist. For example:
>>> hasattr(api, 'Command')
False
>>> api.finalize()
>>> hasattr(api.Command, 'my_command')
True
>>> api.Command.my_command.doc
'My example plugin.'
Notice that your plugin instance in accessed through an attribute named
'my_command', the same name as your plugin class name.
------------------------------
Make your command do something
------------------------------
This simplest way to make your example command plugin do something is to
implement a ``run()`` method, like this:
>>> class my_command(Command):
... """My example plugin with run()."""
...
... def run(self):
... return 'My run() method was called!'
...
>>> api = get_standard_api()
>>> api.register(my_command)
>>> api.finalize()
>>> api.Command.my_command() # Call your plugin
'My run() method was called!'
When `frontend.Command.__call__()` is called, it first validates any arguments
and options your command plugin takes (if any) and then calls its ``run()``
method.
------------------------
Forwarding vs. execution
------------------------
However, unlike the example above, a typical command plugin will implement an
``execute()`` method instead of a ``run()`` method. Your command plugin can
be loaded in two distinct contexts:
1. In a *client* context - Your command plugin is only used to validate
any arguments and options it takes, and then ``self.forward()`` is
called, which forwards the call over XML-RPC to an IPA server where
the actual work is done.
2. In a *server* context - Your same command plugin validates any
arguments and options it takes, and then ``self.execute()`` is called,
which you should implement to perform whatever work your plugin does.
The base `frontend.Command.run()` method simply dispatches the call to
``self.execute()`` if ``self.env.in_server`` is True, or otherwise
dispatches the call to ``self.forward()``.
For example, say you have a command plugin like this:
>>> class my_command(Command):
... """Forwarding vs. execution."""
...
... def forward(self):
... return 'in_server=%r; forward() was called.' % self.env.in_server
...
... def execute(self):
... return 'in_server=%r; execute() was called.' % self.env.in_server
...
If ``my_command`` is loaded in a *client* context, ``forward()`` will be
called:
>>> api = get_standard_api()
>>> api.env.in_server = False # run() will dispatch to forward()
>>> api.register(my_command)
>>> api.finalize()
>>> api.Command.my_command() # Call your command plugin
'in_server=False; forward() was called.'
On the other hand, if ``my_command`` is loaded in a *server* context,
``execute()`` will be called:
>>> api = get_standard_api()
>>> api.env.in_server = True # run() will dispatch to execute()
>>> api.register(my_command)
>>> api.finalize()
>>> api.Command.my_command() # Call your command plugin
'in_server=True; execute() was called.'
Normally there should be no reason to override `frontend.Command.forward()`,
but, as above, it can be done for demonstration purposes. In contrast, there
*is* a reason you might want to override `frontend.Command.run()`: if it only
makes sense to execute your command locally, if it should never be forwarded
to the server. In this case, you should implement your *do-stuff* in the
``run()`` method instead of in the ``execute()`` method.
For example, the ``ipa`` command line script has a ``help`` command
(`ipalib.cli.help`) that is specific to the command-line-interface and should
never be forwarded to the server.
---------------
Backend plugins
---------------
There are two types of plugins:
1. *Frontend plugins* - These are loaded in both the *client* and *server*
contexts. These need to be installed with any application built atop
the `ipalib` library. The built-in frontend plugins can be found in
`ipalib.plugins`. The ``my_command`` example above is a frontend
plugin.
2. *Backend plugins* - These are only loaded in a *server* context and
only need to be installed on the IPA server. The built-in backend
plugins can be found in `ipa_server.plugins`.
Backend plugins should provide a set of methods that standardize how IPA
interacts with some external system or library. For example, all interaction
with LDAP is done through the ``ldap`` backend plugin defined in
`ipa_server.plugins.b_ldap`. As a good rule of thumb, anytime you need to
import some package that is not part of the Python standard library, you
should probably interact with that package via a corresponding backend
plugin you implement.
Backend plugins are much more free-form than command plugins. Aside from a
few reserved attribute names, you can define arbitrary public methods on your
backend plugin (in contrast, frontend plugins get wrapped in a
`plugable.PluginProxy`, which allow access to only specific attributes on the
frontend plugin).
Here is a simple example:
>>> from ipalib import Backend
>>> class my_backend(Backend):
... """My example backend plugin."""
...
... def do_stuff(self):
... """Part of your API."""
... return 'Stuff got done.'
...
>>> api = get_standard_api()
>>> api.register(my_backend)
>>> api.finalize()
>>> api.Backend.my_backend.do_stuff()
'Stuff got done.'
-------------------------------
How your command should do work
-------------------------------
We now return to our ``my_command`` plugin example.
Plugins are separated into frontend and backend plugins so that there are not
unnecessary dependencies required by an application that only uses `ipalib` and
its built-in frontend plugins (and then forwards over XML-RPC for execution).
But how do we avoid introducing additional dependencies? For example, the
``user_add`` command needs to talk to LDAP to add the user, yet we want to
somehow load the ``user_add`` plugin on client machines without requiring the
``python-ldap`` package (Python bindings to openldap) to be installed. To
answer that, we consult our golden rule:
**The golden rule:** A command plugin should implement its ``execute()``
method strictly via calls to methods on one or more backend plugins.
So the module containing the ``user_add`` command does not itself import the
Python LDAP bindings, only the module containing the ``ldap`` backend plugin
does that, and the backend plugins are only installed on the server. The
``user_add.execute()`` method, which is only called when in a server context,
is implemented as a series of calls to methods on the ``ldap`` backend plugin.
When `plugable.Plugin.set_api()` is called, each plugin stores a reference to
the `plugable.API` instance it has been loaded into. So your plugin can
access the ``my_backend`` plugin as ``self.api.Backend.my_backend``.
Additionally, convenience attributes are set for each namespace, so your
plugin can also access the ``my_backend`` plugin as simply
``self.Backend.my_backend``.
This next example will tie everything together. First we create our backend
plugin:
>>> api = get_standard_api()
>>> api.env.in_server = True # We want to execute, not forward
>>> class my_backend(Backend):
... """My example backend plugin."""
...
... def do_stuff(self):
... """my_command.execute() calls this."""
... return 'my_backend.do_stuff() indeed did do stuff!'
...
>>> api.register(my_backend)
Second, we have our frontend plugin, the command:
>>> class my_command(Command):
... """My example command plugin."""
...
... def execute(self):
... """Implemented against Backend.my_backend"""
... return self.Backend.my_backend.do_stuff()
...
>>> api.register(my_command)
Lastly, we call ``api.finalize()`` and see what happens when we call
``my_command()``:
>>> api.finalize()
>>> api.Command.my_command()
'my_backend.do_stuff() indeed did do stuff!'
When not in a server context, ``my_command.execute()`` never gets called, so
it never tries to access the non-existent backend plugin at
``self.Backend.my_backend.`` To emphasize this point, here is one last
example:
>>> api = get_standard_api()
>>> api.env.in_server = False # We want to forward, not execute
>>> class my_command(Command):
... """My example command plugin."""
...
... def execute(self):
... """Same as above."""
... return self.Backend.my_backend.do_stuff()
...
... def forward(self):
... return 'Just my_command.forward() getting called here.'
...
>>> api.register(my_command)
>>> api.finalize()
Notice that the ``my_backend`` plugin has certainly not be registered:
>>> hasattr(api.Backend, 'my_backend')
False
And yet we can call ``my_command()``:
>>> api.Command.my_command()
'Just my_command.forward() getting called here.'
----------------------------------------
Calling other commands from your command
----------------------------------------
It can be useful to have your ``execute()`` method call other command plugins.
Among other things, this allows for meta-commands that conveniently call
several other commands in a single operation. For example:
>>> api = get_standard_api()
>>> api.env.in_server = True # We want to execute, not forward
>>> class meta_command(Command):
... """My meta-command plugin."""
...
... def execute(self):
... """Calls command_1(), command_2()"""
... return '%s; %s.' % (
... self.Command.command_1(),
... self.Command.command_2()
... )
>>> class command_1(Command):
... def execute(self):
... return 'command_1.execute() called'
...
>>> class command_2(Command):
... def execute(self):
... return 'command_2.execute() called'
...
>>> api.register(meta_command)
>>> api.register(command_1)
>>> api.register(command_2)
>>> api.finalize()
>>> api.Command.meta_command()
'command_1.execute() called; command_2.execute() called.'
Because this is quite useful, we are going to revise our golden rule somewhat:
**The revised golden rule:** A command plugin should implement its
``execute()`` method strictly via what it can access through ``self.api``,
most likely via the backend plugins in ``self.api.Backend`` (which can also
be conveniently accessed as ``self.Backend``).
-----------------------------------------------
Defining arguments and options for your command
-----------------------------------------------
You can define a command can accept arbitrary arguments and options.
For example:
>>> from ipalib import Param
>>> class nudge(Command):
... """Takes one argument, one option"""
...
... takes_args = ['programmer']
...
... takes_options = [Param('stuff', default=u'documentation')]
...
... def execute(self, programmer, **kw):
... return '%s, go write more %s!' % (programmer, kw['stuff'])
...
>>> api = get_standard_api()
>>> api.env.in_server = True
>>> api.register(nudge)
>>> api.finalize()
>>> api.Command.nudge('Jason')
u'Jason, go write more documentation!'
>>> api.Command.nudge('Jason', stuff='unit tests')
u'Jason, go write more unit tests!'
The ``args`` and ``options`` attributes are `plugable.NameSpace` instances
containing a command's arguments and options, respectively, as you can see:
>>> list(api.Command.nudge.args) # Iterates through argument names
['programmer']
>>> api.Command.nudge.args.programmer
Param('programmer', Unicode())
>>> list(api.Command.nudge.options) # Iterates through option names
['stuff']
>>> api.Command.nudge.options.stuff
Param('stuff', Unicode())
>>> api.Command.nudge.options.stuff.default
u'documentation'
The arguments and options must not contain colliding names. They are both
merged together into the ``params`` attribute, another `plugable.NameSpace`
instance, as you can see:
>>> api.Command.nudge.params
NameSpace(<2 members>, sort=False)
>>> list(api.Command.nudge.params) # Iterates through the param names
['programmer', 'stuff']
When calling a command, its positional arguments can also be provided as
keyword arguments, and in any order. For example:
>>> api.Command.nudge(stuff='lines of code', programmer='Jason')
u'Jason, go write more lines of code!'
When a command plugin is called, the values supplied for its parameters are
put through a sophisticated processing pipeline that includes steps for
normalization, type conversion, validation, and dynamically constructing
the defaults for missing values. The details wont be covered here; however,
here is a quick teaser:
>>> from ipalib import Int
>>> class create_player(Command):
... takes_options = [
... 'first',
... 'last',
... Param('nick',
... normalize=lambda value: value.lower(),
... default_from=lambda first, last: first[0] + last,
... ),
... Param('points', type=Int(), default=0),
... ]
...
>>> cp = create_player()
>>> cp.finalize()
>>> cp.convert(points=" 1000 ")
{'points': 1000}
>>> cp.normalize(nick=u'NickName')
{'nick': u'nickname'}
>>> cp.get_default(first='Jason', last='DeRose')
{'nick': u'jderose', 'points': 0}
For the full details on the parameter system, see the
`frontend.parse_param_spec()` function, and the `frontend.Param` and
`frontend.Command` classes.
------------------------
Logging from your plugin
------------------------
After `plugable.Plugin.set_api()` is called, your plugin will have a
``self.log`` attribute. Plugins should only log through this attribute.
For example:
>>> class paint_house(Command):
...
... takes_args = ['color']
...
... def execute(self, color):
... """Uses self.log.error()"""
... if color not in ('red', 'blue', 'green'):
... self.log.error("I don't have %s paint!", color) # Log error
... return
... return 'I painted the house %s.' % color
...
Some basic knowledge of the Python ``logging`` module might be helpful. See:
http://www.python.org/doc/2.5.2/lib/module-logging.html
The important thing to remember is that your plugin should not configure
logging itself, but should instead simply use the ``self.log`` logger.
Also see the `plugable.API.bootstrap()` method for details on how the logging
is configured.
---------------------
Environment variables
---------------------
Plugins access various environment variables and run-time information through
``self.api.env`` (for convenience, ``self.env`` is equivalent).
When you create a fresh `plugable.API` instance, its ``env`` attribute is
likewise a freshly created `config.Env` instance, which will already be
populated with certain run-time information. For example:
>>> api = get_standard_api()
>>> list(api.env)
['bin', 'dot_ipa', 'home', 'ipalib', 'mode', 'script', 'site_packages']
Here is a quick overview of the run-time information:
============= ================================ =======================
Key Source or example value Description
============= ================================ =======================
bin /usr/bin Dir. containing script
dot_ipa ~/.ipa User config directory
home os.environ['HOME'] User home dir.
ipalib .../site-packages/ipalib Dir. of ipalib package
mode 'production' or 'unit_test' The mode ipalib is in
script sys.argv[0] Path of script
site_packages /usr/lib/python2.5/site-packages Dir. containing ipalib/
============= ================================ =======================
After `plugable.API.bootstrap()` has been called, the env instance will be
populated with all the environment information used by the built-in plugins.
This will typically be called before any plugins are registered. For example:
>>> len(api.env)
7
>>> api.bootstrap(in_server=True) # We want to execute, not forward
>>> len(api.env)
33
If your plugin requires new environment variables *and* will be included in
the freeIPA built-in plugins, you should add the defaults for your variables
in `ipalib.constants.DEFAULT_CONFIG`. Also, you should consider whether your
new environment variables should have any auto-magic logic to determine their
values if they haven't already been set by the time `config.Env._bootstrap()`,
`config.Env._finalize_core()`, or `config.Env._finalize()` is called.
On the other hand, if your plugin requires new environment variables and will
be installed in a 3rd-party package, your plugin should set these variables
in the module it is defined in.
`config.Env` values work on a first-one-wins basis... after a value has been
set, it can not be overridden with a new value. As any variables can be set
using the command-line ``-e`` global option or set in a configuration file,
your module must check whether a variable has already been set before
setting its default value. For example:
>>> if 'message_of_the_day' not in api.env:
... api.env.message_of_the_day = 'Hello, world!'
...
Your plugin can access any environment variables via ``self.env``.
For example:
>>> class motd(Command):
... """Print message of the day."""
...
... def execute(self):
... return self.env.message_of_the_day
...
>>> api.register(motd)
>>> api.finalize()
>>> api.Command.motd()
'Hello, world!'
Also see the `plugable.API.bootstrap_with_global_options()` method.
---------------------------------------------
Indispensable ipa script commands and options
---------------------------------------------
The ``console`` command will launch a custom interactive Python interpreter
session. The global environment will have an ``api`` variable, which is the
standard `plugable.API` instance found at ``ipalib.api``. All plugins will
have been loaded (well, except the backend plugins if ``in_server`` is False)
and ``api`` will be fully initialized. To launch the console from within the
top-level directory in the the source tree, just run ``ipa console`` from a
terminal, like this:
::
$ ./ipa console
By default, ``in_server`` is False. If you want to start the console in a
server context (so that all the backend plugins are loaded), you can use the
``-e`` option to set the ``in_server`` environment variable, like this:
::
$ ./ipa -e in_server=True console
You can specify multiple environment variables by including the ``-e`` option
multiple times, like this:
::
$ ./ipa -e in_server=True -e mode=dummy console
The space after the ``-e`` is optional. This is equivalent to the above command:
::
$ ./ipa -ein_server=True -emode=dummy console
The ``env`` command will print out the full environment in key=value pairs,
like this:
::
$ ./ipa env
If you use the ``--server`` option, it will forward the call to the server
over XML-RPC and print out what the environment is on the server, like this:
::
$ ./ipa env --server
The ``plugins`` command will show details of all the plugin that are loaded,
like this:
::
$ ./ipa plugins
----------------
Learning more...
----------------
To learn more about writing plugins, you should:
1. Look at some of the built-in plugins, like the frontend plugins in
`ipalib.plugins.f_user` and the backend plugins in
`ipa_server.plugins.b_ldap`.
2. Learn about the base classes for frontend plugins in `ipalib.frontend`.
3. Learn about the core plugin framework in `ipalib.plugable`.
'''
import plugable
from backend import Backend, Context
@@ -35,11 +640,34 @@ from frontend import Command, Object, Method, Property, Application
from ipa_types import Bool, Int, Unicode, Enum
from frontend import Param, DefaultFrom
def get_standard_api():
return plugable.API(
def get_standard_api(mode='dummy'):
"""
Return standard `plugable.API` instance.
This standard instance allows plugins that subclass from the following
base classes:
- `frontend.Command`
- `frontend.Object`
- `frontend.Method`
- `frontend.Property`
- `frontend.Application`
- `backend.Backend`
- `backend.Context`
"""
api = plugable.API(
Command, Object, Method, Property, Application,
Backend, Context,
)
if mode is not None:
api.env.mode = mode
return api
api = get_standard_api()
api = get_standard_api(mode=None)

12
ipalib/config.py
View File

@@ -202,8 +202,8 @@ class Env(object):
"""
self.__doing('_finalize_core')
self.__do_if_not_done('_bootstrap')
self._merge_config(self.conf)
if self.conf_default != self.conf:
if self.__d.get('mode', None) != 'dummy':
self._merge_config(self.conf)
self._merge_config(self.conf_default)
if 'in_server' not in self:
self.in_server = (self.context == 'server')
@@ -335,7 +335,13 @@ class Env(object):
"""
return key in self.__d
def __iter__(self): # Fix
def __len__(self):
"""
Return number of variables currently set.
"""
return len(self.__d)
def __iter__(self):
"""
Iterate through keys in ascending order.
"""

4
ipalib/plugable.py
View File

@@ -808,7 +808,7 @@ class API(DictProxy):
log.addHandler(stderr)
# Add file handler:
if self.env.mode == 'unit_test':
if self.env.mode in ('dummy', 'unit_test'):
return # But not if in unit-test mode
log_dir = path.dirname(self.env.log)
if not path.isdir(log_dir):
@@ -860,7 +860,7 @@ class API(DictProxy):
"""
self.__doing('load_plugins')
self.__do_if_not_done('bootstrap')
if self.env.mode == 'unit_test':
if self.env.mode in ('dummy', 'unit_test'):
return
util.import_plugins_subpackage('ipalib')
if self.env.in_server:

2
make-test
View File

@@ -11,7 +11,7 @@ do
if [[ -f $executable ]]; then
echo "[ $name: Starting tests... ]"
((runs += 1))
if $executable /usr/bin/nosetests -v --with-doctest
if $executable /usr/bin/nosetests -v --with-doctest --stop
then
echo "[ $name: Tests OK ]"
else

3
tests/util.py
View File

@@ -219,8 +219,7 @@ def get_api(**kw):
instance and a `TempHome` instance.
"""
home = TempHome()
api = ipalib.get_standard_api()
api.env.mode = 'unit_test'
api = ipalib.get_standard_api(mode='unit_test')
api.env.in_tree = True
for (key, value) in kw.iteritems():
api.env[key] = value