Introduction¶

Reg lets you write generic functions that dispatch on some of their arguments. To support this, Reg provides an implementation of multiple dispatch in Python. Reg goes beyond dispatching on the class of the arguments but can also dispatch on other aspects of arguments.

In other words: Reg lets you define methods outside their classes as plain Python functions. Reg in its basic use is like the single dispatch implementation described in Python PEP 443, but Reg provides a lot more flexibility.

Reg supports loose coupling. You can define a function in your core application or framework but provide implementations of this function outside of it.

Reg gives developers fine control over how to find implemenations of these functions. You can have multiple independent dispatch registries. For special use cases you can also register and look up other objects instead of functions.

What is Reg for? Reg offers infrastructure that lets you build more powerful frameworks – frameworks that can be extended and overridden in a general way. The Morepath web framework is built on top of Reg. Reg may seem like overkill to you. You may very well be right; it depends on what you’re building.

Example¶

Here is an example of Reg. First we define a generic function that dispatches based on the class of its obj argument:

import reg
@reg.dispatch('obj')
def title(obj):
   return "we don't know the title"

Now we create a few example classes for which we want to be able to use the title function we defined above.

class TitledReport(object):
   def __init__(self, title):
      self.title = title

class LabeledReport(object):
   def __init__(self, label):
      self.label = label

In TitledReport there’s an attribute called title but in the LabeledReport case we have an attribute label we want to use as the title. We will implement this behavior in a few plain python functions:

def titled_report_title(obj):
    return obj.title

def labeled_report_title(obj):
    return obj.label

We now create a Reg reg.Registry, and tell it about a few implementations for the title function:

registry = reg.Registry()
registry.register_function(
    title, titled_report_title, obj=TitledReport)
registry.register_function(
    title, labeled_report_title, obj=LabeledReport)

We then tell Reg to use it automatically using reg.implicit.Implicit.initialize():

from reg import implicit
implicit.initialize(registry.lookup())

Once we’ve done this, our generic title function works on both titled and labeled objects:

>>> titled = TitledReport('This is a report')
>>> labeled = LabeledReport('This is also a report')
>>> title(titled)
'This is a report'
>>> title(labeled)
'This is also a report'

Our example is over, so we reset the implicit registry set up before:

implicit.clear()

Why not just use plain functions or methods instead of generic functions? Often plain functions or methods will be the right solution. But not always – in this document we will examine a situation where generic functions are useful.

A Hypothetical CMS¶

Let’s look at how Reg works in the context of a hypothetical content management system (CMS).

This hypothetical CMS has two kinds of content item (we’ll add more later):

• a Document which contains some text.
• a Folder which contains a bunch of content entries, for instance Document instances.

This is the implementation of our CMS:

class Document(object):
   def __init__(self, text):
       self.text = text

class Folder(object):
   def __init__(self, entries):
       self.entries = entries

size methods¶

Now we want to add a feature to our CMS: we want the ability to calculate the size (in bytes) of any content item. The size of the document is defined as the length of its text, and the size of the folder is defined as the sum of the size of everything in it.

If we have control over the implementation of Document and Folder we can implement this feature easily by adding a size method to both classes:

class Document(object):
   def __init__(self, text):
       self.text = text

   def size(self):
       return len(self.text)

class Folder(object):
   def __init__(self, entries):
       self.entries = entries

   def size(self):
       return sum([entry.size() for entry in self.entries])

And then we can simply call the .size() method to get the size:

>>> doc = Document('Hello world!')
>>> doc.size()
12
>>> doc2 = Document('Bye world!')
>>> doc2.size()
10
>>> folder = Folder([doc, doc2])
>>> folder.size()
22

Note that the Folder size code is generic; it doesn’t care what the entries inside it are; if they have a size method that gives the right result, it will work. If a new content item Image is defined and we provide a size method for this, a Folder instance that contains Image instances will still be able to calculate its size. Let’s try this:

class Image(object):
    def __init__(self, bytes):
        self.bytes = bytes

    def size(self):
        return len(self.bytes)

When we add an Image instance to the folder, the size of the folder can still be calculated:

>>> image = Image('abc')
>>> folder.entries.append(image)
>>> folder.size()
25

Adding size from outside¶

So far we didn’t need Reg at all. But in the real world things may be a lot more complicated. We may be dealing with a content management system core where we cannot control the implementation of Document and Folder. What if we want to add a size calculation feature in an extension package?

We can fall back on good-old Python functions instead. We separate out the size logic from our classes:

def document_size(item):
    return len(item.text)

def folder_size(item):
    return sum([document_size(entry) for entry in item.entries])

Generic size¶

There is a problem with the above implementation however: folder_size is not generic anymore, but now depends on document_size. It would fail when presented with a folder with an Image in it:

>>> folder_size(folder)
Traceback (most recent call last):
  ...
AttributeError: ...

To support Image we first need an image_size function:

def image_size(item):
   return len(item.bytes)

We can now write a generic size function to get the size for any item we give it:

def size(item):
    if isinstance(item, Document):
        return document_size(item)
    elif isinstance(item, Image):
        return image_size(item)
    elif isinstance(item, Folder):
        return folder_size(item)
    assert False, "Unknown item: %s" % item

With this, we can rewrite folder_size to use the generic size:

def folder_size(item):
    return sum([size(entry) for entry in item.entries])

Now our generic size function will work:

>>> size(doc)
12
>>> size(image)
3
>>> size(folder)
25

All a bit complicated and hard-coded, but it works!

New File content¶

What if we now want to write a new extension to our CMS that adds a new kind of folder item, the File, with a file_size function?

class File(object):
   def __init__(self, bytes):
       self.bytes = bytes

def file_size(item):
    return len(item.bytes)

We would need to remember to adjust the generic size function so we can teach it about file_size as well. Annoying, tightly coupled, but sometimes doable.

But what if we are actually yet another party, and we have control of neither the basic CMS nor its size extension? We cannot adjust generic_size to teach it about File now! Uh oh!

Perhaps the implementers of the size extension were wise and anticipated this use case. They could have implemented size like this:

size_function_registry = {
   Document: document_size,
   Image: image_size,
   Folder: folder_size
}

def register_size(class_, function):
   size_function_registry[class_] = function

def size(item):
   return size_function_registry[item.__class__](item)

We can now use register_size to teach size how to get the size of a File instance:

register_size(File, file_size)

And it would work:

>>> size(File('xyz'))
3

This is quite a bit of custom work on the parts of the implementers, though. The API to manipulate the size registry is also completely custom. But you can do it.

New HtmlDocument content¶

What if we introduce a new HtmlDocument item that is a subclass of Document?

class HtmlDocument(Document):
    pass # imagine new html functionality here

Let’s try to get its size:

>>> htmldoc = HtmlDocument('<p>Hello world!</p>')
>>> size(htmldoc)
Traceback (most recent call last):
   ...
KeyError: ...

Uh oh, that doesn’t work! There’s nothing registered for the HtmlDocument class.

We need to remember to also call register_size for HtmlDocument. We can reuse document_size:

>>> register_size(HtmlDocument, document_size)

Now size will work:

>>> size(htmldoc)
19

This is getting rather complicated, requiring not only foresight and extra implementation work for the developers of size but also extra work for the person who wants to subclass a content item.

Hey, we should write a system that generalizes this and automates a lot of this, and gives us a more universal registry API, making our life easier! And that’s Reg.

Doing this with Reg¶

Let’s see how we could implement size using Reg.

First we need our generic size function:

def size(item):
    raise NotImplementedError

This function raises NotImplementedError as we don’t know how to get the size for an arbitrary Python object. Not very useful yet. We need to be able to hook the actual implementations into it. To do this, we first need to transform the size function to a generic one:

import reg
size = reg.dispatch('item')(size)

We can actually spell these two steps in a single step, as reg.dispatch() can be used as decorator:

@reg.dispatch('item')
def size(item):
    raise NotImplementedError

We can now register the various size functions for the various content items in a registry:

r = reg.Registry()
r.register_function(size, document_size, item=Document)
r.register_function(size, folder_size, item=Folder)
r.register_function(size, image_size, item=Image)
r.register_function(size, file_size, item=File)

We can now use our size function:

>>> size(doc, lookup=r.lookup())
12

>>> size(doc)
Traceback (most recent call last):
  ...
NoImplicitLookupError: Cannot lookup without explicit lookup argument because no implicit lookup was configured.

>>> size(folder, lookup=r.lookup())
Traceback (most recent call last):
  ...
NoImplicitLookupError: Cannot lookup without explicit lookup argument because no implicit lookup was configured.

Using reg.implicit.Implicit.initialize() we can specify an implicit lookup argument for all generic lookups so we don’t have to pass it in anymore:

from reg import implicit
implicit.initialize(r.lookup())

Now we can just call our new generic size:

>>> size(doc)
12

And it will work for folder too:

>>> size(folder)
25

It will work for subclasses too:

>>> size(htmldoc)
19

Reg knows that HtmlDocument is a subclass of Document and will find document_size automatically for you. We only have to register something for HtmlDocument if we would want to use a special, different size function for HtmlDocument.

Using classes¶

The previous example worked well for a single function to get the size, but what if we wanted to add a feature that required multiple methods, not just one?

Let’s imagine we have a feature to get the icon for a content object in our CMS, and that this consists of two methods, with a way to get a small icon and a large icon. We want this API:

from abc import ABCMeta, abstractmethod

class Icon(object):
    __metaclass__ = ABCMeta
    @abstractmethod
    def small(self):
        """Get the small icon."""

    @abstractmethod
    def large(self):
        """Get the large icon."""

Let’s implement the Icon API for Document:

def load_icon(path):
    return path # pretend we load the path here and return an image obj

class DocumentIcon(Icon):
   def __init__(self, document):
      self.document = document

   def small(self):
      if not self.document.text:
          return load_icon('document_small_empty.png')
      return load_icon('document_small.png')

   def large(self):
      if not self.document.text:
          return load_icon('document_large_empty.png')
      return load_icon('document_large.png')

The constructor of DocumentIcon receives a Document instance as its first argument. The implementation of the small and large methods uses this instance to determine what icon to produce depending on whether the document is empty or not.

We can call DocumentIcon an adapter, as it adapts the original Document class to provide an icon API for it. We can use it manually:

>>> icon_api = DocumentIcon(doc)
>>> icon_api.small()
'document_small.png'
>>> icon_api.large()
'document_large.png'

But we want to be able to use the Icon API in a generic way, so let’s create a generic function that gives us an implementation of Icon back for any object:

@reg.dispatch('obj')
def icon(obj):
    raise NotImplementedError

We can now register the DocumentIcon adapter class for this function and Document:

r.register_function(icon, DocumentIcon, obj=Document)

We can now use the generic icon to get Icon API for a document:

>>> api = icon(doc)
>>> api.small()
'document_small.png'
>>> api.large()
'document_large.png'

We can also register a FolderIcon adapter for Folder, a ImageIcon adapter for Image, and so on. For the sake of brevity let’s just define one for Image here:

class ImageIcon(Icon):
    def __init__(self, image):
        self.image = image

    def small(self):
        return load_icon('image_small.png')

    def large(self):
        return load_icon('image_large.png')

r.register_function(icon, ImageIcon, obj=Image)

Now we can use icon to retrieve the Icon API for any item in the system for which an adapter was registered:

>>> icon(doc).small()
'document_small.png'
>>> icon(doc).large()
'document_large.png'
>>> icon(image).small()
'image_small.png'
>>> icon(image).large()
'image_large.png'

Multiple dispatch¶

Sometimes we want to adapt more than one thing at the time. The canonical example for this is a web view lookup system. Given a request and a model, we want to find a view that represents these. The view needs to get the request, for parameter information, POST body, URL information, and so on. The view also needs to get the model, as that is what will be represented in the view.

You want to be able to vary the view depending on the type of the request as well as the type of the model.

Let’s imagine we have a Request class:

class Request(object):
    pass

We’ll use Document as the model class.

We want a generic view function that given a request and a model generates content for it:

@reg.dispatch('request', 'model')
def view(request, model):
    raise NotImplementedError

We now define a concrete view for Document:

def document_view(request, document):
    return "The document content is: " + document.text

Let’s register the view in the registry:

r.register_function(view, document_view,
                    request=Request, model=Document)

We now see why the second argument to register() is a list; so far we only supplied a single entry in it, but here we supply two, as we have two parameters on which to do dynamic dispatch.

Given a request and a document, we can now call view:

>>> request = Request()
>>> view(request, doc)
'The document content is: Hello world!'

replacing class methods¶

Reg generic functions can be used to replace methods, so that you can follow the open/closed principle and add functionality to a class without modifying it. This works for instance methods, but what about classmethod? This takes the class as the first argument, not an instance. You can configure @reg.dispatch decorator with a special Predicate instance that lets you dispatch on a class argument instead of an instance argument.

Here’s what it looks like:

@reg.dispatch(reg.match_class('cls', lambda cls: cls))
def something(cls):
    raise NotImplementedError()

Note the call to match_class() here. This lets us specify that we want to dispatch on the class, and we supply a lambda function that shows how to extract this from the arguments to something; in this case we simply want the cls argument.

Let’s use it:

def something_for_object(cls):
    return "Something for %s" % cls

r.register_function(something, something_for_object, cls=object)

class DemoClass(object):
    pass

When we now call something() with DemoClass as the first argument we get the expected output:

>>> something(DemoClass)
"Something for <class 'DemoClass'>"

This also knows about inheritance. So, you can write more specific implementations for particular classes:

class ParticularClass(object):
    pass

def something_particular(cls):
    return "Particular for %s" % cls

r.register_function(something, something_particular,
                    cls=ParticularClass)

When we call something now with ParticularClass as the argument, then something_particular is called:

>>> something(ParticularClass)
"Particular for <class 'ParticularClass'>"

Component lookup¶

You can look up the function that a function would dispatch to without calling it. You do this using the component method on the dispatch function:

>>> size.component(doc) is document_size
True

Getting all¶

As we’ve seen, Reg supports inheritance. size for instance was registered for Document instances, and is therefore also available of instances of its subclass, HtmlDocument:

>>> size.component(doc) is document_size
True
>>> size.component(htmldoc) is document_size
True

Using the special all function we can also get an iterable of all the components registered for a particular instance, including those of base classes. Right now this is pretty boring as there’s only one of them:

>>> list(size.all(doc))
[<function document_size at ...>]
>>> list(size.all(htmldoc))
[<function document_size at ...>]

We can make this more interesting by registering a special htmldocument_size to handle HtmlDocument instances:

def htmldocument_size(doc):
   return len(doc.text) + 1 # 1 so we can see a difference

r.register_function(size, htmldocument_size,
                    item=HtmlDocument)

size.all() for htmldoc now also gives back the more specific htmldocument_size:

>>> list(size.all(htmldoc))
[<function htmldocument_size at ...>, <function document_size at ...>]

Using the Registry directly¶

The key under which we register something in a registry in fact doesn’t need to be a function. We can register predicate for any immutable key such as a string:

r.register_predicates('some key', [reg.match_argname('obj')])

We can now register something for this key:

r.register_value('some key', [Document], 'some registered')

We can’t get it at it using a generic dispatch function anymore now. We can use the reg.Registry API instead. Here’s what to do:

>>> r.component('some key', Document)
'some registered'
>>> list(r.all('some key', Document))
['some registered']

Caching¶

We can turn a plain reg.Registry into a faster, caching class lookup using reg.CachingKeyLookup:

>>> caching = reg.CachingKeyLookup(r, 100, 100, 100)

Turning it back into a lookup gives us a caching version of what we had before:

>>> caching_lookup = caching.lookup()
>>> size(doc, lookup=caching_lookup)
12
>>> size(doc, lookup=caching_lookup)
12

You’ll have to trust us on this, but it’s faster the second time as the dispatch to document_size was cached!

Install Reg for development¶

First make sure you have virtualenv installed for Python 3.5.

Now create a new virtualenv somewhere for Reg’s development:

$ virtualenv /path/to/ve_reg

The goal of this is to isolate you from any globally installed versions of setuptools, which may be incompatible with the requirements of the buildout tool. You should also be able to recycle an existing virtualenv, but this method guarantees a clean one.

Clone Reg from github (https://github.com/morepath/reg) and go to the reg directory:

$ git clone git@github.com:morepath/reg.git
$ cd reg

Now we need to run bootstrap.py to set up buildout, using the Python from the virtualenv we’ve created before:

$ /path/to/ve_reg/bin/python bootstrap.py

This installs buildout, which can now set up the rest of the development environment:

$ bin/buildout

This will download and install various dependencies and tools.

Running the tests¶

You can run the tests using py.test. Buildout will have installed it for you in the bin subdirectory of your project:

$ bin/py.test reg

To generate test coverage information as HTML do:

$ bin/py.test --cov reg --cov-report html

You can then point your web browser to the htmlcov/index.html file in the project directory and click on modules to see detailed coverage information.

Running the documentation tests¶

The documentation contains code. To check these code snippets, you can run this code using this command:

$ bin/sphinxpython bin/sphinx-build  -b doctest doc out

Building the HTML documentation¶

To build the HTML documentation (output in doc/build/html), run:

$ bin/sphinxbuilder

Various checking tools¶

The buildout will also have installed flake8, which is a tool that can do various checks for common Python mistakes using pyflakes, check for PEP8 style compliance and can do cyclomatic complexity checking. To do pyflakes and pep8 checking do:

$ bin/flake8 reg

To also show cyclomatic complexity, use this command:

$ bin/flake8 --max-complexity=10 reg

Reg History¶

The Reg code went through a bit of history:

The core registry (mapping) code was conceived by Thomas Lotze and myself as a speculative sandbox project in January of 2010. It was called iface then:

http://svn.zope.org/Sandbox/faassen/iface/

In early 2012, I was at a sprint in Nürnberg, Germany organized by Novareto. Inspired by discussions with the sprint participants, particularly the Cromlech developers Souheil Chelfouh and Alex Garel, I created a project called Crom:

https://github.com/faassen/crom

Crom focused on rethinking component and adapter registration and lookup APIs, but was still based on zope.interface for its fundamental AdapterRegistry implementation and the Interface metaclass. I worked a bit on Crom after the sprint, but soon I moved on to other matters.

At the Plone conference held in Arnhem, the Netherlands in October 2012, I gave a lightning talk about Crom. I figured what Crom needed was a rewrite of the core adapter registry, i.e. what was in the iface project. In the end of 2012 I mailed Thomas Lotze asking whether I could merge iface into Crom, and he gave his kind permission.

The core registry code of iface was never quite finished however, and while the iface code was now in Crom, Crom didn’t use it yet. Thus it lingered some more.

In July 2013 in development work for CONTACT (contact.de), I found myself in need of clever registries. Crom also had some configuration code intermingled with the component architecture code and I didn’t want this anymore.

So I reorganized the code yet again into another project, this one: Reg. I then finished the core mapping code and hooked it up to the Crom-style API, which I refactored further. For interfaces, I used Python’s abc module.

For a while during internal development this codebase was called Comparch, but this conflicted with another name so I decided to call it Reg, short for registry, as it’s really about clever registries more than anything else.

After my first announcement of Reg to the world in September 2013 I got the question why I shouldn’t just use PEP 443, which has a generic function implementation (single dispatch). I started thinking I should convert Reg to a generic function implementation, as it was already very close. After talking to some people about this at PyCon DE in october, I did the refactoring to use generic functions throughout and no interfaces for lookup. I used this version of Reg in Morepath for about a year.

In October 2014 I had some experience with using Reg and found some of its limitations:

• Reg would try to dispatch on all non-keyword arguments of a function. This is not what is desired in many cases. We need a way to dispatch only on specified arguments and leave others alone.
• Reg had an undocumented predicate subsystem used to implement view lookup in Morepath. I realized that it could not be properly used to dispatch on the class of an instance without reorganizing Reg in a major way.
• I realized that such a reorganized predicate system could actually be used to generalize the way Reg worked based on how predicates worked.
• This would allow predicates to play along in Reg’s caching infrastructure, which could then speed up Morepath’s view lookups.
• A specific use case to replace class methods caused me to introduce reg.classgeneric. This could be subsumed in a generalized predicate infrastructure as well.

So in October 2014, I refactored Reg once again in the light of this. This results in a smaller, more unified codebase that has more features and is hopefully also faster.

Brief history of Zope Component Architecture¶

Reg is heavily inspired by zope.interface and zope.component, by Jim Fulton and a lot of Zope developers. zope.interface has a long history, going all the way back to December 1998, when a scarecrow interface package was released for discussion:

http://old.zope.org/Members/jim/PythonInterfaces/Summary/

http://old.zope.org/Members/jim/PythonInterfaces/Interface/

A later version of this codebase found itself in Zope, as Interface:

http://svn.zope.org/Zope/tags/2-8-6/lib/python/Interface/

A new version called zope.interface was developed for the Zope 3 project, somewhere around the year 2001 or 2002 (code historians, please dig deeper and let me know). On top of this a zope.component library was constructed which added registration and lookup APIs on top of the core zope.interface code.

zope.interface and zope.component are widely used as the core of the Zope 3 project. zope.interface was adopted by other projects, such as Zope 2, Twisted, Grok, BlueBream and Pyramid.

0.9.3 (2016-07-18)¶

• Minor fixes to documentation.
• Add tox test environments for Python 3.4 and 3.5, PyPy 3 and PEP 8.
• Make Python 3.5 the default Python environment.
• Changed location NoImplicitLookupError was imported from in __init__.py.

0.9.2 (2014-11-13)¶

• Reg was a bit too strict; when you had multiple (but not single) predicates, Reg would raise KeyError when you put in an unknown key. Now they’re just being silently ignored, as they don’t do any harm.
• Eliminated a check in ArgExtractor that could never take place.
• Bring test coverage back up to 100%.
• Add converage configuration to ignore test files in coverage reporting.

0.9.1 (2014-11-11)¶

• A bugfix in the behavior of the fallback logic. In situations with multiple predicates of which one is a class predicate it was possible for a fallback not to be found even though a fallback was available.

0.9 (2014-11-11)¶

Total rewrite of Reg! This includes a range of changes that can break code. The primary motivations for this rewrite:

• unify predicate system with class-based lookup system.
• extract dispatch information from specific arguments instead of all arguments.

Some specific changes:

• Replaced @reg.generic decorator with @reg.dispatch() decorator. This decorator can be configured with predicates that extract information from the arguments. Rewrite this:

  @reg.generic
  def foo(obj):
     pass
  

  to this:

  @reg.dispatch('obj')
  def foo(obj):
     pass
  

  And this:

  @reg.generic
  def bar(a, b):
      pass
  

  To this:

  @reg.dispatch('a', 'b')
  def bar(a, b):
      pass
  

  This is to get dispatch on the classes of these instance arguments. If you want to match on the class of an attribute of an argument (for instance) you can use match_instance with a function:

  @reg.dispatch(match_instance('a', lambda a: a.attr))
  

  The first argument to match_instance is the name of the predicate by which you refer to it in register_function.

  You can also use match_class to have direct dispatch on classes (useful for replicating classmethods), and match_key to have dispatch on the (immutable) value of the argument (useful for a view predicate system). Like for match_instance, you supply functions to these match functions that extract the exact information to dispatch on from the argument.

• The register_function API replaces the register API to register a function. Replace this:

  r.register(foo, (SomeClass,), dispatched_to)
  

  with:

  r.register_function(foo, dispatched_to, obj=SomeClass)
  

  You now use keyword parameters to indicate exactly those arguments specified by reg.dispatch() are actually predicate arguments. You don’t need to worry about the order of predicates anymore when you register a function for it.

• The new classgeneric functionality is part of the predicate system now; you can use reg.match_class instead. Replace:

  @reg.classgeneric
  def foo(cls):
     pass
  

  with:

  @reg.dispatch(reg.match_class('cls', lambda cls: cls))
  def foo(cls):
      pass
  

  You can do this with any argument now, not just the first one.

• pep443 support is gone. Reg is focused on its own dispatch system.

• Compose functionality is gone – it turns out Morepath doesn’t use lookup composition to support App inheritance. The cached lookup functionality has moved into registry.py and now also supports caching of predicate-based lookups.

• Dependency on the future module is gone in favor of a small amount of compatibility code.

0.8 (2014-08-28)¶

• Added a @reg.classgeneric. This is like @reg.generic, but the first argument is treated as a class, not as an instance. This makes it possible to replace @classmethod with a generic function too.
• Fix documentation on running documentation tests. For some reason this did not work properly anymore without running sphinxpython explicitly.
• Optimization: improve performance of generic function calls by employing lookup_mapply instead of general mapply, as we only care about passing in the lookup argument when it’s defined, and any other arguments should work as before. Also added a perf.py which is a simple generic function timing script.

0.7 (2014-06-17)¶

• Python 2.6 compatibility. (Ivo van der Wijk)
• Class maps (and thus generic function lookup) now works with old style classes as well.
• Marked as production/stable now in setup.py.

0.6 (2014-04-08)¶

• Removed unused code from mapply.py.
• Typo fix in API docs.

0.5 (2014-01-21)¶

• Make reg.ANY public. Used for predicates that match any value.

0.4 (2014-01-14)¶

• arginfo has been totally rewritten and is now part of the public API of reg.

0.3 (2014-01-06)¶

• Experimental Python 3.3 support thanks to the future module.

0.2 (2013-12-19)¶

• If a generic function implementation defines a lookup argument that argument will be the lookup used to call it.
• Added reg.mapply(). This allows you to call things with more keyword arguments than it accepts, ignoring those extra keyword args.
• A function that returns None is not assumed to fail, so no fallback to the original generic function is triggered anymore.
• An optional precalc facility is made available on Matcher to avoid some recalculation.
• Implement a specific PredicateMatcher that matches a value on predicate.

0.1 (2013-10-28)¶

• Initial public release.