The PEAK Developers' Center   PEAK-Rules UserPreferences
 
HelpContents Search Diffs Info Edit Subscribe XML Print View

Creating Generic Functions using PEAK-Rules

PEAK-Rules is a highly-extensible framework for creating and using generic functions, from the very simple to the very complex. Out of the box, it supports multiple-dispatch on positional arguments using tuples of types, full predicate dispatch using strings containing Python expressions, and CLOS-like method combining. (But the framework allows you to mix and match dispatch engines and custom method combinations, if you need or want to.)

Basic usage:

>>> from peak.rules import abstract, when, around, before, after

>>> @abstract()
... def pprint(ob):
...     """A pretty-printing generic function"""

>>> @when(pprint, (list,))
... def pprint_list(ob):
...     print "pretty-printing a list"

>>> @when(pprint, "isinstance(ob,list) and len(ob)>50")
... def pprint_long_list(ob):
...     print "pretty-printing a long list"

>>> pprint([1,2,3])
pretty-printing a list

>>> pprint([42]*1000)
pretty-printing a long list

>>> pprint(42)
Traceback (most recent call last):
  ...
NoApplicableMethods: ...

PEAK-Rules works with Python 2.3 and up -- just omit the @ signs if your code needs to run under 2.3. Also, note that with PEAK-Rules, any function can be generic: you don't have to predeclare a function as generic. (The abstract decorator is used to declare a function with no default method; i.e., one that will give a NoApplicableMethods if no rules match the arguments it's invoked with, as opposed to executing a default implementation.)

PEAK-Rules is still under development; it lacks much in the way of error checking, so if you mess up your rules, it may not be obvious where or how you did. User documentation is also lacking, although there are extensive doctests describing and testing most of its internals, including:

(Please note that these documents are still in a state of flux and some may still be incomplete or disorganized, prior to the first official release.)

Source distribution snapshots are generated daily, but you can also update directly from the development version in SVN.

Table of Contents

Developer's Guide

XXX basics tutorial should go here

Method Combination and Custom Method Types

Sometimes, more than one method of a generic function applies in a given circumstance. For example, you might need to sum the results of a series of pricing rules in order to compute a product's price. Or, sometimes you'd like a method to be able to modify the result of a less-specific method.

For these scenarios, you will want to use "method combination", either using PEAK-Rules' built-in method decorators, or custom method types of your own.

Using next_method

By default, a generic function will only invoke the most-specific applicable method. However, if you add a next_method argument to the beginning of an individual method's signature, you can use it to call the "next method" that applies. That is, the second-most-specific method. If that method also has a next_method argument, it too will be able to invoke the next method after it, and so on, down through all the applicable methods. For example:

>>> from peak.rules import DispatchError

>>> @abstract()
... def foo(bar, baz):
...     """Foo bar and baz"""

>>> @when(foo, "bar>1 and baz=='spam'")
... def foo_one_spam(next_method, bar, baz):
...     return bar + next_method(bar, baz)

>>> @when(foo, "baz=='spam'")
... def foo_spam(bar, baz):
...     return 42

>>> @when(foo, "baz=='blue'")
... def foo_spam(next_method, bar, baz):
...     # if next_method is an instance of DispatchError, it means
...     # that calling it will raise that error (NoApplicableMethods
...     # or AmbiguousMethods)
...     assert isinstance(next_method, DispatchError)
... 
...     # but we'll call it anyway, just to demo the error
...     return 22 + next_method(bar, baz)

>>> foo(2,"spam")   # 2 + 42
44

>>> foo(2,"blue")   # 22 + ...no next method!
Traceback (most recent call last):
  File ... combiners.txt... in foo_spam
    return 22 + next_method(self,bar,baz)
...
NoApplicableMethods: ...

Notice that next_method comes before self in the arguments if the generic function is an instance method. (If used, it must be the very first argument of the method.) Its value is supplied automatically by the generic function machinery, so when you call next_method you do not have to care whether the next method needs to know its next method; just pass in all of the other arguments (including self if applicable) and the next_method implementation will do the rest.

Also notice that methods that do not call their next method do not need to have a next_method argument. If a method calls next_method when there are no further methods available, NoApplicableMethods is raised. Similarly, if there is more than one "next method" and they are all equally specific (i.e. ambiguous), then AmbiguousMethods is raised.

Most of the time, you will know when writing a routine whether it's safe to call next_method. But sometimes you need a routine to behave differently depending on whether a next method is available. If calling next_method will raise an error, then next_method will be an instance of the error class, so you can detect it with isinstance(). If there are no remaining methods, then next_method will be an instance of NoApplicableMethods, and if the next method is ambiguous, it will be an AmbiguousMethods instance. In either case, calling next_method will raise that error with the supplied arguments. (And DispatchError is a base class of both AmbiguousMethods and NoApplicableMethods, so you can just check for that.)

Before/After Methods

Sometimes you'd like for some additional validation or notification to occur before or after the "normal" or "primary" methods. This is what "before", "after", and "around" methods are for. For example:

>>> class BankAccount:
...
...     def __init__(self,balance,protection=0):
...         self.balance = balance
...         self.protection = protection
...
...     def withdraw(self,amount):
...         """Withdraw 'amount' from bank"""
...         self.balance -= amount      # nominal case
...
...     @before(withdraw, "amount>self.balance and self.protection==0")
...     def prevent_overdraft(self, amount):
...         raise ValueError("Insufficient funds")
...
...     @after(withdraw, "amount>self.balance")
...     def automatic_overdraft(self, amount):
...         print "Transferring",-self.balance,"from overdraft protection"
...         self.protection += self.balance
...         self.balance = 0

>>> acct = BankAccount(200)
>>> acct.withdraw(400)
Traceback (most recent call last):
...
ValueError: Insufficient funds

>>> acct.protection = 300
>>> acct.withdraw(400)
Transferring 200 from overdraft protection
>>> acct.balance
0
>>> acct.protection
100

This specific example could have been written entirely with normal when() methods, by using more complex conditions. But, in more complex scenarios, where different modules may be adding rules to the same generic function, it's not possible for one module to predict whether its conditions will be more specific than another's, and whether it will need to call next_method, etc.

So, generic functions offer before() and after() methods, that run before and after the when() (aka "primary") methods, respectively. Unlike primary methods, before() and after() methods:

  • Are allowed to have ambiguous conditions (and if they do, they execute in the order in which they were added to the generic function)
  • Are always run when their conditions apply, with no need to call next_method to invoke the next method
  • Cannot return a useful value and do not have access to the return value of any other method

The overall order of method execution is:

  1. All applicable before() methods, from most-specific to least-specific, methods at the same level of specificity execute in the order they were added.
  2. Most-specifc primary method, which may optionally chain to less-specific primary methods. AmbiguousMethods or NoApplicableMethods may be raised if the most-specific method is ambiguous or no primary methods are applicable.
  3. All applicable after() methods, from least-specific to most-specific, with methods at the same level of specificity executing in the reverse order from the order they were added. (In other words, the more specific the after() condition, the "more after" it gets run!)

If any of these methods raises an uncaught exception, the overall function execution terminates at that point, and methods later in the order are not run.

"Around" Methods

Sometimes you need to recognize certain special cases, and perhaps not run the entire generic function, or need to alter its return value in some way, or perhaps trap and handle certain exceptions, etc. You can do this with "around" methods, which run "around" the entire "before/primary/after" sequence described in the previous section.

A good way to think of this is that it's as if the "around" methods form a separate generic function, whose default (least-specific) method is the original, "inner" generic function.

When "around" methods are applicable on a given invocation of the generic function, the most-specific "around" method is invoked. It may then choose to call its next_method to invoke the next-most-specific "around" method, and so on. When there are no more "around" methods, calling next_method instead invokes the "before", "primary", and "after" methods, according to the sequence described in the previous section. For example:

>>> @around(BankAccount.withdraw, "amount > self.balance")
... def overdraft_fee(next_method,self,amount):
...     print "Adding overdraft fee of $25"
...     return next_method(self,amount+25)

>>> acct.withdraw(20)
Adding overdraft fee of $25
Transferring 45 from overdraft protection

Values

Sometimes, if you're defining a generic function whose job is to classify things, it can get to be a pain defining a bunch of functions or lambdas just to return a few values -- especially if the generic function has a complex signature! So peak.rules provides a convenience function, value() for doing this:

>>> from peak.rules import value
>>> value(42)
value(42)

>>> value(42)('whatever')
42

>>> classify = abstract(lambda age:None)

>>> when(classify, "age<2")(value("infant"))
value('infant')

>>> when(classify, "age<13")(value("preteen"))
value('preteen')

>>> when(classify, "age<5")(value("preschooler"))
value('preschooler')

>>> when(classify, "age<20")(value("teenager"))
value('teenager')

>>> when(classify, "age>=20")(value("adult"))
value('adult')

>>> when(classify, "age>=55")(value("senior"))
value('senior')

>>> when(classify, "age==16")(value("sweet sixteen"))
value('sweet sixteen')

>>> classify(17)
'teenager'

>>> classify(42)
'adult'

Method Combination

The combine_using() decorator marks a function as yielding its method results (most-specific to least-specific, with later-defined methods taking precedence), and optionally specifies how the resulting iteration will be post-processed:

>>> from peak.rules import combine_using

Let's take a look at how it works, by trying it with different ways of postprocessing on an example generic function. We'll start by defining a function to recreate a generic function with the same set of methods, so you can see what happens when we pass different arguments to combine_using:

>>> class A: pass
>>> class B(A): pass
>>> class C(A, B): pass
>>> class D(B, A): pass

>>> def demo(*args):
...     """We'll be setting this function up multiple times, so we do it in
...        a function.  In normal code, you won't need this outer function!
...     """
...     @combine_using(*args)
...     def func(ob):
...         return "default"
...
...     when(func, (object,))(value("object"))
...     when(func, (int,))   (value("int"))
...     when(func, (str,))   (value("str"))
...     when(func, (A,))     (value("A"))
...     when(func, (B,))     (value("B"))
...
...     return func

In the simplest case, you can just call @combine_using() with no arguments, and get a generic function that yields the results returned by its methods, in order from most-specific to least-specific:

>>> func = demo()

>>> list(func(A()))
['A', 'object', 'default']

>>> list(func(42))
['int', 'object', 'default']

In the event of ambiguity between methods, methods defined later are called first:

>>> list(func(C()))
['B', 'A', 'object', 'default']

>>> list(func(D()))
['B', 'A', 'object', 'default']

Passing a function to @combine_using(), however, makes it wrap the result iterator with that function, e.g.:

>>> func = demo(list)

>>> func(A())
['A', 'object', 'default']

While including abstract anywhere in the wrapper sequence makes the function abstract (i.e., it omits the original function's body from the defined methods):

>>> func = demo(abstract, list)

>>> func(A())  # 'default' isn't included any more:
['A', 'object']

You can also include more than one function in the wrapper list, and they will be called on the result iterator, first function outermost, ignoring any abstract in the sequence:

>>> func = demo(str.title, ' '.join)

>>> func(B())
'B A Object Default'

>>> func = demo(str.title, abstract, ' '.join)

>>> func(B())
'B A Object'

Some stdlib functions you might find useful for combine_using() include:

  • itertools.chain
  • sorted
  • reversed
  • list
  • set
  • "".join (or other string)
  • any
  • all
  • sum
  • min
  • max

(And of course, you can write and use arbitrary functions of your own.)

By the way, when using "around" methods with a method combination, the innermost next_method will return the fully processed combination of all the "when" methods, with the "before/after" methods running before and after the result is returned:

>>> from peak.rules import before, after, around

>>> def b(ob): print "before"
>>> def a(ob): print "after"
>>> def ar(next_method, ob):
...     print "entering around"
...     print next_method(ob)
...     print "leaving around"

>>> b = before(func, ())(b)
>>> a = after(func, ())(a)
>>> ar = around(func, ())(ar) 

>>> func(B())
entering around
before
after
B A Object
leaving around

Custom Method Types

If the standard before/after/around/when/combine_using decorators don't work for your application, you can create custom ones by defining your own "method types" and decorators.

Suppose, for example, that you are using a "pricing rules" generic function that operates by summing its methods' return values to produce a total:

>>> @combine_using(sum)
... def getPrice(product, customer=None, options=()):
...     """Get this product's price"""
...     return 0    # base price for arbitrary items

>>> class Product:
...     @when(getPrice)
...     def __addBasePrice(self, customer, options):
...         """Always include the product's base price"""
...         return self.base_price

>>> @when(getPrice, "'blue suede' in options")
... def blueSuedeUpcharge(product,customer,options):
...     return 24

>>> getPrice("arbitrary thing")
0

>>> shoes = Product()
>>> shoes.base_price = 42

>>> getPrice(shoes)
42

>>> getPrice(shoes, options=['blue suede'])
66

This is useful, sure, but what if you also want to be able to compute discounts or tax as a percentage of the total, rather than as flat additional amounts?

We can do this by implementing a custom "method type" and a corresponding decorator, to let us mark rules as computing a discount instead of a flat amount.

We'll start by defining the template that will be used to generate our method's implementation.

This format for method templates is taken from the DecoratorTools package's @template_method decorator. $args is used in places where the original generic function's calling signature is needed, and all local variables should be named so as not to conflict with possible argument names. The first argument of the template method will be the generic function the method is being used with, and all other arguments are defined by the method type's creator.

In our case, we'll need two arguments: one for the "body" (the discount method being decorated) and one for the "next method" that will be called to get the base price:

>>> def discount_template(__func, __body, __next_method):
...     return """
...     __price = __next_method($args)
...     return __price - (__body($args) * __price)
...     """

Okay, that's the easy bit. Now we need to define a bunch of other stuff to turn it into a method type and a decorator:

>>> from peak.rules.core import Around, MethodList, compile_method, \
...     combine_actions

>>> class DiscountMethod(Around):
...     """Subtract a discount"""
...
...     def override(self, other):
...         if self.__class__ == other.__class__:
...             return self.override(other.tail)  # drop the other one
...         return self.tail_with(combine_actions(self.tail, other))
...
...     def compiled(self, engine):
...         body = compile_method(self.body, engine)
...         next = compile_method(self.tail, engine)
...         return engine.apply_template(discount_template, body, next)

>>> discount_when = DiscountMethod.make_decorator(
...     "discount_when", "Discount price by the returned multiplier"
... )

>>> DiscountMethod >> MethodList    # mark precedence
<class 'peak.rules.core.MethodList'>

The make_decorator() method of Method objects lets you create decorators similar to when(), that we can now use to add a discount:

>>> @discount_when(getPrice, 
...    "customer=='Elvis' and 'blue suede' in options and product is shoes"
... )
... def ElvisGetsTenPercentOff(product,customer,options):
...     return .1

>>> getPrice(shoes)
42

>>> print getPrice(shoes, 'Elvis', options=['blue suede'])
59.4

>>> getPrice(shoes, 'Elvis')     # no suede, no discount!
42
XXX
This is still pretty hard; but without some real-world use cases for custom methods, it's hard to tell how to streamline the common cases.

Porting Code from RuleDispatch

The major design differences between PEAK-Rules and RuleDispatch are:

  1. It's designed for extensibility/pluggability from the ground up
  2. It's built from the ground up using generic functions instead of adaptation, so its code is a lot more straightforward. (The current implementation, combined with all its dependencies, is roughly the same number of lines as RuleDispatch without any of its dependencies -- and already has features that can't even be added to RuleDispatch.)
  3. It generates custom bytecode for each generic function, to minimize calling and interpreter overhead, and to potentially allow compatibility with Psyco and PyPy in the future. (Currently, neither Psyco nor PyPy support the "computed jump" trick used in the generated code, so don't try to Psyco-optimize any generic functions yet - it'll probably core dump!)

Because of its exensible design, PEAK-Rules can use custom-tuned engines for specific application scenarios, and over time it may evolve the ability to accept "tuning hints" to adjust the indexing techniques for special cases.

PEAK-Rules also supports the full method combination semantics of RuleDispatch using a new decentralized approach, that allows you to easily create new method types or combination semantics, complete with their own decorators (like when, around, etc.)

These decorators also all work with existing functions; you do not have to predeclare a function generic in order to use it. You can also omit the condition from the decorator call, in which case the effect is the same as RuleDispatch's strategy.default, i.e. there is no condition. Thus, you can actually use PEAK-Rules's around() as a quick way to monkeypatch existing functions, even ones defined by other packages. (And the decorators use the DecoratorTools package, so you can omit the @ signs for Python 2.3 compatibility.)

RuleDispatch was always conceived as a single implementation of a single dispatch algorithm intended to be "good enough" for all uses. Guido's argument on the Py3K mailing list, however, was that applications with custom dispatch needs should write custom dispatchers. And I almost agree -- except that I think they should get a RuleDispatch-like dispatcher for free, and be able to tune or write ones to plug in for specialized needs.

The kicker was that Guido's experiment with type-tuple caching (a predecessor algorithm to the Chambers-and-Chen algorithm used by RuleDispatch) showed it to be fast enough for common uses, even without any C code, as long as you were willing to do a little code generation. The code was super-small, simple, and fast enough that it got me thinking it was good enough for maybe 50% of what you need generic functions for, especially if you added method combination.

And thus, PEAK-Rules was born, and RuleDispatch doomed to obsolescence. (It didn't help that RuleDispatch was a hurriedly-thrown-together experiment, with poor testing and little documentation, either.)

So, if you are currently using RuleDispatch, we strongly advise that you port your code. To convert the most common RuleDispatch usages, simply do the following:

  • Replace @dispatch.on() and @dispatch.generic() with @abstract()
  • Replace @func.when(sig) with @when(func, sig) (and the same for before, after, and around)
  • When replacing @func.when(type) calls where func was defined with @dispatch.on, use @func.when("isinstance(arg, type)"), where arg is the argument that was named in the @dispatch.on() call.

RuleDispatch Emulation

If your code doesn't use much of the RuleDispatch API, you may be able to use PEAK-Rules' "emulation API", which supports the following RuleDispatch APIs:

  • dispatch.on, dispatch.generic, and dispatch.as``
  • strategy.default, strategy.Min, strategy.Max
  • DispatchError, NoApplicableMethods, and AmbiguousMethod errors
  • The when(), before(), after() and around() methods of generic functions.

(Note that some APIs may issue deprecation warnings (e.g. dispatch.as), and over time, the entire API will be deprecated. Please update your code as soon as practical.)

The emulation API does NOT support:

  • custom combiners (use custom method types instead)
  • The addMethod or __setitem__ APIs for adding rules
  • the clone() method of generics created with dispatch.on
  • PyProtocols (i.e., interfaces cannot be used for dispatching)

In the future, a PyProtocols emulation API may be added, but it doesn't exist yet.

To use the emulation API, simply import dispatch from peak.rules:

>>> from peak.rules import dispatch

>>> @dispatch.generic()     # roughly equivalent to @abstract()
... def a_function(an_arg, other_arg):
...     """Blah"""

>>> @a_function.when((int, str))
... def a_when_int_str(an_arg, other_arg):
...     print "int and str"

>>> a_function(42, "blue")
int and str

>>> a_function("blue", 42)
Traceback (most recent call last):
  ...
NoApplicableMethods: (('blue', 42), {})

Whether you use dispatch.generic or dispatch.on to define a generic function, you can begin using peak.rules.when to declare methods immediately:

>>> @when(a_function, (str, int))
... def a_when_str_int(an_arg, other_arg):
...     print "str and int"

>>> a_function("blue", 42)
str and int

This means that you don't have to update your entire codebase at once; you can port your method definitions incrementally, if desired.

Mailing List

Please direct questions regarding this package to the PEAK mailing list; see http://www.eby-sarna.com/mailman/listinfo/PEAK/ for details.


PythonPowered
EditText of this page (last modified 2010-08-17 19:58:51)
FindPage by browsing, title search , text search or an index
Or try one of these actions: AttachFile, DeletePage, LikePages, LocalSiteMap, SpellCheck