Total Network Tools 2002

home *** CD-ROM | disk | FTP | other *** search

/ Total Network Tools 2002 / NextStepPublishing-TotalNetworkTools2002-Win95.iso / Archive / Misc Servers / Zope.exe / EXTENSIONCLASS.REF < prev next >

Wrap

Text File | 2000-02-16 | 33.0 KB | 772 lines

<h1>Extension Classes, Python Extension Types Become Classes</h1> Jim Fulton, Digital Creations, Inc. jim@digicool.com <a href="COPYRIGHT.html">Copyright (C) 1996-1998, Digital Creations</a>. <h2>Abstract</h2> A lightweight mechanism has been developed for making Python extension types more class-like. Classes can be developed in an extension language, such as C or C++, and these classes can be treated like other python classes: <ul><li>They can be sub-classed in python, <li>They provide access to method documentation strings, and <li>They can be used to directly create new instances. </ul> An example class shows how extension classes are implemented and how they differ from extension types. Extension classes provide additional extensions to class and instance semantics, including: <ul><li>A protocol for accessing subobjects "in the context of" their containers. This is used to implement custom method types and <a href="Acquisition.html">environmental acquisition</a>. <li>A protocol for overriding method call semantics. This is used to implement "synchonized" classes and could be used to implement argument type checking. <li>A protocol for class initialization that supports execution of a special <code>__class_init__</code> method after a class has been initialized. </ul> Extension classes illustrate how the Python class mechanism can be extended and may provide a basis for improved or specialized class models. <h2>Releases</h2> To find out what's changed in this release, see the <a href="release.html">release notes</a>. <h2>Problem</h2> Currently, Python provides two ways of defining new kinds of objects: <ul><li>Python classes <li>Extension types </ul> Each approach has it's strengths. Extension types provide much greater control to the programmer and, generally, better performance. Because extension types are written in C, the programmer has greater access to external resources. (Note that Python's use of the term type has little to do with the notion of type as a formal specification.) Classes provide a higher level of abstraction and are generally much easier to develop. Classes provide full inheritance support, while support for inheritance when developing extension types is very limited. Classes provide run-time meta-data, such as method documentation strings, that are useful for documentation and discovery. Classes act as factories for creating instances, while separate functions must be provided to create instances of types. It would be useful to combine the features of the two approaches. It would be useful to be able to have better support for inheritance for types, or to be able to subclass from types in Python. It would be useful to be able to have class-like meta-data support for types and the ability to construct instances directly from types. Our software is developed in Python. When necessary, we convert debugged Python routines and classes to C for improved performance. In most cases, a small number of methods in a class is responsible for most of the computation. It should be possible to convert only these methods to C, while leaving the other method in Python. A natural way to approach this is to create a base class in C that contains only the performance-critical aspects of a class' implementation and mix this base class into a Python class. We have need, in a number of projects, for semantics that are slightly different than the usual class and instance semantics, yet we don't want to do most of our development in C. For example, we have developed a persistence mechanism <a href="#1">[1]</a> that redefines <code>__getattr__</code> and <code>__setattr__</code> to take storage-related actions when object state is accessed or modified. We want to be able to take certain actions on every attribute reference, but for python class instances, <code>__getattr__</code> is only called when attribute lookup fails by normal means. As another example, we would like to have greater control over how methods are bound. Currently, when accessing a class instance attribute, the attribute value is bound together with the instance in a method object if and only if the attribute value is a python function. For some applications, we might also want to be able to bind extension functions, or other types of callable objects, such as HTML document templates <a href="#2">[2]</a>. Furthermore, we might want to have greater control over how objects are bound. For example, we might want to bind instances and callable objects with special method objects that assure that no more than one thread accesses the object or method at one time. We can provide these special semantics in extension types, but we wish to provide them for classes developed in Python. <h2>Background</h2> At the first Python Workshop, Don Beaudry presented work <a href="#3">[3]</a> done at V.I. Corp to integrate Python with C++ frameworks. This system provided a number of important features, including: <ul><li>Definition of extension types that provide class-like meta-data and that can be called to create instances. <li>Ability to subclass in python from C types. <li>Ability to define classes in python who's data are stored as C structures rather than in dictionaries to better interface to C and C++ libraries, and for better performance. <li>Less dynamic data structures. In particular, the data structure for a class is declared during class definition. <li>Support for enumeration types. </ul> This work was not released, initially. Shortly after the workshop, changes were made to Python to support the sub-classing features described in <a href="#3">[3]</a>. These changes were not documented until the fourth Python Workshop <a href="#4">[4]</a>. At the third Python workshop, I presented some work I had done on generating module documentation for extension types. Based on the discussion at this workshop, I developed a meta-type proposal <a href="#5">[5]</a>. This meta-type proposal was for an object that simply stored meta-information for a type, for the purpose of generating module documentation. In the summer of 1996, Don Beaudry released the system described in <a href="#3">[3]</a> under the name MESS <a href="#6">[6]</a>. MESS addresses a number of needs but has a few drawbacks: <ul><li>Only single inheritance is supported. <li>The mechanisms for defining MESS extension types is very different from and more complicated than the standard Python type creation mechanism. <li>Defining MESS types requires the use of an extensive C applications programming interface. This presents problems for configuring dynamically-loaded extension modules unless the MESS library is linked into the Python interpreter. <li>Because the system tries to do a number of different things, it is fairly large, about 15,000 lines. <li>There is very little documentation, especially for the C programming interface. <li>The system is a work in progress, with a number of outstanding bugs. </ul> As MESS matures, we expect most of these problems to be addressed. <h2>Extension Classes</h2> To meet short term needs for a C-based persistence mechanism <a href="#1">[1]</a>, an extension class module was developed using the mechanism described in <a href="#4">[4]</a> and building on ideas from MESS <a href="#6">[6]</a>. The extension class module recasts extension types as "extension classes" by seeking to eliminate, or at least reduce semantic differences between types and classes. The module was designed to meet the following goal: <ul><li>Provide class-like behavior for extension types, including interfaces for meta information and for constructing instances. <li>Support sub-classing in Python from extension classes, with support for multiple inheritance. <li>Provide a small hardened implementation that can be used for current products. <li>Provide a mechanism that requires minimal modification to existing extension types. <li>Provide a basis for research on alternative semantics for classes and inheritance. </ul> Note: I use non-standard terminology here. By standard python terminology, only standard python classes can be called classes. ExtensionClass "classes" are technically just "types" that happen to swim, walk and quack like python classes. <h3>Base extension classes and extension subclasses</h3> Base extension classes are implemented in C. Extension subclasses are implemented in Python and inherit, directly or indirectly from one or more base extension classes. An extension subclass may inherit from base extension classes, extension subclasses, and ordinary python classes. The usual inheritance order rules apply. Currently, extension subclasses must conform to the following two rules: <ul><li>The first super class listed in the class statement defining an extension subclass must be either a base extension class or an extension subclass. This restriction will be removed in Python-1.5. <li>At most one base extension direct or indirect super class may define C data members. If an extension subclass inherits from multiple base extension classes, then all but one must be mix-in classes that provide extension methods but no data. </ul> <h3>Meta Information</h3> Like standard python classes, extension classes have the following attributes containing meta-data: <dl><dt> <code>__doc__</code> <dd>a documentation string for the class, <dt> <code>__name__</code> <dd>the class name, <dt> <code>__bases__</code><dd>a sequence of base classes, <dt> <code>__dict__</code> <dd>a class dictionary, and <dt> <code>__module__</code><dd>the name of the module in which the class was defined. </dl> The class dictionary provides access to unbound methods and their documentation strings, including extension methods and special methods, such as methods that implement sequence and numeric protocols. Unbound methods can be called with instance first arguments. <h3>Subclass instance data</h3> Extension subclass instances have instance dictionaries, just like Python class instances do. When fetching attribute values, extension class instances will first try to obtain data from the base extension class data structure, then from the instance dictionary, then from the class dictionary, and finally from base classes. When setting attributes, extension classes first attempt to use extension base class attribute setting operations, and if these fail, then data are placed in the instance dictionary. <h2>Implementing base extension classes</h2> A base extension class is implemented in much the same way that an extension type is implemented, except: <ul><li>The include file, <code>ExtensionClass.h</code>, must be included. <li>The type structure is declared to be of type <code>PyExtensionClass</code>, rather than of type <code>PyTypeObject</code>. <li>The type structure has an additional member that must be defined after the documentation string. This extra member is a method chain (<code>PyMethodChain</code>) containing a linked list of method definition (<code>PyMethodDef</code>) lists. Method chains can be used to implement method inheritance in C. Most extensions don't use method chains, but simply define method lists, which are null-terminated arrays of method definitions. A macro, <code>METHOD_CHAIN</code> is defined in <code>ExtensionClass.h</code> that converts a method list to a method chain. (See the example below.) <li>Module functions that create new instances must be replaced by <code>__init__</code> methods that initialize, but does not create storage for instances. <li>The extension class must be initialized and exported to the module with:: <PRE> PyExtensionClass_Export(d,"name",type); where 'name' is the module name and 'type' is the extension class type object. </PRE> </ul> <h3>Attribute lookup</h3> Attribute lookup is performed by calling the base extension class <code>getattr</code> operation for the base extension class that includes C data, or for the first base extension class, if none of the base extension classes include C data. <code>ExtensionClass.h</code> defines a macro <code>Py_FindAttrString</code> that can be used to find an object's attributes that are stored in the object's instance dictionary or in the object's class or base classes: <PRE> v = Py_FindAttrString(self,name); </PRE> where <code>name</code> is a C string containing the attribute name. In addition, a macro is provided that replaces <code>Py_FindMethod</code> calls with logic to perform the same sort of lookup that is provided by <code>Py_FindAttrString</code>. If an attribute name is contained in a Python string object, rather than a C string object, then the macro <code>Py_FindAttr</code> should be used to look up an attribute value. <h3>Linking</h3> The extension class mechanism was designed to be useful with dynamically linked extension modules. Modules that implement extension classes do not have to be linked against an extension class library. The macro <code>PyExtensionClass_Export</code> imports the <code>ExtensionClass</code> module and uses objects imported from this module to initialize an extension class with necessary behavior. <h3>Example: MultiMapping objects</h3> An <a href="MultiMapping.html">example</a> is provided that illustrates the changes needed to convert an existing type to an ExtensionClass. <h2>Implementing base extension class constructors</h2> Some care should be taken when implementing or overriding base class constructors. When a Python class overrides a base class constructor and fails to call the base class constructor, a program using the class may fail, but it will not crash the interpreter. On the other hand, an extension subclass that overrides a constructor in an extension base class must call the extension base class constructor or risk crashing the interpreter. This is because the base class constructor may set C pointers that, if not set properly, will cause the interpreter to crash when accessed. This is the case with the <code>MultiMapping</code> extension base class shown in the example above. If no base class constructor is provided, extension class instance memory will be initialized to 0. It is a good idea to design extension base classes so that instance methods check for uninitialized memory and perform initialialization if necessary. This was not done above to simplify the example. <h2>Overriding methods inherited from Python base classes</h2> A problem occurs when trying to overide methods inherited from Python base classes. Consider the following example: <PRE> from ExtensionClass import Base class Spam: def __init__(self, name): self.name=name class ECSpam(Base, Spam): def __init__(self, name, favorite_color): Spam.__init__(self,name) self.favorite_color=favorite_color </PRE> This implementation will fail when an <code>ECSpam</code> object is instantiated. The problem is that <code>ECSpam.__init__</code> calls <code>Spam.__init__</code>, and <code>Spam.__init__</code> can only be called with a Python instance (an object of type <code>"instance"</code>) as the first argument. The first argument passed to <code>Spam.__init__</code> will be an <code>ECSpam</code> instance (an object of type <code>ECSPam</code>). To overcome this problem, extension classes provide a class method <code>inheritedAttribute</code> that can be used to obtain an inherited attribute that is suitable for calling with an extension class instance. Using the <code>inheritedAttribute</code> method, the above example can be rewritten as: <PRE> from ExtensionClass import Base class Spam: def __init__(self, name): self.name=name class ECSpam(Base, Spam): def __init__(self, name, favorite_color): ECSpam.inheritedAttribute('__init__')(self,name) self.favorite_color=favorite_color </PRE> This isn't as pretty but does provide the desired result. <h2>New class and instance semantics</h2> <h3>Context Wrapping</h3> It is sometimes useful to be able to wrap up an object together with a containing object. I call this "context wrapping" because an object is accessed in the context of the object it is accessed through. We have found many applications for this, including: <ul><li>User-defined method objects, <li><a href="Acquisition.html">Acquisition</a> and <li>Computed attributes </ul> <h4>User-defined method objects</h4> Python classes wrap Python function attributes into methods. When a class has a function attribute that is accessed as an instance attribute, a method object is created and returned that contains references to the original function and instance. When the method is called, the original function is called with the instance as the first argument followed by any arguments passed to the method. Extension classes provide a similar mechanism for attributes that are Python functions or inherited extension functions. In addition, if an extension class attribute is an instance of an extension class that defines an <code>__of__</code> method, then when the attribute is accessed through an instance, it's <code>__of__</code> method will be called to create a bound method. Consider the following example: <PRE> import ExtensionClass class CustomMethod(ExtensionClass.Base): def __call__(self,ob): print 'a %s was called' % ob.__class__.__name__ class wrapper: def __init__(self,m,o): self.meth, self.ob=m,o def __call__(self): self.meth(self.ob) def __of__(self,o): return self.wrapper(self,o) class bar(ExtensionClass.Base): hi=CustomMethod() x=bar() hi=x.hi() </PRE> Note that <code>ExtensionClass.Base</code> is a base extension class that provides very basic ExtensionClass behavior. When run, this program outputs: <code>a bar was called</code>. <h4>Computed Attributes</h4> It is not uncommon to wish to expose information via the attribute interface without affecting implementation data structures. One can use a custom <code>__getattr__</code> method to implement computed attributes, however, this can be a bit cumbersome and can interfere with other uses of <code>__getattr__</code>, such as for persistence. The <code>__of__</code> protocol provides a convenient way to implement computed attributes. First, we define a ComputedAttribute class. a ComputedAttribute is constructed with a function to be used to compute an attribute, and calls the function when it's <code>__of__</code> method is called: import ExtensionClass class ComputedAttribute(ExtensionClass.Base): def __init__(self, func): self.func=func def __of__(self, parent): return self.func(parent) Then we can use this class to create computed attributes. In the example below, we create a computed attribute, <code>radius</code>: from math import sqrt class Point(ExtensionClass.Base): def __init__(self, x, y): self.x, self.y = x, y radius=ComputedAttribute(lambda self: sqrt(self.x**2+self.y**2)) which we can use just like an ordinary attribute: p=Point(2,2) print p.radius <h3>Overriding method calls</h3> Normally, when a method is called, the function wrapped by the method is called directly by the method. In some cases, it is useful for user-defined logic to participate in the actual function call. Extension classes introduce a new protocol that provides extension classes greater control over how their methods are called. If an extension class defines a special method, <code>__call_method__</code>, then this method will be called to call the functions (or other callable object) wrapped by the method. The method. <code>__call_method__</code> should provide the same interface as provided by the Python builtin <code>apply</code> function. For example, consider the expression: <code>x.meth(arg1, arg2)</code>. The expression is evaluated by first computing a method object that wraps <code>x</code> and the attribute of <code>x</code> stored under the name <code>meth</code>. Assuming that <code>x</code> has a <code>__call_method__</code> method defined, then the <code>__call_method__</code> method of <code>x</code> will be called with two arguments, the attribute of <code>x</code> stored under the name <code>meth</code>, and a tuple containing <code>x</code>, <code>arg1</code>, and <code>arg2</code>. To see how this feature may be used, see the Python module, <code>Syn.py</code>, which is included in the ExtensionClass distribution. This module provides a mix-in class that provides Java-like "synchonized" classes that limit access to their methods to one thread at a time. An interesting application of this mechanism would be to implement interface checking on method calls. <h3>Method attributes</h3> Methods of ExtensionClass instances can have user-defined attributes, which are stored in their associated instances. For example: <PRE> class C(ExtensionClass.Base): def get_secret(self): "Get a secret" .... c=C() c.f.__roles__=['Trusted People'] print c.f.__roles__ # outputs ['Trusted People'] print c.f__roles__ # outputs ['Trusted People'] print C.f.__roles__ # fails, unbound method </PRE> A bound method attribute is set by setting an attribute in it's instance with a name consisting of the concatination of the method's <code>__name__</code> attribute and the attribute name. Attributes cannot be set on unbound methods. <h3>Class initialization</h3> Normal Python class initialization is similar to but subtley different from instance initialization. An instance <code>__init__</code> function is called on an instance immediately after it is created. An instance <code>__init__</code> function can use instance information, like it's class and can pass the instance to other functions. On the other hand, the code in class statements is executed immediately before the class is created. This means that the code in a class statement cannot use class attributes, like <code>__bases__</code>, or pass the class to functions. Extension classes provide a mechanism for specifying code to be run after a class has been created. If a class or one of it's base classes defines a <code>__class_init__</code> method, then this method will be called just after a class has been created. The one argument passed to the method will be the class, not an instance of the class. <h2>Useful macros defined in ExtensionClass.h</h2> A number of useful macros are defined in ExtensionClass.h. These are documented in <code>ExtensionClass.h</code>. <h2>Pickleability</h2> Classes created with ExtensionClass, including extension base classes are automatically pickleable. The usual gymnastics necessary to pickle <code>non-standard</code> types are not necessray for types that have been modified to be extension base classes. <h2>Status</h2> The current release of the extension class module is <a href="http://www.digicool.com/releases/ExtensionClass/ExtensionClass-1.1.tar.gz">1.1</a>. The core implementation has less than four thousand lines of code, including comments. This release requires Python 1.4 or higher. To find out what's changed in this release, see the <a href="release.html">release notes</a>. <a href="Installation.html">Installation instructions</a> are provided. <h2>Issues</h2> There are a number of issues that came up in the course of this work and that deserve mention. <ul><li>In Python 1.4, the class extension mechanism described in <a href="#4">[4]</a> required that the first superclass in a list of super-classes must be of the extended class type. This may not be convenient if mix-in behavior is desired. If a list of base classes starts with a standard python class, but includes an extension class, then an error was raised. It would be more useful if, when a list of base classes contains one or more objects that are not python classes, the first such object was used to control the extended class definition. To get around this, the <code>ExtensionClass</code> module exports a base extension class, <code>Base</code>, that can be used as the first base class in a list of base classes to assure that an extension subclass is created. Python 1.5 allows the class extension even if the first non-class object in the list of base classes is not the first object in the list. This issue appears to go away in Python 1.5, however, the restriction that the first non-class object in a list of base classes must be the first in the list may reappear in later versions of Python. <li>Currently, only one base extension class can define any data in C. The data layout of subclasses-instances is the same as for the base class that defines data in C, except that the data structure is extended to hold an instance dictionary. The data structure begins with a standard python header, and extension methods expect the C instance data to occur immediately after the object header. If two or more base classes defined C data, the methods for the different base classes would expect their data to be in the same location. A solution might be to allocate base class instances and store pointers to these instances in the subclass data structure. The method binding mechanism would have to be a more complicated to make sure that methods were bound to the correct base data structure. Alternatively, the signature of C methods could be expanded to allow pointers to expected class data to be passed in addition to object pointers. <li>There is currently no support for sub-classing in C, beyond that provided by method chains. <li>Rules for mixed-type arithmetic are different for python class instances than they are for extension type instances. Python classes can define right and left versions of numeric binary operators, or they can define a coercion operator for converting binary operator operands to a common type. For extension types, only the latter, coercion-based, approach is supported. The coercion-based approach does not work well for many data types for which coercion rules depend on the operator. Because extension classes are based on extension types, they are currently limited to the coercion-based approach. It should be possible to extend the extension class implementation to allow both types of mixed-type arithmetic control. <li>I considered making extension classes immutable, meaning that class attributes could not be set after class creation. I also considered making extension subclasses cache inherited attributes. Both of these are related and attractive for some applications, however, I decided that it would be better to retain standard class instance semantics and provide these features as options at a later time. <li>The extension class module defines new method types to bind C and python methods to extension class instances. It would be useful for these method objects to provide access to function call information, such as the number and names of arguments and the number of defaults, by parsing extension function documentation strings. </ul> <h2>Applications</h2> Aside from test and demonstration applications, the extension class mechanism has been used to provide an extension-based implementation of the persistence mechanism described in <a href="#1">[1]</a>. We have developed this further to provide features such as automatic deactivation of objects not used after some period of time and to provide more efficient persistent-object cache management. Acquisition has been heavily used in our recent products. Synchonized classes have also been used in recent products. <h2>Summary</h2> The extension-class mechanism described here provides a way to add class services to extension types. It allows: <ul><li>Sub-classing extension classes in Python, <li>Construction of extension class instances by calling extension classes, <li>Extension classes to provide meta-data, such as unbound methods and their documentation string. </ul> In addition, the extension class module provides a relatively concise example of the use of mechanisms that were added to Python to support MESS <a href="#6">[6]</a>, and that were described at the fourth Python Workshop <a href="#4">[4]</a>. It is hoped that this will spur research in improved and specialized models for class implementation in Python. References <a name="1">[1]</a> Fulton, J., <a href="http://www.digicool.com/papers/Persistence.html">Providing Persistence for World-Wide-Web Applications</a> Proceedings of the 5th Python Workshop. <a name="2">[2]</a> Page, R. and Cropper, S., <a href="http://www.digicool.com/papers/DocumentTemplate.html">Document Template</a> Proceedings of the 5th Python Workshop. <a name="3">[3]</a> Beaudry, D., <a href="http://www.python.org/workshops/1994-11/BuiltInClasses/BuiltInClasses_1.html">Deriving Built-In Classes in Python</a> Proceedings of the First International Python Workshop. <a name="4">[4]</a> Van Rossum, G., <a href="http://www.python.org/workshops/1996-06/notes/thursday.html">Don Beaudry Hack - MESS</a> presented in the Developer's Future Enhancements session of the 4th Python Workshop. <a name="5">[5]</a> Fulton, J., <a href="http://www.digicool.com/jim/MetaType.c">Meta-Type Object</a> This is a small proposal, the text of which is contained in a sample implementation source file, <a name="6">[6]</a> Beaudry, D., and Ascher, D., "The Meta-Extension Set", http://starship.skyport.net/~da/mess/.