Tryag File Manager

//proc/self/root/usr/share/doc/mx-2.0.6/Proxy/Doc/mxProxy.html

<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<HTML>
<HEAD>
    <TITLE>mxProxy - Generic Proxy Wrapper Type for Python</TITLE>
    <META HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=iso-8859-1">
    <STYLE TYPE="text/css">
      p { text-align: justify; }
      ul.indent { }
      body { }
    </STYLE>
</HEAD>

<HR NOSHADE WIDTH="100%">
    <H2>mxProxy - Generic Proxy Wrapper Type for Python</H2>

<HR SIZE=1 NOSHADE WIDTH="100%">
    <TABLE WIDTH="100%">
      <TR>
	<TD>
	  <SMALL>
	    <A HREF="#AccessProtocol">Access Protocol</A> :
	    <A HREF="#CleanupProtocol">Cleanup Protocol</A> :
	    <A HREF="#ImplicitAccess">Implicit Access</A> :
	    <A HREF="#WeakRefs">Weak References</A> :
	    <A HREF="#Interface">Interface</A> :
	    <A HREF="#Examples">Examples</A> :
	    <A HREF="#Structure">Structure</A> :
	    <A HREF="#Support">Support</A> :
	    <A HREF="http://www.egenix.com/files/python/eGenix-mx-Extensions.html#Download-mxBASE"><B>Download</B></A> :
	    <A HREF="#Copyright">Copyright &amp; License</A> :
	    <A HREF="#History">History</A> :
	    <A HREF="" TARGET="_top">Home</A>
	</SMALL>
	</TD>
	<TD ALIGN=RIGHT VALIGN=TOP>
	  <SMALL>
	    <FONT COLOR="#FF0000">Version 2.0.3</FONT>
	  </SMALL>
	</TD>
    </TABLE>
    <HR SIZE=1 NOSHADE WIDTH="100%">

<H3>Introduction</H3>

<P>
	  A while ago, I ran into a security problem within Python:
	  there was no apparent way to hide attributes from
	  unauthorized access other than running in an restricted
	  environment and using the standard lib's Bastion object
	  wrapper.

<P>
	  Since I couldn't run the code in restricted mode (e.g. it
	  does assignments to __class__ and __dict__ in a few vital
	  places), the only reasonable way seemed building a new type
	  that hides a few attributes in the C implementation.  These
	  attributes are not accessible from within Python. Of course,
	  a debugger will give you access and some special extension
	  module that knows about the internal structure of the types
	  could too. But for my purpose this level security suffices.

<P>
	  Instead of creating a special Bastion like type I figured
	  that it would be a good idea to provide a more generic and
	  extendable version of a <B>Proxy</B> object. The object
	  itself doesn't provide too much funtionality. It basically
	  hides the wrapped object and provides controlled access to
	  its attributes.

<P>
	  Another subject that comes up every now and then in the
	  Python community is that of <B>weak references</B>,
	  i.e. references to objects that don't prevent them to be
	  garbage collected. Since Python uses a reference count
	  garbage collection scheme, circular references cause memory
	  leakage as soon as the reference to the cycle is dropped
	  (the involved objects are not garbage collected because
	  their reference count never drops to zero).

<P>
	  Proxy objects can use two flavors of references starting
	  with version 0.2: strong references which act just like any
	  other reference you use to the object in Python and weak
	  references. The <CODE>WeakProxy()</CODE> constructor must be
	  used for the latter. Details are described below.

<H4>Access Protocol</H4>

<P>
      To control access to the wrapped objects attributes (these also
      include its methods), the Proxy provides two features:
    <P>
    <OL>

<LI>Access is only granted if the attributes name appears in a
      special <B>interface list</B> of accessible names.
	<P>

<LI>Public access can be further controlled if the object
      provides the methods <CODE><B>__public_getattr__</B></CODE>
      and/or <CODE><B>__public_setattr__</B></CODE>.
	<P>

</OL>

<P>
      The access scheme first checks the interface list. If given,
      only attributes with names appearing in it can be set or
      fetched. All other accesses are cancelled by raising an
      <CODE>AccessError</CODE> exception (which is a subclass of the
      standard <CODE>AttributeError</CODE>).

<P>
      The presence of the interface list also indicates whether
      methods that are to be returned to the requesting object should
      be Proxy-wrapped on-the-fly or not.  Wrapped methods only have
      the callable slot enabled, thus inhibiting access to the
      enclosed object reference.

<P>
      In case an attribute it to be fetched, the Proxy checks the
      availability of a <CODE>__public_getattr__</CODE> method. If
      found, this method is called with the name as parameter and
      whatever this method returns is returned to the requesting
      object. Otherwise, the Proxy uses the standard
      getattr()-functionality to fetch the attribute.

<P>
      Setting attributes is done in a similar way: the Proxy checks
      the availability of a <CODE>__public_setattr__</CODE> method. If
      found, this method is called with parameters name and value.
      Otherwise, the standard setattr() functionality is used.

<H4>Cleanup Protocol</H4>

<P>
      When dealing with large object systems circular references are
      sometimes introduced. Since these reference loops cannot
      automatically be broken by the system, the user has to provide
      means of breaking the circles at the proper places. One such place
      is the object destructor of objects that explicitly contain such
      references.

<P>
      The Proxy object can be used to provide an indirect pointer into
      such a circle of objects that reference each other. When the
      system deletes the Proxy object, its destructor tries to find a
      special method <CODE><B>__cleanup__</B></CODE> in the wrapped
      object and calls it before continuing the destruction. This
      allows the object to break the reference circles, e.g. by
      performing a <CODE>self.__dict__.clear()</CODE> or something
      similar.

<P>
      Errors raised in <CODE>__cleanup__</CODE> calls are
      ignored. Warnings are printed to stderr in case Python is run in
      verbose mode (python -v) and tracebacks printed if run in debug
      mode (python -d; note that printing tracebacks can sometimes
      cause core dumps e.g. during finalization of the interpreter).

<H4>Implicit Access</H4>

<P>
      On many occasions object attributes are not explicitly accessed
      by e.g. using <CODE>object.attribute</CODE>, but indirect
      through builtin functions or the interpreter itself. This poses
      a problem to the Proxy, since there are different ways to reach
      an objects implementation. For Python instances all important
      hooks are reachable via <CODE>getattr()</CODE>, e.g. __len__ and
      __add__. This is different for extension and builtin types: they
      use a system of slots for providing access to their data.

<P>
      As of version 0.2, Proxy objects also allow managing the type
      slot interface for most applications. This means that you can
      transparently use the Proxy to wrap builtin types such as lists
      or tuples and use the Proxy just like you would use the
      referenced builtin object itself.

<P>
      Since Proxies are about access management, you can also restrict
      access to these slots. For simplicity, the names you have to use
      for different slots are exactly those you would define for
      Python classes, e.g. the length slot is named '__len__', the
      sequence and mapping get item slot are grouped under the name
      '__getitem__'. These names have to be explicitly stated in the
      interface list you pass to the Proxy constructor when creating
      the proxy in case you do define an interface list. If you don't
      specify such a list, no direct interface restriction is applied.

<P>
      These type slot names are defined:
    <P>
    <DL>
      <DT>Generic:
      <DD>__call__, __hash__, __str__, __cmp__

<DT>Mapping:
      <DD>__getitem__, __setitem__, __len__

<DT>Sequence:
      <DD>__add__, __repeat__, __getitem__, __getslice__
	__setitem__, __getitem__

<DT>Number: 
      <DD>__add__, __sub__, __mul__, __div__
      __mod__, __divmod__, __pow__, __neg__, __pos__, __abs__, __true__
      __invert__, __lshift__, __rshift__, __and__, __str__, __or__, __int__
      __long__, __float__
	<BR> Note: number coercion does not yet
	work, so most of these are currently useless !

</DL>
    
    <P>
      Omissions currently are: __coerce__, __hex__ and
      __oct__. __repr__ is handled by the Proxy object itself since it
      would otherwise possibly expose address information about the
      underlying object.
	
    <P>
      Proxies also work transparently for Python instances. To achieve
      this, another proxy-like object which is a real Python instance
      is to be put in front of the Proxy object. Access to this
      instance then gets translated into getattr-calls by the
      interpreter, these calls are filtered through the Proxy and the
      wrapped instance object's attribute then gets accessed in the
      usual way. The result is passed back to the requesting object.

<H4>Weak References</H4>

<P>
      Proxy objects work in two modes: keeping a strong or a weak reference
      to the object. The <CODE>Proxy()</CODE> constructor returns a
      Proxy object using a strong reference, the <CODE>WeakProxy()</CODE> constructor one
      using a weak reference.

<P>
      Weak references are called weak because they don't keep the
      object alive by incrementing the reference count on the
      referenced object. Since objects get garbage collected when this
      reference count falls down to 0, a weak reference can become
      invalid at any time. The mxProxy implementation raises a
      <CODE>LostReferenceError</CODE> in case a weak reference to such
      an object is used.

<P>
      This may sound like a pretty flaky feature at first, but the
      main pro argument for these weak references is that you can
      build up circular references without having to fear about them
      not being properly garbage collected. Using strong references
      (which do increment the reference count and thus keep the object
      alive as long as the reference is around) would produce cycles
      in the referencing scheme which the Python garbage collection
      (GC) mechanism cannot automatically break causing the cycle to
      become unreachable from the Python namespaces: a severe memory
      leak.

<P>
      Weak references in mxProxy work by using a global dictionary of
      all objects handled through weak reference Proxies. This
      dictionary is checked prior to every action on a weak Proxy and
      after its deletion. You can also force a check by calling the
      <CODE>checkweakrefs()</CODE> anytime you like, e.g. at regular
      intervals.

<P>
      The dictionary holds a strong reference to the object keeping it
      alive until the next check is done. During the check all handled
      objects are inspected to see if their reference count has gone
      down to 1 (<I>phantom objects</I>: only the dictionary
      references them). If this is the case, all weak proxies are
      marked defunct and the object is removed from the dictionary
      causing it to be garbage collected by the Python GC
      mechanism. All subsequent actions on the weak references to this
      object will then cause a <CODE>LostReferenceError</CODE>
      exception to be raised.

</UL>

<H3>Interface</H3>

<H4>Proxy Constructors</H4>

<P>
	  These constructors are available in the package:

<P><DL>

<DT><CODE><FONT COLOR="#000099">
		Proxy(object,interface=None,passobj=None)</FONT></CODE></DT>

<DD>
	    Returns a new Proxy instance that wraps object.

<P>
	      interface can be given as sequence of strings and/or
	      objects with __name__ attribute or as dictionary with
	      string keys (only the keys are currently used) and tells
	      the Proxy to only allow access to these names. If not
	      given or None, no filtering is done by the Proxy (see
	      above on how access is managed).

<P>
	      passobj can be provided to retrieve the wrapped object
	      from the Proxy at a later point using the
	      <CODE>.proxy_object()</CODE> method.  </DD><P>

<DT><CODE><FONT COLOR="#000099">
		InstanceProxy(object,interface=None,passobj=None)</FONT></CODE></DT>

<DD>
	    Same as above, except that a Python instance wraps the
	    Proxy object.

<P>
	      This makes the Proxy transparent for access to wrapped
	      Python instances, meaning that the Proxy will act as if
	      it were the wrapped object itself (with the added
	      features mentioned above). </DD><P>

<DT><CODE><FONT COLOR="#000099">
		CachingInstanceProxy(object,interface=None,passobj=None)</FONT></CODE></DT>

<DD>
	    Same as InstanceProxy(), except that a read cache is used
	    by the Proxy which caches all queried attributes in the
	    Proxy instance's dictionary.

<P>
	      Note that this may introduce circular references if not
	      used properly. Cached attributes are not looked up in
	      the wrapped instance after the first lookup -- if their
	      value changes, this won't be noticed by objects that
	      access the object through this wrapper.

</DD><P>

<DT><CODE><FONT COLOR="#000099">
		SelectiveCachingInstanceProxy(object,interface=None,passobj=None)</FONT></CODE></DT>

<DD>
	    Same as InstanceProxy(), except that a read cache is used
	    by the Proxy which caches certain queried attributes in
	    the Proxy instance's dictionary depending on their type.

<P>
	      The cached types are defined by the
	      <CODE>.proxy_cacheable_types</CODE> attribute. It
	      defaults to only cache Python methods.

</DD><P>

<DT><CODE><FONT COLOR="#000099">
		MethodCachingProxy(object,interface=None,passobj=None)</FONT></CODE></DT>

<DD>
	    Alias for SelectiveCachingInstanceProxy().
	  </DD><P>

<DT><CODE><FONT COLOR="#000099">
		ReadonlyInstanceProxy(object,interface=None,passobj=None)</FONT></CODE></DT>

<DD>
	    Same as InstanceProxy, except that write access will
	    result in an <CODE>AccessError</CODE> being raised
	    </DD><P>

<DT><CODE><FONT COLOR="#000099">
		ProxyFactory(Class,interface=None)</FONT></CODE></DT>

<DD>
	    Returns a factory object for producing Class instances
	    that are automatically wrapped using Proxy-instances.

<P>
	      interface is passed to the Proxy constructor,
	      pass-objects are not supported.  </DD><P>

<DT><CODE><FONT COLOR="#000099">
		InstanceProxyFactory(Class,interface=None)</FONT></CODE></DT>

<DD>
	    Returns a factory object for producing Class instances
	    that are automatically wrapped using
	    InstanceProxy-instances.</DD><P>

<DT><CODE><FONT COLOR="#000099">
		WeakProxy(object,interface=None,passobj=None)</FONT></CODE></DT>

<DD>
	    Returns a new weak referencing Proxy instance that points
	    to object.

<P>
	      passobj can be provided to retrieve the wrapped object
	      from the Proxy at a later point using the
	      <CODE>.proxy_object()</CODE> method.

<P>
	      For details on weak references and how they work, see
	      the above discussion.
	  
	  </DD><P>

</DL>

</UL>

<H4>Proxy Instance Methods</H4>

<UL CLASS="indent">
      
	<P>
	  A <TT>Proxy</TT> instance <CODE>proxy</CODE> defines these
	  methods in addition to the ones available through restricted
	  access to the wrapped object:

<P><DL>

<DT><CODE><FONT COLOR="#000099">
		proxy_object(passobj)</FONT></CODE></DT>

<DD>
	    Returns the wrapped object provided the given passobj is
	    identical to the one used for creating the Proxy.

<P>
	      Simple equality is not enough -- it has to be the same
	      object.  </DD><P>

<DT><CODE><FONT COLOR="#000099">
		proxy_getattr(name)</FONT></CODE></DT>

<DD>
	    Same as getattr(proxy,name).  </DD><P>

<DT><CODE><FONT COLOR="#000099">
		proxy_setattr(name,value)</FONT></CODE></DT>

<DD>
	    Same as setattr(proxy,name,value). </DD><P>

<DT><CODE><FONT COLOR="#000099">
		proxy_defunct()</FONT></CODE></DT>

<DD>
	    Return 1 iff the referenced object has already been
	    garbage collected. </DD><P>

</DL>

<P>
	  Note that all attribute names starting with
	  '<CODE>proxy_</CODE>' are interpreted as being Proxy
	  attributes and are not passed to the wrapped object. Access
	  to these attributes is <I>not</I> subject to the
	  restrictions explained above.

</UL>

<H4>Proxy Instance Variables</H4>

<P>
	  <TT>Proxy</TT> instances do not provide any instance
	  variables themselves. They do provide restricted access to
	  the variables of the wrapped object though.

<P>
	  Note that all attribute names starting with
	  '<CODE>proxy_</CODE>' are interpreted as being Proxy
	  attributes and are not passed to the wrapped object. Also,
	  access to these attributes is <I>not</I> subject to the
	  restrictions explained above.

</UL>
    
    <H4>Functions</H4>

<P>
	  These functions are available:

<P><DL>

<DT><CODE><FONT COLOR="#000099">
		checkweakrefs()</FONT></CODE></DT>

<DD>
	    Calling this function causes the global dictionary used
	    for managing weak references to be checked for phantom
	    objects. If such objects are found, they are garbage
	    collected during the call and weak referencing Proxies
	    pointing to them are defunct.
	    <P>
	      Weak references cause the objects to stay alive until
	      either a proxy is used on them (causing an exception), a
	      proxy referencing them is deleted or this function is
	      called. To ensure the timely garbage collection of the
	      objects, call this function at regular intervals.

</DD><P>

<DT><CODE><FONT COLOR="#000099">
		initweakrefs()</FONT></CODE></DT>
	  
	  <DD>
	    For internal use only:<BR> Initializes or reinitializes
	    the weak reference implementation.  This first forces a
	    finalization of the previous state if the implementation
	    has already been used and then starts again with a clean
	    weak reference dictionary.

</DD><P>

<DT><CODE><FONT COLOR="#000099">
		finalizeweakrefs()</FONT></CODE></DT>

<DD>
	    For internal use only:<BR> Forces finalization of the weak
	    reference implementation. Subsequent usage of weak
	    references will cause errors to be raised.
	    <P>
	      Calling this function after finalization is not an
	      error.

</DD><P>

</DL>

</UL>
    
    <H4>Constants</H4>

<P>
	  These constants are available:

<P><DL>

<DT><CODE><FONT COLOR="#000099">
		AccessError</FONT></CODE></DT>

<DD>
	    Exception object used for access specific errors that the
	    module raises. It is a subclass of
	    exceptions.AttributeError. </DD><P>

</DL>

</UL>

</UL>
    
    <A NAME="Examples">

<H3>Examples of Use</H3>

<P>
	  Here is a very simple one:

<PRE><FONT COLOR="#000066">import Proxy

class DataRecord:
    a = 2
    b = 3
    # Make read-only:
    def __public_setattr__(self,what,to):
	  raise Proxy.AccessError,'read-only'
    # Cleanup protocol
    def __cleanup__(self):
	  print 'cleaning up',self

o = DataRecord()

# Wrap the instance:
p = Proxy.InstanceProxy(o,('a',))
# Remove o from the accessible system:
del o

print 'Read p.a through Proxy:',p.a

# This will cause an exception, because the object is read-only
p.a = 3

# This will cause an exception, because no access is given to .b
print p.b

# Clear all traces of the provious exceptions (they might contain
# references to p) by issuing another one. Note that not doing
# so will cause the following 'del p' to not destroy the final
# reference to p... (don't ask me why).
1/0

# Deleting the Proxy will also delete the wrapped object, if there
# is no other reference to it in the system. It will invoke
# the __cleanup__ method in that case.
del p

# If you want to have the wrapping done automatically, you can use
# the InstanceProxyFactory:
DataRecord = Proxy.InstanceProxyFactory(DataRecord,('a',))

# This gives the same behaviour...
p = DataRecord()
print p.a
p.a = 3
print p.b

</FONT></PRE>

<P>
	  More examples will eventually appear in the Examples
	  subdirectory of the package.

</UL>

<H3>Package Structure</H3>
    
    <UL CLASS="indent">

<PRE>
[Proxy]
       Doc/
       Examples/
       [mxProxy]
       Proxy.py
	</PRE>

<P>
	  Entries enclosed in brackets are packages (i.e. they are
	  directories that include a <TT>__init__.py</TT> file). Ones
	  without brackets are just simple subdirectories that are not
	  accessible via <CODE>import</CODE>.

<P>
	  The package imports all symbols from the Proxy sub module
	  which in turn imports the extension module, so you only need
	  to '<CODE>import Proxy</CODE>' to start working.

<P>

</UL>

<H3>Support</H3>

<P>
	  eGenix.com is providing commercial support for this
	  package. If you are interested in receiving information
	  about this service please see the <A
	  HREF="http://www.egenix.com/files/python/eGenix-mx-Extensions.html#Support">eGenix.com
	  Support Conditions</A>.

</UL>

<H3>Copyright &amp; License</H3>

<P>
	  &copy; 1998-2000, Copyright by Marc-Andr&eacute; Lemburg;
	  All Rights Reserved.  mailto: <A
	  HREF="mailto:mal@lemburg.com">mal@lemburg.com</A>
	<P>
	  &copy; 2000-2001, Copyright by eGenix.com Software GmbH,
	  Langenfeld, Germany; All Rights Reserved.  mailto: <A
	  HREF="mailto:info@egenix.com">info@egenix.com</A>

<P>
	  This software is covered by the <A
	  HREF="mxLicense.html#Public"><B>eGenix.com Public
	  License Agreement</B></A>. The text of the license is also
	  included as file "LICENSE" in the package's main directory.

<P>
	  <B> By downloading, copying, installing or otherwise using
	  the software, you agree to be bound by the terms and
	  conditions of the eGenix.com Public License
	  Agreement. </B>

</UL>

<H3>History & Future</H3>

<P>Things that still need to be done:

<P><UL>

<LI> Get number coercion to work. This may not be possible
	    in general due to deficiencies in Python internal coercion
	    mechanism.

<P><LI> Provide a callback hook for weak referencing
	    proxies which gets called when the referenced object dies.

<P><LI> Implement getattr-translators that map __XXX__
	    style names to type slot calls.

<P><LI> Make automatic method and function wrapping (and
	    proxy object hiding) optional.

</UL>

<P>There were no significant changes between 2.0.0 and 2.0.3.

<P>Changes from <A HREF="mxProxy-0.2.0.zip">0.2.0</A> to 2.0.0:

<UL>

<LI> Fixed a bug in the finalization of the weak
	    reference implemenation. Thanks to Michael Muller for
	    pointing me to this quirk. Finalization of the weak ref
	    dict previously occurred in an atexit() function, but at
	    this time, the Python interpreter is already gone, so
	    __del__ hooks would cause a fatal error due to a missing
	    thread state. The implementation now uses explicit
	    function APIs to handle finalization at module cleanup
	    time.

<P><LI> Removed __weakref reference to the weak ref dict
	    from the module interface.

<P><LI> Added initweakrefs() and finalizeweakrefs().

<P><LI> AccessError is now a subclass of
	    exceptions.AttributeError to enhance integration with the
	    getattr() protocol.

<P><LI> Exceptions raised during .__cleanup__() are only
	    reported in debug or verbose mode (python -v or -d).

<P><LI> AccessErrors now contain the name of the attribute
	    to which access was denied (in case the name is a string).

<P><LI> <B>Moved</B> the package under a new top-level
	    package 'mx'. It is part of the <I>eGenix.com mx BASE
	    distribution</I>.

</UL>

<P>Changes from <A HREF="mxProxy-0.1.0.zip">0.1.0</A> to 0.2.0:

<UL>

<LI>AccessError is a subclass of
	    exceptions.StandardError.

<P><LI>Fixed a bug where a proxy without interface would
	    not wrap bound and unbound methods.

<P><LI>Added CachingInstanceProxy.

<P><LI>Added weak referencing mechanism.

<P><LI>Added support for nearly all low level type slots.

</UL>

<P>

<P>
	  Version 0.1.0 was the intial release.

<P>

</UL>

<HR WIDTH="100%">
    <CENTER><FONT SIZE=-1>&copy; 1999, Copyright by Marc-Andr&eacute;
    Lemburg; All Rights Reserved. mailto: <A
    HREF="mailto:mal@lemburg.com?subject=mxProxy">mal@lemburg.com</A>
    </FONT></CENTER>

</BODY>
</HTML>