[Subversion] / PEAK / CHANGES.txt  

Diff of /PEAK/CHANGES.txt

Parent Directory | Revision Log

version 1104, Sat May 10 18:23:04 2003 UTC version 1803, Thu Sep 23 21:07:44 2004 UTC
Line 1 
Line 1 
 Fixes and Enhancements since Version 0.5 alpha 1  Fixes and Enhancements since Version 0.5 alpha 3
   
  Changed, Enhanced, or Newly Deprecated Features   - Fixed several 'peak.events' bugs, as reported by Vladimir Iliev, Yaroslav
      Samchuk, and Alexander Smishlajev:
   
  - DEPRECATED use of '__implements__' and '__class_implements__' to declare     * 'events.AnyOf' could hold multiple references to a single event source,
    support for interfaces.  Use 'implements()', 'classProvides()',       and nesting 'AnyOf()' calls could leak references to the nested events.
    'directlyProvides()', 'moduleProvides()', etc. to do this now; they  
    are now available automatically from 'peak.api'.  
   
  - The signatures of the 'getObjectInstance()', 'getStateToBind()', and  
    'getURLContext()' methods in the 'peak.naming' package have changed, to  
    place the context or parent component as the first, non-optional argument.  
    (If you don't know what these methods are for, you don't need to do anything  
    about this, as they are part of the naming package's extensibility  
    framework.)  
   
  - 'EigenRegistry' and 'PropertyMap' no longer attempt to figure out whether  
    implied (i.e. inherited) interfaces are more or less general with respect  
    to a previous registration.  This was behavior that emulated Zope adapter  
    registries, but what we really wanted was more akin to a Zope "type"  
    registry.  The only parts of the test suite that used the old behavior  
    were the tests specifically written to ensure that behavior!  
   
  - 'binding.bindTo()' now accepts a 'default=' argument, whose value will be  
    used in case of a 'NameNotFound' error.  
   
  - DEPRECATED 'naming.ParsedURL'.  It will disappear in 0.5 alpha 3 or beta.  
    It is replaced by the new 'naming.URL.Base'.  The 'naming.URL' package  
    provides a new URL parsing framework based on 'peak.model'.  Upgrading from  
    'ParsedURL' to 'URL.Base' is trivial for ParsedURL subclasses that used  
    only the 'scheme' and 'body' fields, and in fact may not require any  
    changes except for the choice of base class.  
   
    For more complex URL classes, the '__init__' methods go away, 'parse'  
    methods change slightly, and explicit field definitions (using  
    'model.structField' or similar) are required.  See PEAK's 'URL.Base'  
    subclasses for examples.  There is also a sophisticated parsing and  
    formatting framework (see the 'peak.naming.URL' and 'peak.util.fmtparse'  
    modules) that can be used in place of the old regex-based approach.  
   
  - Added 'peak.util.fmtparse', a parsing and formatting framework, and  
    integrated it with 'peak.model' so that any element type can have a  
    syntax for parsing from, or formatting to, a string.  
   
  - Added 'binding.whenAssembled(...)' as syntax sugar for  
    'binding.Once(...,activateUponAssembly=True)'.  
   
  - Removed 'LOG_XYZ' convenience functions from 'peak.api', and refactored  
    'peak.running.logs' to use a PEP 282-like interface, 'running.ILogger'.  
    Under the new scheme, messages must be sent to a specific entry point  
    (e.g. 'self.logger.warning("foo")').  Components can bind an attribute  
    directly to a logger object, or via configuration properties or utilities.  
    PEAK components that do logging all define a 'logger' attribute, bound  
    to a configuration property in the 'peak.logs' property namespace.  By  
    a default in 'peak.ini', 'peak.logs.*' is configured to output messages  
    of 'WARNING' priority or higher to 'sys.stderr'.  
   
    For compatibility with the PEP 282 logging package, a 'logging.logger:'  
    URL scheme has been added; looking up the URL '"logging.logger:foo.bar"'  
    is equivalent to 'logging.getLogger("foo.bar")', unless the 'logging'  
    package is not available, in which case the configuration property  
    'peak.logs.foo.bar' will be looked up in the target context of the  
    lookup.  Optionally, you can configure the 'logging.logger' URL scheme so  
    that it only uses PEAK loggers, and never uses the PEP 282 loggers.  
   
  - Added 'binding.metamethod()' wrapper for metaclass methods that might  
    not be accessible from their instances if the instances (classes) also  
    defined the method for *their* instances.  You must now use this wrapper  
    on any such metaclass-defined methods, as PEAK no longer works around  
    this via the 'x.__class__.foo(x,...)' trick that was used previously.  
    In particular, if you have metaclass definitions of 'getParentComponent',  
    '_getConfigData', 'getComponentName', or 'notifyUponAssembly', you need  
    to wrap them with 'binding.metamethod' now.  
   
  - Made 'NOT_GIVEN' and 'NOT_FOUND' recognizable by humans (they 'repr'  
    and 'str' to their names) and by Python (they can be pickled, and  
    when restored they come back as the same object).  
   
   
  Corrected Problems  
   
  - Fixed 'naming.lookup()' and related APIs not setting the parent component  
    of created objects without an explicitly supplied 'creationParent' keyword  
    argument.  This used to "sort of work" when we had implicit configuration  
    parents, but was broken when we went "all explicit" for 0.5 alpha 1.  
   
  - Fixed a problem where initializing single-valued immutable fields of  
   'peak.model' types did not perform type/value normalization.  
   
  - Fixed a problem where bindTo would use the attribute name as the  
    default value for a lookup, if the requested name/property/utility  
    was not found.  
   
  - Fixed 'mof2py' generator script not working  
   
  - Fixed model.Element not getting parent component set when passed as a     * 'events.subscribe()' had a potential race condition wherein a callback
    constructor argument.       could be invoked after its weak reference was garbage collected, leading
        to bizarre error messages about 'self' being 'None'.
   
      * 'select()' could be called on select event objects even if there were
        no current subscribers to the event, potentially leading to calling
        'select()' on a closed socket.
   
      * Non-default signal handlers were remaining installed even when there
        were no current subscribers to the applicable event, as long as a
        reference to the event object existed.
   
      As a result of these changes, certain I/O event types (esp. signals and
      stream readable/writeable events) are now longer-lived.  For example,
      signal event objects are now immortal, and the read/write event for a
      particular 'fileno()' will be reused for as long as its supplying
      'Selector' or 'EventLoop' instance exists.  (Previously, weak references
      were used so that these objects would be recycled when not in use.)
   
    - Added 'config.registeredProtocol()' API, that supports defining named and
      local protocols.  This allows easy emulation of Zope 3's "named" and "local"
      adapters and views.
   
    - 'binding.Component' objects no longer support instance configuration at
      runtime (i.e., they no longer implement 'config.IConfigurable').  If you
      need a component to be configurable at runtime, you must now derive from
      (or mix in) 'binding.Configurable' instead.  If you get errors about
      a missing 'registerProvider' attribute, or about being unable to adapt to
      'IConfigurable', try changing your base class from 'binding.Component'
      to 'binding.Configurable', or add it as a mixin if you're deriving from
      a class that uses 'binding.Component' as its base.
   
    - 'binding.IComponent' no longer derives from 'config.IConfigurable' or
      'config.IConfigMap', only 'config.IConfigSource'.  This means that
      'IComponent' no longer guarantees or requires the presence of the
      'registerProvider()' method: now only 'config.IConfigurable' does that.
   
    - The 'config.IConfigMap' interface is now DEPRECATED.  Use
      'config.IConfigurable' instead.  The '_configKeysMatching()' method
      of 'IConfigMap' was moved to 'config.IConfigSource', so if you've
      implemented a custom 'IConfigSource', be sure to add this method.
   
    - 'web.ISkinService' and 'web.ILayerService' were consolidated into
      'web.IInteractionPolicy', because the need to have configurable
      implementations of these services is negligible.  That is, the
      corresponding property namespaces ('peak.web.skins' and 'peak.web.layers')
      are more than adequate as registries.
   
    - Removed 'peak.running.timers' and 'peak.util.dispatch'.  Neither was in
      active use, and both are being replaced by the new generic functions
      package in PyProtocols.
   
    - The 'config.iterParents' API is now moved to 'binding.iterParents', and all
      'binding' functions that walk the component hierarchy use it.  It has also
      been changed to avoid infinite loops in the case of a pathological
      component structure.
   
    - The 'persistence' package has been moved to 'peak.persistence' to avoid
      conflicts with ZODB3 and the latest version of Zope 3.  It will eventually
      be phased out, but for now this move is the simplest way to get it out of
      the way.
   
    - The 'peak.util.SOX' module now uses only one parser, based directly on
      'expat', instead of using SAX.  The new parser expects a new node interface,
      'IXMLBuilder', but adapters from the previous interfaces ('ISOXNode' and
      'ISOXNode_NS') are supplied for backward compatibility.  All of PEAK's
      direct XML handling (currently just 'peak.storage.xmi' and
      'peak.web.templates') have been refactored to use the new interface.  Some
      parsing classes (such as 'ObjectMakingHandler', 'NSHandler', and
      'DOMletParser') are no longer available.
   
    - 'peak.web' no longer uses Zope X3 for HTTP publishing support; it has been
      refactored to use a "simpler, more uniform architecture":http://www.eby-sarna.com/pipermail/peak/2004-May/001462.html
      See also "more on the architecture":http://www.eby-sarna.com/pipermail/peak/2004-June/001482.html
      and subsequent posts in that thread.
   
      As a consequence, "various features have been removed":http://www.eby-sarna.com/pipermail/peak/2004-June/001500.html
      from 'peak.web', for possible return at a future date.  Here is a rough
      outline of the changes made so far:
   
       * The 'pageProtocol', 'pathProtocol', and 'errorProtocol' machinery are
         gone.  They will be replaced in the future with an explicit "controller"
         wrapping mechanism to allow application-specific renderings of the same
         underlying components.
   
       * The Zope 'request' and 'response' objects are gone, along with all of
         their special handling for cookies, character sets, form variables,
         automatically marshalling parameters to functions, etc.  These items of
         functionality will be gradually replaced by functions in 'peak.web.api'.
   
         As a result of this, arbitrary functions and methods can no longer be
         used as web pages; instead, functions and methods to be published must
         use the same inputs and outputs as the 'IHTTPHandler.handle_http()'
         method.
   
       * The 'IWebPage', 'IWebInteraction', 'ITraversalContext', 'Traversal',
         'TraversalContext', and 'Interaction' interfaces and classes no longer
         exist, as they are unneeded in the new architecture.  Instead of
         having a central 'IWebInteraction' that's referenced by numerous
         'ITraversalContext' objects, the new approach uses an 'environ' mapping
         for most functions.  For access control, a 'security.IInteraction' is
         now used, whose function is limited to security checks.  Most
         functions previously performed by 'IWebInteraction' have moved to
         'IInteractionPolicy' or to 'peak.web.api' functions operating on
         'environ' mappings.
   
       * Web exceptions can define a 'levelName' attribute that determines the
         severity level with which the exception will be logged.  This allows
         one to e.g. avoid logging tracebacks for 'NotFound' errors.
   
       * Various interface calling signatures have changed slightly.  For example,
         'IAuthService.getUser()' now accepts an 'environ' mapping instead of
         an interaction.  'IInteractionPolicy.newInteraction()' now takes keyword
         arguments, but not a 'request'.  The 'IWebTraversable' interface no longer
         has a 'getObject()' method, and the 'IWebException.handleException()'
         method signature has changed as well.  Finally, all methods that
         previously accepted 'ITraversalContext' (such as
         'IDOMletState.renderFor()') now expect 'environ' mappings.
   
       * 'web.TestInteraction' was replaced with 'web.TestPolicy', and
         'web.Interaction' was removed, since 'IWebInteraction' is no longer part
         of the architecture.
   
    - The 'log()' method of PEAK loggers ('logs.ILogger') now accepts a level name
      *or* a number, for convenient invocation.
   
    - SQL transaction semantics have changed.  Now, issuing an SQL statement
      *always* causes the connection to join the active PEAK transaction, even if
      you request that the SQL be issued "outside" a transaction.  Such SQL will
      be issued outside of the *database* transaction, but not outside of the
      PEAK transaction.  This simplifies the overall processing model for dealing
      with "untransacted" SQL such as Sybase DDL or read-only Oracle transactions.
      (In particular, the requirement that triggered this change was to allow
      Oracle read-only transactions to be released at the end of the current PEAK
      transaction.)  Also, got rid of the now-meaningless 'begin' command in n2.
   
    - The 'events.IEventSource' interface now returns a 'canceller' function from
      the 'addCallback()' method, allowing you to cancel a previously-scheduled
      callback.  This fixes a memory leak and performance problem with
      'events.AnyOf()', which previously could accumulate unneeded callbacks on
      the sources it was monitoring.  Note that if you have developed any custom
      event sources with 'addCallback()' methods, you must make sure that they
      return a canceller from now on.
   
    - Added 'ref:factory@addr1||addr2' URL scheme that maps to a corresponding
      'naming.Reference("factory",["addr1","addr2"])'.  'factory' can be either a
      dotted import string referencing a 'naming.IObjectFactory', or you can
      define a factory in the 'peak.naming.factories' property space.
   
    - Added a 'zconfig.schema' factory, so that 'ref:zconfig.schema@streamURL'
      will load a schema loader.  Schema loaders are themselves object factories,
      so you can do something like:
   
        [Named Services]
        peak.naming.factories.myschema = \
            naming.LinkRef('ref:zconfig.schema@pkgfile:mypkg/Schema.xml')
   
      in order to make URLs like 'ref:myschema@filename' work.  Note, by the way,
      that the above could also read:
   
        [Named Services]
        peak.naming.factories.myschema = \
            naming.Reference('zconfig.schema',['pkgfile:mypkg/Schema.xml'])
   
      which runs somewhat faster at lookup time.  Similarly, one can also use
      'naming.Reference("myschema",["somefile"])' in place of a
      'naming.LinkRef("ref:myschema@filename")'.  As well as being faster, for
      some use cases it's easier to 'Reference' directly than to glue together
      a 'ref:' URL string.
   
  - Fixed property/utility lookups not working correctly on model.*  
    objects.  
   
  - Fixed IndentedStream generating all-whitespace lines  
   


Generate output suitable for use with a patch program
Legend:
Removed from v.1104  
changed lines
  Added in v.1803

cvs-admin@eby-sarna.com

Powered by ViewCVS 1.0-dev

ViewCVS and CVS Help