abdera-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From James Snell <jasn...@gmail.com>
Subject Abdera2 Change Summary
Date Wed, 21 Sep 2011 15:54:24 GMT
Ok, so I've checked in the initial run at a proposed Abdera 2.0...
there are quite a few changes in this package so I thought it would be
good to summarize things for those of you checking things out.

1. First off, all of the dependencies have been updated to the latest,
greatest versions. In several cases (such as the upgrade to Apache
HTTP Client 4.x) this meant making major changes to the code. In fact,
the entire Client code component has been overhauled and improved. The
basic API will still be familiar to those familiar with Abdera 1.x
tho.

2. Second, the java package layout has been rearranged significantly.
Because of the many non-backwards compatible changes that have been
introduced, the base java package for abdera 2.0 is now
org.apache.abdera2.*. With the exception of possible differences in
configuration files, both the 1.x and 2.x versions can sit within the
same classpath without conflicting.

3. Many of the "common" utility classes have been updated. For
instance, org.apache.abdera.model.AtomDate has been moved to
org.apache.abdera2.common.date.DateTime and has been expanded to
support ISO8601 Duration and Intervals as well. The URI Template
library has been overhauled to support the latest version of the URI
Templates specification. The Lang tag implementation has been
simplified. Language tag validation code has been removed. The
language tag registry has grown significantly and the code just wasn't
keeping pace with it so it was better, and more scalable just to
remove the validation code. The new org.apache.abdera2.common.http.*
package contains a number of utilities for working directly with
various HTTP Header values.. these include EntityTags, Q-value
qualified tokens (e.g. for content negotiation), custom Authentication
headers, cache controls and web links
(http://tools.ietf.org/html/rfc5988).

4. A very important change is that the NamedWriter and NamedParser
mechanism has been revamped. In fact, the NamedWriter and NamedParser
interfaces have been dropped completely and replaced with just Writer
and Parser implementations that use the @Name annotation from
org.apache.abdera2.common.anno.* package. The functionality for
supporting named writers and parsers is still there, but if you've
implemented a NamedWriter and NamedParser, then you'll have to port
those over when moving to Abdera2

5. There is now a DefaultImplementation annotation that can be used
with the classpath discovery mechanism for specifying the default
implementation of an interface.

6. There is a new Selector interface that provides a mechanism of
filtering requested lists. New methods have been added to the Feed
Object Model API to support the use of selectors... For instance,
entry.getLinks(new MySelector()) ... the Selector interface can
greatly simplify the code necessary for iterating through a list of
elements and can improve overall performance since the FOM method
implementations will apply the selector internally when it is building
the list to return, avoiding the need to iterate through the list
multiple times.

7. There is an experimental new Pusher interface. This is essentially
a pub/sub style interface for pushing objects (e.g. atom entries or
activities) to a collection of listeners. It can, for instance, be
used to implement a long-polling async servlet based on the new
Servlet 3.0 asynchronous request support. An experimental long polling
servlet has been included in the
org.apache.abdera2.common.protocol.servlet.async package. The default
implementation of the Pusher interface is based on a
ConcurrentLinkedQueue and mechanisms from the java.util.concurrent.*
package, but alternative implementations can be plugged in using the
usual Abdera classpath discovery model. You could, for instance, back
it up with a message queue for a more enterprisey implementation
approach.

8. The Unicode stuff has been completely replaced with ICU4J. This
should have happened a long time ago but I'd never gotten around to
it.

9. An implementation of http://tools.ietf.org/html/rfc5987 has been
added to the org.apache.abdera2.common.test.Codec class and has been
made the default encoding for HTTP Headers that include non-ascii
characters. Q and B Codec support is still included as a fallback.

10. ExtensionFactory implementations that extend from
AbstractExtensionFactory have been simplified using Annotations.
Whereas before you had to call a number of methods in the constructor
to register the namespace and extension implementation classes handled
by the ExtensionFactory, now you can use a number of simple
annotations. For instance,

 @Namespace({"http://example.org/foo/ns"})
 @Impls({@Impl(Foo.class)})
 public final class MyExtensionFactory extends AbstractExtensionFactory {}

 @QName(value="foo", ns="http://example.org/foo/ns")
 public class Foo extends ElementWrapper { ... }

And that's it... configuration of ExtensionFactory impls using
classpath discovery remains unchanged.

11. The ExtensionFactoryMap implementation has been improved. Before,
when an extension needed to be created, it would iterate through ALL
of the extension factories until it found one that could create the
extension, regardless of whether the extensionfactory was capable of
handling extensions of that particular namespace. This caused a
significant performance lag and really was a bug that should have been
fixed a long time ago. Now it checks first to see if the
extensionfactory is capable of handling the namespace before it
attempts to create the element.

12. Performance in the FOMFactory implementation has been greatly
improved by building an internal hashmap of QName to Constructor
pairs. That is, whenever the factory needed to create an element, it
would go through and check the qname of the element being created
against each of the known core QNames. Each one of these was an
equals(...) check that would be performed sequentially every time,
causing poor linear performance as we were parsing. Obviously, when
dealing with large amounts of elements, it was enough to really slow
things down. So instead, when the FOMFactory class is created, a
static mapping of QNames to the appropriate Constructor methods on
their respective classes is created and cached in memory. When a
request to create an element instance of received, we simply need to
do a constant time lookup of the appropriate constructor, call
newInstance(...) on it and off we go. The performance improvement is
significant.

13. the ListParseFilter has been changed to a SetParseFilter... using
Set internally for that just made more sense and generally improves
performance.

14. The server code has undergone a major overhaul, primarily to
support the addition of the Activity Streams component that I'll talk
about in a minute. Basically, in Abdera 1.x, the server assumed the
use of the Atom format throughout every layer. The Feed Object Model
was baked in to nearly every class. In Abdera2, the core of the Abdera
Server Framework has been abstracted out so as to not depend or assume
any specific data format. This allows us, for instance, to create a
publishing service for JSON Activity Streams that uses the Activity
Streams format as a direct replacement for an Atom Feed. This
essentially gives us an out of the box mechanism for working with
Activity Streams the same way we work with Atompub

15. The major new addition to Abdera2 is the Activities Streams
component. This is based entirely on the JSON Activity Streams
specification and does not include support for the Atom-based
Activities Streams specification (the Atom-based flavor can be added
later within the extensions module if necessary). The activities
module includes support for producing and consuming activity streams,
provides a rich api for interacting with the various Activity Streams
objects, include a Client and Server implementation based on the exact
same model used for Atompub, and provides a number of utilities and
extensions to the core spec. A rich set of examples are included.

16. I'm sure there are other things that I'm missing and haven't
mentioned. Overall, there are a number of API changes throughout that
I will continue to work on documenting. Example documentation will
need to be updated.

Things yet to be done...

1. I did not port any of the existing Adapters or the Spring module
over to the new code. This is primarily because the dependencies for
those have changed significantly as well and I'm just not an expert in
each of those specific components. I'm also not sure which backend
data storage mechanisms we should be primarily targeting right now. It
would be great if we could get some others to start taking a look at
those.

2. The Servlet 3.0 support is still highly experimental. The basic
servlet works fine, but the async servlet stuff needs work and
testing. Right now there's a bad memory leak in the default
implementation that I haven't had a chance to track down yet.

3. Extensive testing and even more testing. I've run through all the
test cases and have them working with a few minor exceptions for the
async servlet stuff, but given how many changes have gone into this,
it needs to be thoroughly put through it's paces.

Mime
View raw message