Recent changes to patches

trivial improvment to getFieldsToSerialize

toby cabot — Tue, 29 Jul 2003 17:52:00 -0000

Looking through the code to chase down a different
problem (that probably isn't in Skaringa) I noticed
that getFieldsToSerialize() could be improved a little,
and since it gets called a lot it might be worthwhile.
First, it can use an ArrayList instead of a Vector to
save the synchronization overhead, second it can use a
zero-element array as the parameter to toArray().

fast pipelined transformer

toby cabot — Sun, 27 Apr 2003 23:11:53 -0000

Hi,

Here's some code that implements a SAX-pipelined
stripped down hot-rod version of ObjectTransformer. It
takes an XSLT stylesheet in the constructor and caches
Templates so it only needs to prepare the stylesheet
once and can build Transformers quickly on a per-call
basis. With Saxon it's about 2.5 times faster than
passing strings between Skaringa and XSLT; with Xalan
there's less of a difference.

Not sure how to fit it into Skaringa since it doesn't
explicitly implement ObjectTransformer but I thought
that you might be interested in the code.

Regards,
Toby

allow input of dates with no time part

toby cabot — Thu, 06 Mar 2003 23:22:25 -0000

this patch modifies Skaringa's date parser to allow it
to deserialize dates that have no time component, i.e.

add more xml data types

toby cabot — Fri, 21 Feb 2003 21:51:25 -0000

This patch adds some additional XML data types to make
Skaringa more flexible at deserializing XML files that
it didn't create.

Since Skaringa will never use these types for
serialization I added them to the _deserializers map
directly instead of calling addSerializer(). mk,
please let me know if this is a bad approach.

The patch modifes SerializerRegistry.java and
JavaXMLTestClass.java. There's also a new file called
StrangeTypesObj.java which goes in the
com.skaringa.javaxml.test package.

more descriptive eror messages

toby cabot — Tue, 18 Feb 2003 20:49:21 -0000

We're using Skaringa for a purpose for which it might
not have been intended: deserializing from non-Skaringa
XML into Java objects. We get an input message in some
strange format and then run it through XSLT to get XML
which Skaringa can push into our hand-coded Java
classes. Skaringa works well for this - it's very fast.

This patch gives better diagnostics when the types
specified in the input XML don't match the types in the
Java classes.

support for java.util.Locale

toby cabot — Fri, 17 Jan 2003 19:17:26 -0000

Locale doesn't have a no-arg constructor so
ObjectSerializer stubs its toe when it tries to
instantiate it. Here's code to explicitly handle Locale.

Patch to unit tests and the serializer registry, new
serializer class for Locale.

support for serializing exceptions

toby cabot — Mon, 13 Jan 2003 15:08:06 -0000

I've seen problems where Skaringa gets into an infinite
recursion when it tries to marshal certain objects,
specifically Exception and descendants. After some
investigation it appears that the problem is that the
"cause" of the exception is initialized by default to
itself, so when Skaringa goes to serialize the cause it
gets into an infinite recursion (well, not infinite,
only until the stack overflows).

I'm not sure what the real solution is, but I've
implemented a work-around that seems to work well for
the cases that I've seen. Since the circularity is one
object that references itself in one of its members we
can just check that the member that we're about to
marshal isn't "self". Granted, larger loops (a -> b ->
a) will still cause problems, but those loops would
involve a lot more code to check, and marshalling
Exception can be useful.

This patch is:

- a new Exception test file
src/com/skaringa/javaxml/test/ExceptionObj.java
- a patch to ObjectSerializer to check for objects
that reference themselves
- a patch to the test cases

serializers for BigInteger, BigDecimal,

toby cabot — Mon, 30 Dec 2002 20:01:08 -0000

Here's a patch (and some source files) that adds
serializers for java.math.BigInteger,
java.math.BigDecimal, and java.util.Set.

The java.math classes don't have zero-arg constructors
so they seem to make Skaringa unhappy when it tries to
unmarshal them. Set, too causes problems in
unmarshalling because a Set is not necessarily a
Collection.

Note that the JavaXMLTestClass patch here depends on
the bug fixes that I submitted earlier, so you'll
probably want to apply those first.