Adding a test case for ancestor-or-self along the lines of the existing one for ancestor showed runaway, memory-eating behaviour. In changing CollectResults to remove the livelock, it seemed simplest to remove the extra yield of the root from MoveNextCore. I was unable to ascertain a good reason for producing this element in this special way compared to the AncestorIterator.
According to http://www.w3.org/TR/xpath#predicates, a predicate [n] should be equivalent to [position()=n] for a number expression n. However, we currently have a PredicateIterator intepretation of [n] that has been corrected for the issue outlined in the bug report, leaving the position() function behind with an inappropriate, document-order value for reverse axes. Note that [n] is not treated by parsing or transforming to the same form as [position()=n] before evaluation, valid as that would be.
The discussion in that same section of the recommendation about forward and reverse axes and further remarks in http://www.w3.org/TR/xpath#node-sets make it clear that the predicate [1] has the sense of "nearest" rather than "first in the document" which then, depending on the axis type, may be the same or reverse of the document order. It does not seem to me to be so clear whether the order given by a navigator for a location path expression should mirror this or always give the document order. The .Net implementation seems to always give the document order, as well as predicates having the correct reverse/forward senses, so I have added test cases that explore both.
The old code cause incorrect duplicate ID errors when XmlSchemas is involved
(global components were not resolved correctly and it's not simple to make
minor fix).
This is part of bugfix #670945.
* XmlAnyElementAttributes.cs:
* XmlArrayItemAttributes.cs:
* XmlElementAttributes.cs:
Under Silverlight API this implement IList only (does not
inherit from CollectionBase).
* XmlCustomFormatter.cs:
* XmlSerializationWriterInterpreter.cs:
Use a Convert.ChangeType override that exists in SL API
* XmlReflectionImporter.cs: Provide an alternative for (missing in SL)
Enum.GetNames and select available (in SL) overloads for parsing and
formatting for enums
* XmlSerializationReader.cs: Fix ToEnum API to use IDictionary (in SL)
instead of Hashtable (even if the SL4 docs still states the later)
* XmlSerializationReaderInterpreter.cs:
* XmlTypeMapping.cs:
Use the default (identical to nonPublic being false) when using
Activator.CreateInstance under SL
* XmlSerializationWriter.cs: Internally use a Queue<T> instead a a Queue
(not available in SL). Also Namespaces (ArrayList) is replaced by
XmlNamespaces (IList) in SL API.
* XmlSerializer.cs: Do not use environment variables in MOONLIGHT (like
the rest of MOBILE). Use Encoding.UTF8 instead of ENcoding.Default since
the later does not exist in SL.
* XmlSerializerImplementation.cs: SL version use IDictionary instead of
Hashtable for properties.
* XmlSerializerNamespaces.cs: Internally use a generic Dictionary
* instead
of a ListDictionary (which is not available in SL)
Non-nullable web service parameters can be made optional by
adding an additional parameter with the same name and a
'Specified' suffix. This was supported when serializing
classes, but not when serializing web service parameters.
This might fix bug 688496.
Namespaces in nested external schemas have to be validated against
immediately-containing schema. If there is a schema A with urn:x
targetNamespace imports a schema B with urn:y targetNamespace which
in turn includes another schema C, then C must be validated its namespace
against B, not A.