There's a lot of hard work gone into fixing all the problems we're seeing with sequenced and open types, which is much appreciated. Can we now discuss another set of problems we're seeing with the schema support, concerned with performance issues? Although you may not want to do much detailed work on performance at this stage, with a newish project like Tuscany, there are likely to be some "low-hanging fruit" which could make a lot of difference if picked off.
Performance seems fine with simple schema, but more complex schema such as Atom are taking a surprisingly long time to load. Some of the reasons are apparently: - cyclic dependencies within the schema cause recursive type-checking. - schemas are loaded too many times. I think this is a problem exclusively in the way our code uses Tuscany, but I'll add it to this list in case you think Tuscany may be contributing to it too. - schemas once loaded are not cached. - Tuscany looks in too many places to find the schema, for example http://issues.apache.org/jira/browse/TUSCANY-907. I realise the spec is quite open in the area of how to interpret the schemaLocation value and what to do if it is missing, and perhaps some configuration options are needed here.
