I spoke to Jonathan about this and it seems your use case may be legal although it is very
unusual.
In most cases, the xid will end it's association with one resource before joining on
the next resource. If you change your test to do the "normal" thing, it passes.
Are you actually seeing this behaviour with a transaction manager?
The XA spec says it permissable to block the second start call until the first has
finished, so I suggest we do that. I don't want to go down the path of having to make
transactions mult-threaded on the server.
View the original post :
http://www.jboss.org/index.html?module=bb&op=viewtopic&p=4217329#...
Reply to the post :
http://www.jboss.org/index.html?module=bb&op=posting&mode=reply&a...