OK,

sorry for getting back to this discussion late.

I modified the test case so that the InvalidItemStateException indeed 
occurs with jackrabbit-core (see attachment).

I think we have consensus that JSR-170 doesn't require a store to behave 
like that; it's truly optional.

Now, with the same test and with jcr2spi (connecting to jackrabbit-core 
through spi2jcr) the InvalidItemStateException is *not* thrown.

This doesn't come as a surprise, because -- as far as I understand -- 
JCR2SPI lacks the necessary information to do so.

It seems to me that the SPI APIshould *allow* stores to expose the same 
behaviour as jackrabbit-core, and the simplest way to achieve that would 
(IMHO) be to allow the SPI implementation to keep state. (*)

Feedback appreciated,

Julian

(*) Right now, when it does keep state, we see TCK failures, as JCR's 
refresh() method is not delegated to the SPI SessionInfo. I think we 
need to fix that.