TreeCache 2.0 [CVS] - Bug in optimistic locking mode?

[chasta]

Hi,
I'm talking about the regular Cache (TreeCache), not PojoCache.

It seems that if I try to use the Node interface directly (after getting some sub-node from the Cache interface), in optimistic locking mode, a simple node.put(k, v) will not make a subsequent node.get(k) return v.

However, if I work against the Cache interface directly, it seems to work fine.

So...
1) What do you say? Is this true, or am I missing something?
2) Can I at least rely on node additions (e.g. addChild()) to reflect immediately?
3) If I want to iterate on the node contents (presumably using Node.getData() ) - and make sure the recent changes are reflected - how can I accomplish that? The Cache interface does not offer me an alternative for that...

[manik]

Yes, there were a few issues with the Node interfaces in ALPHA1. ALPHA2, which should be out very soon, has a number of changes in this department.

[chasta]

Great to hear that! However, I'm talking about the latest version from CVS (as of today) - so assuming the ALPHA2 changes have already been committed, this issue persists in the updated version...

Are these changes indeed yet to be committed?

[manik]

Yes, they have. If you got these off CVS today, then it should be fixed. Will have a look.

[manik]

I see your problem. Note that this is specifically when checked within the same transaction. If you don't use transactions (I.e., allow optimistic locking to create implicit transactions) the problem isn't there.

Added some more unit tests to check this. See NodeAPITest and it's subclass, NodeAPIOptimisticTest. Two specific tests depict this: testAddingData() and testAddingDataTx().

[manik]

And regarding your comment about adding children, the bug exists there too (see testGetChildrenUnderTx() in the same test classes)

[manik]

FYI, I've fixed these and will check them into CVS after running a few more regressions.

[chasta]

Great! Thank you for the quick round-up.

I've taken a look at NodeAPITest (and NodeAPIOptimisticTest) and indeed, the problem is well tested there.

[manik]

No worries.

Have you tried a new build from CVS? Does it solve your problem?