Tests to show getAllLinearVersions() being 'stuck' on first value seen

[fazy]

Using Jackalope-Jackrabbit it seems there is a problem with the version history response: after the first call to getAllLinearVersions(), subsequent calls always return the same value as the first call, even if the history has now changed.

The same problem doesn't seem to happen with getAllVersions(). I've created test cases (using new nodes for each) to show the working, getAllVersions(), and failing, getAllLinearVersions(), cases.

See https://groups.google.com/d/msg/symfony-cmf-devs/--3J2hbMsa8/bupTUgqeb-wJ

[dbu]

ok this is quite obviously wrong in jackalope: https://github.com/jackalope/jackalope/blob/master/src/Jackalope/Version/VersionHistory.php#L62 - we just cache and do never invalidate. i wonder how we would best invalidate. we should invalidate on every checkin/checkpoint operation. the tricky part is that if versioning is cascaded to children, so its not straight forward what to invalidate.

[fazy]

I think the difference between handling linearVersions and versions is the notifyHistoryChanged() method here: https://github.com/jackalope/jackalope/blob/master/src/Jackalope/Version/VersionHistory.php#L243

Jackalope\Version\VersionManager calls that method from checkin() and restore(). Can it be the same for linear versions?

[dbu]

ah so we already do invalidate! yes indeed, there is no reason not to kill the cache of linear versions as well, that should fix it. can you try and do a PR on jackalope if it fixes the problem?

btw the existing code does not handle the situation when creating versions is cascaded by the node type. for this we would need another method similar to ObjectManager::getCachedNode that gets all potential children of that node, by looping through the array and comparing the path string. but we do not need to fix that in the same moment, can also create an issue and fix it when somebody finds time to do that.

[fazy]

Ok, PR sent - jackalope/jackalope#148

I haven't looked at the cascading side of things.