Comment 5 for bug 185066

I'm backing out my fix because, as you point out, it doesn't go far enough. Calling the base deactivate isn't appropriate because there are some types, especially persistent classes, that can't be deactivated.

The current implementation badly mixes up _p_deactivate and _p_invalidate. Worse, there are some persistent objects that seem to rely, somewhat, on the default implementation of _p_invalidate calling _p_deactivate. :( I think this can be sorted out, but doing so is too risky for 3.9.