@mrab
gah! I could swear I originally had the _odict_clear_node first and had switched them due to a segfault. It even crossed my mind on Friday but I didn't pursue it. I'm guessing I did put the _odict_clear_node call first at some point but lost that fix along the way. :( That's the "benefit" of having a patch languish for so long; you end up working on it from multiple hosts and sometimes lose bits and pieces. Unfortunately only recently did I think to create a feature repo on which to track the on-going work.
Anyway, thanks for helping with the investigation. The patch should be just about ready to commit at this point. :) I'm going to give it a once over, check for any lingering ref leaks, and double-check with Raymond. So I'm hopeful we can land this in the next few days.