Happens in tests:
It looks like MariaDB would do the same (use binlog as a trx coordinator) in this scenario.
Ahh, that makes sense. I thought I was missing something here. So then the real issue is that the assumptions in the assert are not correct and should be removed. I wonder if this should be mentioned upstream?
Thanks for the feedback.
Yes this should be logged upstream and your patch sent by OCA.
Cool, I'll take care of that tomorrow.