>> Erm -- obviously that BEGIN was unnecessary and you see the error >> without it. Anyway, the point I was making is that you don't get an >> error if you set it in postgresql.conf, but you do if you set it in >> other contexts, including the parallel worker startup sequence. I
>> wondered if this might have been a consequence of the parallel startup
>> reordering that happened with commit 6c3c9d41, but nope -- >> RestoreGUCState() was always run inside a transaction.