Hello,
I think this demonstrates a bug, tested in 11.5:
https://gist.github.com/wkalt/a298fe82c564668c803b3537561e67a0
The same script succeeds if the index on line 11 is either dropped, made to
be non-partial on b, or shifted to a different column (the others are used
in the partitioning; maybe significant).
This seems somewhat related to
https://www.postgresql.org/message-id/flat/CAFjFpRc0hqO5hc-%3DFNePygo9j8WTtOvvmysesnN8bfkp3vxHPQ%40mail.gmail.com#00ca695e6c71834622a6e42323f5558a
.
Regards,
Wyatt