In reviewing one of my patches[1], Fujii-san has pointed out that I
didn't include checks for being in recovery, or for working on another
backend's temporary index.
I think that brin_summarize_new_values in 9.5.0 commits those same
sins. In its case, I don't think those are critical, as they just
result in getting less specific error messages that one might hope
for, rather than something worse like a panic.
But still, we might want to address them.
Cheers,
Jeff
[1] http://www.postgresql.org/message-id/CAHGQGwH=M1BAEJPQdpJjCneqwg8Xa+P8SB+ZsvhVwH6gL2J46w@mail.gmail.com