Why is infinite_recurse test suddenly failing?

In the past week, four different buildfarm members have shown
non-reproducing segfaults in the "select infinite_recurse()"
test case, rather than the expected detection of stack overrun
before we get to the point of a segfault.

https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=bonito&dt=2019-05-01%2023%3A05%3A36
https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=takin&dt=2019-05-01%2008%3A16%3A48
https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=buri&dt=2019-04-27%2023%3A54%3A46
https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=demoiselle&dt=2019-04-27%2014%3A55%3A52

They're all on HEAD, and they all look like

2019-05-01 23:11:00.145 UTC [13933:65] LOG: server process (PID 17161) was terminated by signal 11: Segmentation fault
2019-05-01 23:11:00.145 UTC [13933:66] DETAIL: Failed process was running: select infinite_recurse();

I scraped the buildfarm database and verified that there are no similar
failures for at least three months back; nor, offhand, can I remember ever
having seen this test fail in many years. So it seems we broke something
recently. No idea what though.

(Another possibility, seeing that these are all members of Mark's PPC64
flotilla, is that there's some common misconfiguration --- but it's hard
to credit that such a problem would only affect HEAD not the back
branches.)

Thoughts?

regards, tom lane