I'm really sorry, David. For some reason I overlooked that part. Thanks.
The following bug has been logged on the website:
Bug reference: 14717
Logged by: Paolo Guagliardo
PostgreSQL version: 9.6.3
Operating system: Scientific Linux release 7.3 (Nitrogen)
Description:
[...]
SELECT (1,NULL)::pair = (1,NULL)::pair ;
?column?
----------
t
(1 row)
while the answer should be NULL (unknown) as before.
Working as documented.
See the last two sections for why (or at least how...) "anonymous row constructor" and "named record" are treated differently.
David J.