It appears to be a PostgreSQL bug from my reading. The return code causing the problem has always been legitimate, even though you never expected to see it in practice.

I did a quick check of the source of two other storage engines that run on both Linux and macOS to see how common this oversight is. They both had code for handling this case correctly for years. It was probably never used, but it was implemented because the syscall docs allowed for the possibility.

This is the kind of thing I would have assumed PostgreSQL handles correctly as a matter of pedantry. Given how syscall heavy PostgreSQL is, it will be a fair amount of work to fix it.