Of course getentropy would be better. But the current mechanism is not wrong or broken: at best, it's inconvenient. And it's certainly no excuse for the LibreSSL authors to write a library that calls raise(SIGKILL) on file descriptor exhaustion. That behavior, in many cases, amounts to a remote DoS. As long as this code is in the library (even if off by default), I'm hesitant to recommend LibreSSL.
Without a way to getentropy(2) [hint] that doesn't use file descriptors, it has no other secure choice but to raise(SIGKILL) in my opinion; a mere error might be overlooked, but continuing to run could expose secrets and keys, which is much worse than a DoS condition (anything in file-descriptor exhaustion when under attack is already being DoSsed). (It's turned off because coredumps could also do that locally.)
It's behind a define so there's no problem, don't turn it on if you don't like it. You'd never execute that code anyway, because you're the smart admin who knows better and always has the right devices in all the chroots. And who cares about other users! If they don't know better, they're doing it wrong. Put the blame on them. There is no problem.
