> Hm, I'm not following? setup_depend runs in initdb, that is on the > client side. It can't invoke backend-internal functions any other > way than via SQL, AFAICS.
Ah, brainfade on my part.
I was also curious about the test case where Andres fixed a regression in the parent thread [1], and there is a noticeable improvement (lowest of 10 measurements):
HEAD: 623ms patch: 567ms
If no one else has anything, I think this is ready for commit.