Martijn van Oosterhout <kleptog(at)svana(dot)org> writes:
> About the only thing in the backend I found interesting was this:
> src/backend/utils/hash/dynahash.c function hash_create
I wonder if we shouldn't just remove the hash_destroy calls in
hash_create's failure paths. hash_destroy is explicitly not gonna
work on a shared-memory hashtable, and in all other cases I'd expect
that any already-allocated table structure will be in a palloc context
that will get cleaned up during error recovery.
regards, tom lane