Re: Sorting out acl fixes

Christopher Kings-Lynne <chriskl(at)familyhealth(dot)com(dot)au> writes:
>> The solution I had in mind was for ALTER OWNER to run through the ACL
>> and replace the old owner ID with the new one wherever the old one
>> appears, in both grantor and grantee positions.

> What about fixing existing bad acls?

When did that get to be part of the requirements? I don't even know
who you expect to do this (backend? pg_dump? user?) or at what level
you think the fixing should happen (GRANT/REVOKE? UPDATE pg_class
SET relacl = fixme(relacl)? direct hacking of the ACL array?). To
say nothing of the semantic problems of deciding what an invalid
ACL is really supposed to mean.

I'll be satisfied if ALTER OWNER does not transform a valid
configuration into an invalid one. Right now it fails to meet that
minimal requirement. Considering we are weeks past feature freeze,
I don't want to get into inventing a magic wand that can fix existing
breakage automatically.

regards, tom lane