On Wed, Jan 30, 2019 at 09:41:04AM -0500, Tom Lane wrote:
> I'd definitely take this on a case-by-case basis. In the planner
> functions you mentioned, for instance, I'd be pretty hesitant to remove
> the "root" parameter even if it happens not to be needed today.
> We'd probably just end up putting it back in the future, because almost
> everything in the planner needs that. I'd only consider removing it in
> cases where there was a solid reason to require the function not to need
> it ever (as for instance what I just did to flatten_join_alias_vars).
Definitely agreed, this is a case-by-case. For the callbacks and
hooks it makes no sense, the planner ones and a couple of layers like
shared memory calculation are just here for symmetry, so most of the
report is really noise (I deliberately discarded anything related to
bison and generated code of course).
> In cases where we can get simplifications of calling layers, and
> it doesn't seem likely that we'd have to undo it in future, then
> probably it's worth the trouble to change.
Some of these are in tablecmds.c, and visibly worth the trouble. The
ones in the btree, gin and gist code also could do for some cleanup at
quick sight. And these are places where we complain a lot about the
complexity of the code.
--
Michael