>>>>> "Tom" == Tom Lane <tgl@sss.pgh.pa.us> writes:
Tom> However, we need to fix this in all active branches, and I
Tom> definitely agree with minimizing the amount of change to back
Tom> branches. The fact that the minimal change breaks (or exposes an
Tom> oversight in) assign_collations_walker makes it very plausible
Tom> that it will also break somebody's third-party code. If we push
Tom> the API change further we increase the risk of breaking stuff.
Tom> That seems OK in HEAD but not in back branches.
We could minimize the chance of breakage in a back-patched fix by having
query_tree_walker/mutator iterate the windowClause list itself and
invoke the walker only on offset expressions; is it worth it?
Walkers that follow the recommended code structure should be unaffected;
it only shows up in the collations walker because that treats
expressions as the "default" case and tries to explicitly handle all
non-expression nodes.
--
Andrew (irc:RhodiumToad)