Hi Julian, Thanks for the quick reply, I'll investigate the different approach but I think the third option would be the most extensible (in case I need other rules with similar requirement).
On Tue, Jan 31, 2017 at 5:41 PM, Julian Hyde <jh...@apache.org> wrote: > If the root is the only place it can match you could almost apply the rule > manually (well, you’d have to write your own implementation of > RelOptRuleCall). > > You could customize HepPlanner so that it doesn’t look below the root node. > > Or, you could invent a new RelNode type (a concrete sub-class of SingleRel > would probably do the trick) called Marker and place it at the root node. > Make your match only if a Marker is on top. Since no other rules match > Marker it would remain at the root. > > Julian > > > > On Jan 31, 2017, at 7:32 AM, Renaud Delbru <renaud@siren.solutions> > wrote: > > > > Hi, > > > > In our scenario, we can have a relational expression that is composed of > a > > single logical table function scan operator (i.e., the logical table > > function scan operator if the root of the relational expression). We > would > > like to be able to differentiate this with scenarios where this operator > > appears as a leaf of some other operators. However, I am not able to > find a > > way to properly express this as a rule, e.g., by restricting a rule to > > match only from the root of the relational expression. Or is there > another > > way to achieve this ? > > > > Thanks, > > Regards > > > > -- > > Renaud Delbru > > -- *Renaud Delbru* Siren Solutions - CTO renaud@siren.solutions | http://siren.solutions/ Unit 13 GTC, Mervue Park, Galway, Ireland
