Hi, Here are my rants on the subject: 1) http://mail-archives.apache.org/mod_mbox/calcite-dev/201411.mbox/%3CCAB%3DJe-GTi1-dABpzxvpVmaDjmEuLwePgRURTQhP9vhk0KSqeaQ%40mail.gmail.com%3E
2) Currently Calcite tries to fully de-correlate the queries. In other words, Calcite "never" produces a "nested loops" execution plan. Well, it does support Correlate plans (it can even execute ones), but typically you want cost-based de-correlation. For instance, it would be nice if Calcite could compare "hashjoin(fullscan, fullscan)" vs "nestedloops(fullscan, indexscan)" on a cost basis. I have not yet moved in the direction, however I hope I reach that goal. Are you going to join hangout on the 13th May 10am pacific? It could be a good discussion topic. Vladimir
