Thank you for your reply. I confirm that this particular bug is gone with the new tarball. I even tested it with the original large CTE request and it worked correctly. Still, there is a regression in a test case involving the same CTE but with the wholenumber virtual table instead of generate_series. In that test case, the query enters into an infinite loop as the limits on wholenumber generated values inside some parts of the CTE seem to not be respected. It is as if the WHERE clause is completely ignored in some chunks of the CTE which look like this: ``` SELECT value FROM integers WHERE value < 10 ``` I am still investigating this particular issue as it looks to be related to shared cursors in materialized views (according to my quick understanding of the new SQLite code) and to query flattening which is not working as before. I am also checking the source code of both virtual tables to see if there are any differences in cursor handling that may explain this behavior. I'll keep you updated about my findings.