adriangb commented on PR #12978: URL: https://github.com/apache/datafusion/pull/12978#issuecomment-2423916764
Okay @alamb I pushed a pretty big rework. Lots of new test cases, lots of comments explaining what's going on. I removed the `not like` part; I'm thinking this is complex enough as is and most of the benefit (maybe even in ClickBench?) will come from `like`. We can tackle `not like`, `ilike` and `not ilike` in future PRs. Especially since it's going to be important to evaluate each test case carefully. I will note that I am a bit concerned about the interaction of truncated stats and how we apply these filters. Take the (absurd) case of stats that were truncated so that all you have is `"",""`. You basically know _nothing_ about the data, there could be anything in there. Yet `col = 'A'` transforms into `'' <= 'A' and 'A' <= ''` which is false. Substitute in non-truncated stats and `'A' <= 'A' and 'A' <= 'Z'` is true. I added a test case for this behavior on the existing code. It doesn't differentiate between `"ABC"` truncated to `""` and `""` actually being the min string but it shows the behavior which would be the same in both cases. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
