Kould
Kould
> It seems that there is something wrong with `HashJoin`, I'm debugging now, I'll merge this first ops, I found the problem. In correlated subqueries, there is no left data...
yep, I made a mistake. This seems to be a feature that has not yet been implemented.
2_00_000 row on case: `select * from t1 where c1 > 500 and c1 < 1000` 2024/3/29 
```rust let fnck_sql = DataBaseBuilder::path("./fncksql_bench") .build() .await?; let guard = pprof::ProfilerGuard::new(100).unwrap(); for _ in 0..10_000 { let _ = fnck_sql.run("select * from t1 where c1 > 500 and c1 <...
I ran the benchmark again, and this pr: https://github.com/KipData/FnckSQL/pull/193 made the performance of `select * from t1 where c1 = 1000` basically the same as sqlite: sqlite: 20us, fncksql: 27us,...
[Skyline-Pruning](https://docs.pingcap.com/zh/tidb/stable/choose-index)
It looks good, but it lacks the test of the rules. Reference: https://github.com/KipData/KipSQL/blob/main/src/optimizer/rule/column_pruning.rs#L166
> When truncate is interrupted by a restart or something else, we may leak some files. We should purge files while opening the region. > > Reopen this issue as...
the problem in ci is similar to https://github.com/tursodatabase/limbo/pull/592#issuecomment-2567306343. executing `make test-compat` locally passes