Support uncorrelated EXISTS #14474
Support uncorrelated EXISTS #14474
Conversation
For the purpose of decorrelation, an uncorrelated plan is a unit. No verification needs to be performed on it.
Extract variable from a long if condition involving a match. Improves readability.
Handle the unhandled case returning immediately. This adds additional return point to the function, but removes subsequent if. At the point of this additional return we know why we bail out (some unhandled situation), later the None filter could be construed as a true condition.
github-actions
bot
added
logical-expr
| fn check_no_outer_references(inner_plan: &LogicalPlan) -> Result<()> { | ||
| if inner_plan.contains_outer_reference() { | ||
| plan_err!( | ||
| "Accessing outer reference columns is not allowed in the plan: {}", |
There was a problem hiding this comment.
I thought it was possible (in theory) to have a correlated subuery in Exists (which would access outder columns)
So in this case, the error might be more clear as a not_impl_err! with "Referencing outer columns (correlated queries) not yet supported: {}" or something
There was a problem hiding this comment.
(although I see this check is simply moved around in this PR, not introduced)
It would also be great if you were able to add a (negative) test for such queries
There was a problem hiding this comment.
(although I see this check is simply moved around in this PR, not introduced)
indeed
It would also be great if you were able to add a (negative) test for such queries
i think the check in analyzer is unsound, i wouldn't want to spend time on it
this is because decorrelability is a pretty complex thing. we have two separate classes/files devoted to it, and we're only scratching the surface (#11773). I understand the desire is to produce nice error messages, but the analyzer cannot, in its brevity, know which query can run and which query cannot. For example, #13523 added as allowed operator in the analyzer, without adding any decorrelation support for it. Thus, the check was either incorrect before or after the change (or, likely, both).
There was a problem hiding this comment.
edit: above the important negation was missed (fixed now)
Support uncorrelated EXISTS
check_correlations_in_subqueryshould not check uncorrelated plan nodes #14461