Hi ZizhuanLiu,
Thanks for the update and the spreadsheet. That analysis is helpful. I agree with the narrowed conclusion full MCV scanning only looks justified when the column collation is deterministic and the expression uses a non deterministic collation. In the other cases, keeping the current first match behavior seems right. If you post a revised patch that does the full scan only in that case, I am happy to retest it. Thanks, Damil Shahzad On Wed, 5 Aug 2026 at 15:37, Zizhuan Liu <[email protected]> wrote: > >That would double the function's runtime on average, without changing > >the results at all in most cases (it could only be different if the > >given operator has different semantics from the equality operator used > >while building the statistics list). > > So I agree with Tom’s reasoning.
