On 1/4/26 12:21 PM, David Rowley wrote:
I'd say this is wishful thinking. You're saying that with the overhead of planning, executor startup, most of executor run and executor shutdown and subsequent cleanup, you're expecting a speedup from swapping the order of operations in this function when you're calling it once!? That could be true if the function was very expensive, but this is not an expensive function. It'll amount to a few dozen instructions for the function out of several 10s of millions, when you consider the cost of processing the entire query.
On top of this the idea the scalars are a less common thing is dubious. Why would that be? I would assume it depends on the input data. So even if it had been a speedup, which perf output does not seem to indicate that it would be, it would have slowed down some workloads while speeding up others.
Andreas
