David Rowley <dgrowleyml@gmail.com> writes:
> Aside from that, I don't have any ideas on how to get rid of the
> possible additional datumCopy() from non-Var arguments to these window
> functions. Should we just suffer it? It's quite likely that most
> arguments to these functions are plain Vars anyway.
No, we shouldn't. I'm pretty sure that we have various window
functions that are deliberately designed to take advantage of the
no-copy behavior, and that they have taken a significant speed
hit from your having disabled that optimization. I don't say
that this is enough to justify reverting the chunk header changes
altogether ... but I'm completely not satisfied with the current
situation in HEAD.
regards, tom lane