Just doing a drive-by... On Fri, May 20, 2016 at 3:50 PM, Jeff Davis <pg...@j-davis.com> wrote:
> Old thread link: > > http://www.postgresql.org/message-id/CA+=vxna5_n1q5q5okxc0aqnndbo2ru6gvw+86wk+onsunjd...@mail.gmail.com > > On Thu, Apr 14, 2016 at 1:29 PM, Stephen Frost <sfr...@snowman.net> wrote: > > Jeff > > > > (Reviving an old thread for 2014...) > > > Would you have time to work on this for 9.7..? I came across a > > real-world use case for exactly this capability and was sorely > > disappointed to discover we didn't support it even though there had been > > discussion for years on it, which quite a few interested parties. > > > First, I think the syntax is still implemented in a bad way. Right now > it's part of the OVER clause, and the IGNORE NULLS gets put into the > frame options. It doesn't match the way the spec defines the grammar, > and I don't see how it really makes sense that it's a part of the > frame options or the window object at all. How does the relatively new FILTER clause play into this, if at all? I think we need a need catalog support to say > whether a function honors IGNORE|RESPECT NULLS or not, which means we > also need support in CREATE FUNCTION. > We already have "STRICT" for deciding whether a function processes nulls. Wouldn't this need to exist on the "CREATE AGGREGATE" Rhetorical question: I presume we are going to punt on the issue, since it is non-standard, but what is supposed to happen with a window invocation that ignores nulls but has a non-strict function that returns a non-null on null input? David J.
