> On Thu, Mar 18, 2021 at 04:50:02PM +0100, Dmitry Dolgov wrote:
> > On Thu, Mar 18, 2021 at 09:38:09AM -0400, David Steele wrote:
> > On 1/5/21 10:51 AM, Zhihong Yu wrote:
> > >
> > > + int lastExprLenght = 0;
> > >
> > > Did you mean to name the variable lastExprLenghth ?
> > >
> > > w.r.t. extracting to helper method, the second and third
> > > if (currentExprIdx == pgss_merge_threshold - 1) blocks are similar.
> > > It is up to you whether to create the helper method.
> > > I am fine with the current formation.
> >
> > Dmitry, thoughts on this review?
>
> Oh, right. lastExprLenghth is obviously a typo, and as we agreed that
> the helper is not strictly necessary I wanted to wait a bit hoping for
> more feedback and eventually to post an accumulated patch. Doesn't make
> sense to post another version only to fix one typo :)
Hi,
I've prepared a new rebased version to deal with the new way of
computing query id, but as always there is one tricky part. From what I
understand, now an external module can provide custom implementation for
query id computation algorithm. It seems natural to think this machinery
could be used instead of patch in the thread, i.e. one could create a
custom logic that will enable constants collapsing as needed, so that
same queries with different number of constants in an array will be
hashed into the same record.
But there is a limitation in how such queries will be normalized
afterwards — to reduce level of surprise it's necessary to display the
fact that a certain query in fact had more constants that are showed in
pgss record. Ideally LocationLen needs to carry some bits of information
on what exactly could be skipped, and generate_normalized_query needs to
understand that, both are not reachable for an external module with
custom query id logic (without replicating significant part of the
existing code). Hence, a new version of the patch.