Hi,
On 2018-12-22 12:20:00 -0500, Tom Lane wrote:
> John Naylor <jcnaylor@gmail.com> writes:
> > Using a single file also gave me another idea: Take value and category
> > out of ScanKeyword, and replace them with an index into another array
> > containing those, which will only be accessed in the event of a hit.
> > That would shrink ScanKeyword to 4 bytes (offset, index), further
> > increasing locality of reference. Might not be worth it, but I can try
> > it after moving on to the core scanner.
>
> I like that idea a *lot*, actually, because it offers the opportunity
> to decouple this mechanism from all assumptions about what the
> auxiliary data for a keyword is.
OTOH, it doubles or triples the number of cachelines accessed when
encountering a keyword. The fraction of keywords to not-keywords in SQL
makes me wonder whether that makes it a good deal.
Greetings,
Andres Freund
