Re: reducing the footprint of ScanKeyword (was Re: Large writable variables)

On 12/17/18, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> John Naylor <jcnaylor(at)gmail(dot)com> writes:
>> Since PL/pgSQL uses the core scanner, we'd need to use offsets in its
>> reserved_keywords[], too. Those don't change much, so we can probably
>> get away with hard-coding the offsets and the giant string in that
>> case. (If that's not acceptable, we could separate that out to
>> pl_reserved_kwlist.h and reuse the above tooling to generate
>> pl_reserved_kwlist_{offset,string}.h, but that's more complex.)
>
> plpgsql isn't as stable as all that: people propose new syntax for it
> all the time. I do not think a hand-maintained array would be pleasant
> at all.

Okay.

> Also, wouldn't we also adopt this technology for its unreserved keywords,
> too?

We wouldn't be forced to, but there might be other reasons to do so.
Were you thinking of code consistency (within pl_scanner.c or
globally)? Or something else?

If we did adopt this setup for plpgsql unreserved keywords,
ecpg/preproc/ecpg_keywords.c and ecpg/preproc/c_keywords.c would be
left using the current ScanKeyword struct for search. Using offset
search for all 5 types of keywords would be globally consistent, but
it also means additional headers, generated headers, and makefile
rules.

-John Naylor