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

On 12/18/18, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> I'd be kind of inclined to convert all uses of ScanKeyword to the new way,
> if only for consistency's sake. On the other hand, I'm not the one
> volunteering to do the work.

That's reasonable, as long as the design is nailed down first. Along
those lines, attached is a heavily WIP patch that only touches plpgsql
unreserved keywords, to test out the new methodology in a limited
area. After settling APIs and name/directory bikeshedding, I'll move
on to the other four keyword types.

There's a new Perl script, src/common/gen_keywords.pl, which takes
pl_unreserved_kwlist.h as input and outputs
pl_unreserved_kwlist_offset.h and pl_unreserved_kwlist_string.h. The
output headers are not installed or symlinked anywhere. Since the
input keyword lists will never be #included directly, they might be
better as .txt files, like errcodes.txt. If we went that far, we might
also remove the PG_KEYWORD macros (they'd still be in the output
files) and rename parser/kwlist.h to common/core_kwlist.txt. There's
also a case for not changing things unnecessarily, especially if
there's ever a new reason to include the base keyword list directly.

To keep the other keyword types functional, I had to add a separate
new struct ScanKeywordOffset and new function
ScanKeywordLookupOffset(), so the patch is a bit messier than the
final will be. With a 4-byte offset, ScankeyWordOffset is 8 bytes,
down from 12, and is now a power of 2.

I used the global .gitignore, but maybe that's an abuse of it.

Make check passes, but I don't know how well it stresses keyword use.
I'll create a commitfest entry soon.

-John Naylor