Hi,
On 2023-11-21 12:52:35 +0900, Amit Langote wrote:
> version gram.o text bytes %change gram.c bytes %change
>
> 9.6 534010 - 2108984 -
> 10 582554 9.09 2258313 7.08
> 11 584596 0.35 2313475 2.44
> 12 590957 1.08 2341564 1.21
> 13 590381 -0.09 2357327 0.67
> 14 600707 1.74 2428841 3.03
> 15 633180 5.40 2495364 2.73
> 16 653464 3.20 2575269 3.20
> 17-sqljson 672800 2.95 2709422 3.97
>
> So if we put SQL/JSON (including JSON_TABLE()) into 17, we end up with a gram.o 2.95% larger than v16, which granted
isa somewhat larger bump, though also smaller than with some of recent releases.
I think it's ok to increase the size if it's necessary increases - but I also
think we've been a bit careless at times, and that that has made the parser
slower. There's probably also some "infrastructure" work we could do combat
some of the growth too.
I know I triggered the use of the .c bytes and text size, but it'd probably
more sensible to look at the size of the important tables generated by bison.
I think the most relevant defines are:
#define YYLAST 117115
#define YYNTOKENS 521
#define YYNNTS 707
#define YYNRULES 3300
#define YYNSTATES 6255
#define YYMAXUTOK 758
I think a lot of the reason we end up with such a big "state transition" space
is that a single addition to e.g. col_name_keyword or unreserved_keyword
increases the state space substantially, because it adds new transitions to so
many places. We're in quadratic territory, I think. We might be able to do
some lexer hackery to avoid that, but not sure.
Greetings,
Andres Freund