> On 6 Mar 2023, at 21:45, Gregory Stark (as CFM) <stark.cfm@gmail.com> wrote:
>
> So.... This patch has been through a lot of commitfests. And it really
> doesn't seem that hard to resolve -- Pavel has seemingly been willing
> to go along whichever way the wind has been blowing but honestly it
> kind of seems like he's just gotten drive-by suggestions and he's put
> a lot of work into trying to satisfy them.
Agreed.
> He implemented --include-tables-from-file=... etc. Then he implemented
> a hand-written parser for a DSL to select objects, then he implemented
> a bison parser, then he went back to the hand-written parser.
Well, kind of. I was trying to take the patch to the finishing line but was
uncomfortable with the hand written parser so I implemented a parser in Bison
to replace it with. Not that hand-written parsers are bad per se (or that my
bison parser was perfect), but reading quoted identifiers across line
boundaries tend to require a fair amount of handwritten code. Pavel did not
object to this version, but it was objected to by two other committers.
At this point [0] I stepped down from trying to finish it as the approach I was
comfortable didn't gain traction (which is totally fine).
Downthread from this the patch got a lot of reviews from Julien with the old
parser back in place.
> Can we get some consensus on whether the DSL looks right
I would consider this pretty settled.
> and whether the hand-written parser is sensible.
This is the part where a committer who wants to pursue the hand-written parser
need to step up. With the amount of review received it's hopefully pretty close.
--
Daniel Gustafsson
[0] 098531E1-FBA9-4B7D-884E-0A4363EEE6DF@yesql.se