po 15. 3. 2021 v 18:18 odesÃlatel Julien Rouhaud <rjuju...@gmail.com>
napsal:
> On Mon, Mar 15, 2021 at 06:05:52PM +0100, Pavel Stehule wrote:
> >
> > Possibility to work with a parser is one main reason for forking
> postgres.
> > Lot of interestings projects fail on the cost of maintaining their own
> fork.
> >
> > Maybe a good enough possibility is the possibility to inject an own
> parser
> > called before Postgres parser. Then it can do a transformation from
> "CREATE
> > PIPE ..." to "SELECT extparse("CREATE PIPE()". There can be a switch if
> > returned content is string for reparsing or already prepared AST.
>
> Having a hook that returns a reformatted query string would definitely be
> easier to write compared to generating an AST, but the overhead of parsing
> the
> query twice plus deparsing it will probably make that approach way too
> expensive in many usecases, so we shouldn't go that way.
>
yes - so it can be nice to have more possibilities.
parsing is expensive - but on today computers, the cost of parsing is low -
the optimization is significantly more expensive.
I wrote some patches in this area (all rejected by Tom :)), and a lot of
work can be done after parser and before the analysis stage. Probably, the
parser hook is not good enough, there should be an analysis stage hook too.
