Обсуждение: Re: [PATCHES] Re: [HACKERS] Query->hasSubLinks is always FALSE...

Bruce Momjian wrote:
>
> You know, after all our lengthy discussions, this SubLink thing looks
> pretty clean to me.
>
> How are subselects going?  Are you having to write a lot of code to get
> it working?  I would say after all our discussions and thinking, it took

Mmm ... =~ 1000 lines of code :)

> me about two or three days to do the parser and rewrite for sublinks.
> How many coding days is the rest taking?  I am curious.

All was near ready Feb 6, but ... I got problems from canonification
routines in optimizer: 'where A AND B OR C ' becomes 'where (A OR C) AND (B OR C)'
and if C is subselect then ... C has to be processed twice (sometime) ...
I remember your question about handling 'A AND (B OR SubSelect)' - unfortunately,
I didn't see problems here (actually, there is no problem with this
conjunctive normal form :), but there are troubles in common case.

Ideas ?

For the moment I could leave this un-optimized, with possibility
of double subselect processing...

Vadim

>
> Bruce Momjian wrote:
> >
> > You know, after all our lengthy discussions, this SubLink thing looks
> > pretty clean to me.
> >
> > How are subselects going?  Are you having to write a lot of code to get
> > it working?  I would say after all our discussions and thinking, it took
>
> Mmm ... =~ 1000 lines of code :)
>
> > me about two or three days to do the parser and rewrite for sublinks.
> > How many coding days is the rest taking?  I am curious.
>
> All was near ready Feb 6, but ... I got problems from canonification
> routines in optimizer: 'where A AND B OR C ' becomes 'where (A OR C) AND (B OR C)'
> and if C is subselect then ... C has to be processed twice (sometime) ...
> I remember your question about handling 'A AND (B OR SubSelect)' - unfortunately,
> I didn't see problems here (actually, there is no problem with this
> conjunctive normal form :), but there are troubles in common case.
>
> Ideas ?
>
> For the moment I could leave this un-optimized, with possibility
> of double subselect processing...

I would leave it for later.  We already have an optimizer explosion
problem when we have lots of OR's so at some point we may have to
revisit the cnf-ify process anyway.  Commerical Ingres has or had the
same problem.


--
Bruce Momjian
maillist@candle.pha.pa.us

Bruce Momjian wrote:
>
> >
> > > me about two or three days to do the parser and rewrite for sublinks.
> > > How many coding days is the rest taking?  I am curious.
> >
> > All was near ready Feb 6, but ... I got problems from canonification
> > routines in optimizer: 'where A AND B OR C ' becomes 'where (A OR C) AND (B OR C)'
> > and if C is subselect then ... C has to be processed twice (sometime) ...
> > I remember your question about handling 'A AND (B OR SubSelect)' - unfortunately,
> > I didn't see problems here (actually, there is no problem with this
> > conjunctive normal form :), but there are troubles in common case.
> >
> > Ideas ?
> >
> > For the moment I could leave this un-optimized, with possibility
> > of double subselect processing...
>
> I would leave it for later.  We already have an optimizer explosion
> problem when we have lots of OR's so at some point we may have to
> revisit the cnf-ify process anyway.  Commerical Ingres has or had the
> same problem.

Ok, follow this way...

Vadim