Обсуждение: Precedence of new phrase search tsquery operator

It appears that the new <-> operator has been made to have exactly the
same grammatical precedence as the existing & (AND) operator.  Thus,
for example, 'a & b <-> c'::tsquery means something different from
'b <-> c & a'::tsquery:

regression=# select 'a & b <-> c'::tsquery;             tsquery              
-----------------------------------( 'a' <-> 'c' ) & ( 'b' <-> 'c' )
(1 row)

regression=# select 'b <-> c & a'::tsquery;       tsquery        
-----------------------( 'b' <-> 'c' ) & 'a'
(1 row)

I find this surprising.  My intuitive feeling is that <-> ought to
bind tighter than & (and therefore also tighter than |).  What's
the reasoning for making it act like this?
        regards, tom lane

On Wed, Jun 8, 2016 at 7:13 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> It appears that the new <-> operator has been made to have exactly the
> same grammatical precedence as the existing & (AND) operator.  Thus,
> for example, 'a & b <-> c'::tsquery means something different from
> 'b <-> c & a'::tsquery:
>
> regression=# select 'a & b <-> c'::tsquery;
>               tsquery
> -----------------------------------
>  ( 'a' <-> 'c' ) & ( 'b' <-> 'c' )
> (1 row)
>
> regression=# select 'b <-> c & a'::tsquery;
>         tsquery
> -----------------------
>  ( 'b' <-> 'c' ) & 'a'
> (1 row)
>
> I find this surprising.  My intuitive feeling is that <-> ought to
> bind tighter than & (and therefore also tighter than |).  What's
> the reasoning for making it act like this?

I don't remember, but it looks like a bug. I found another issue with that

If some dictionary returns two infinitives, like:

select * from to_tsquery('en','leavings');     to_tsquery
----------------------'leavings' | 'leave'
(1 row)


then following query looks like a bug

select to_tsquery('en', 'aa & leavings <-> tut');                           to_tsquery
-------------------------------------------------------------------( 'aa' <-> 'tut' ) & ( 'leavings' <-> 'tut' |
'leave'<-> 'tut' )
 
(1 row)

It should be definitely

select to_tsquery('en', 'aa & leavings <-> tut');                           to_tsquery
------------------------------------------------------------------- 'aa'  & ( 'leavings' <-> 'tut' | 'leave' <-> 'tut'
)
(1 row)

so, yes, <-> should be more tight than &.

>
>                         regards, tom lane

On Wed, Jun 8, 2016 at 7:13 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> It appears that the new <-> operator has been made to have exactly the
> same grammatical precedence as the existing & (AND) operator.  Thus,
> for example, 'a & b <-> c'::tsquery means something different from
> 'b <-> c & a'::tsquery:
>
> regression=# select 'a & b <-> c'::tsquery;
>               tsquery
> -----------------------------------
>  ( 'a' <-> 'c' ) & ( 'b' <-> 'c' )
> (1 row)
>
> regression=# select 'b <-> c & a'::tsquery;
>         tsquery
> -----------------------
>  ( 'b' <-> 'c' ) & 'a'
> (1 row)
>
> I find this surprising.  My intuitive feeling is that <-> ought to
> bind tighter than & (and therefore also tighter than |).  What's
> the reasoning for making it act like this?

ah, now we remember :)   The idea about equivalence of  & and <->
operators appeared in situation when <-> degenerates to & in case of
absence of positional information. Looks like we mixed different
things, will fix.

>
>                         regards, tom lane

> On Wed, Jun 8, 2016 at 7:13 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> It appears that the new <-> operator has been made to have exactly the
>> same grammatical precedence as the existing & (AND) operator.  Thus,
>> for example, 'a & b <-> c'::tsquery means something different from
>> 'b <-> c & a'::tsquery:
>> I find this surprising.  My intuitive feeling is that <-> ought to
>> bind tighter than & (and therefore also tighter than |).  What's
>> the reasoning for making it act like this?
>
> ah, now we remember :)   The idea about equivalence of  & and <->
> operators appeared in situation when <-> degenerates to & in case of
> absence of positional information. Looks like we mixed different
> things, will fix.

Attached patch changes a precedences of operations to |, &, <->, | in ascending
order. BTW, it simplifies a bit a code around printing and parsing of tsquery.

--
Teodor Sigaev                                   E-mail: teodor@sigaev.ru
                                                    WWW: http://www.sigaev.ru/

> Attached patch changes a precedences of operations to |, &, <->, | in ascending
> order. BTW, it simplifies a bit a code around printing and parsing of tsquery.

|, &, <->, ! of course
-- 
Teodor Sigaev                                   E-mail: teodor@sigaev.ru
  WWW: http://www.sigaev.ru/