Thread: Possible bug in parsing

Possible bug in parsing

From
Hannu Krosing
Date:
Can anyone explain the following:

hannu=> select 1 + 2 + 3 + 4;
?column?
--------
      10
(1 row)

works fine, but:

hannu=> select '1' || '2' || '3';
ERROR:  parser: parse error at or near "||"
hannu=> select ('1' || '2') || '3';
NOTICE:  there is more than one operator || for types
NOTICE:  unknown and unknown. You will have to retype this query
ERROR:  using an explicit cast
hannu=> select ('1'::text || '2') || '3';
?column?
--------
     123
(1 row)

I understand the 'there is more than one operator' errors, but
what is so different between operators + and || ?

Hannu


Re: [HACKERS] Possible bug in parsing

From
"Thomas G. Lockhart"
Date:
> > > I understand the 'there is more than one operator' errors, but
> > > what is so different between operators + and || ?
>
> Yes I know that, but why are they so different that I can have
>  ARG1 op ARG2 op ARG3
> with + but not with ||

Ah! I *thought* that was a silly question, and as usual that's because I
didn't understand the question :)

They behave differently because the "+" operator is allowed to be
"left-associative", since its behavior is well-defined mathematically,
while the "||" operator falls into a broad class of multi-character
operators in the parser which are "non-associative". So, the parser does
not know whether to evaluate left-to-right or right-to-left and throws
an error instead.

> I actually found this while trying to find out how characters in
> strings are escaped and if \0 is supported in them.
>
> Is there some docs on escaping characters, and if not then where do
> you think it should go in docs ?

I believe that there are no docs on escaping characters and I would
suggest that docs for it should go into the user's guide in the section
on character data types (since it applies to all of them). That would be
in doc/src/sgml/datatype.sgml, probably in a "Sect2" at the end of the
"Sect1" on "Character Types" and just before the "Sect1" on "Date/Time
Types".

If you find it difficult to work with that source document, you can just
type the information and I can help to integrate it into the doc.

btw, I think that the escape conventions for strings need some work; the
fact that the characters are re-escaped when output makes it easy to do
a dump/reload but sort of defeats the usefulness of escaping anything in
the first place.

                           - Tom

Re: [HACKERS] Possible bug in parsing

From
Hannu Krosing
Date:

Thomas G. Lockhart wrote:

> > I understand the 'there is more than one operator' errors, but
> > what is so different between operators + and || ?
>
> ?? They are different operators, that's why. The "+" is addition, and
> the "||" is defined in SQL92 to be string concatenation.
>
>

Yes I know that, but why are they so different that I can have

  ARG1 op ARG2 op ARG3

with + but not with ||

Hannu




Re: [HACKERS] Possible bug in parsing

From
Hannu Krosing
Date:
Thomas G. Lockhart wrote:

> > I understand the 'there is more than one operator' errors, but
> > what is so different between operators + and || ?
>
> ?? They are different operators, that's why. The "+" is addition, and
> the "||" is defined in SQL92 to be string concatenation.

I actually found this while trying to find out how characters in strings
are escaped
and if \0 is supported in them.

Is there some docs on escaping characters, and if not then where do you
think it should go in docs ?

Hannu




Re: [HACKERS] Possible bug in parsing

From
"Thomas G. Lockhart"
Date:
> I understand the 'there is more than one operator' errors, but
> what is so different between operators + and || ?

?? They are different operators, that's why. The "+" is addition, and
the "||" is defined in SQL92 to be string concatenation.

                        - Tom