Обсуждение: Re: Creating constraint dynamically

sivapostgres@yahoo.com schreef op ma 22-08-2022 om 07:29 [+0000]:
> 
> ALTER TABLE public.tx_barcode_stock 
> ADD CONSTRAINT "tx_barcode_stock_CK1" CHECK 
> ( (branchcode = '1'::bpchar  and barcodeitem = 'Y'::bpchar and
> closingstock >= 0::numeric)  Or (branchcode = '1' and barcodeitem =
> 'N'::bpchar and closingstock >= 0::numeric )  Or (branchcode =
> '2'::bpchar  and barcodeitem = 'Y'::bpchar and closingstock >=
> 0::numeric)  Or (branchcode = '2' and  barcodeitem = 'N'::bpchar and
> closingstock >= 0::numeric ) ) NOT VALID; 
> 
> After creation, when we check what we find is  [ in PgAdmin ]
> ALTER TABLE public.tx_barcode_stock
>     ADD CONSTRAINT "tx_barcode_stock_CK1" CHECK (branchcode::bpchar =
> '1'::bpchar AND barcodeitem = 'Y'::bpchar AND closingstock >=
> 0::numeric OR branchcode::text = '1'::text AND barcodeitem =
> 'N'::bpchar AND closingstock >= 0::numeric OR branchcode::bpchar =
> '2'::bpchar AND barcodeitem = 'Y'::bpchar AND closingstock >=
> 0::numeric OR branchcode::text = '2'::text AND barcodeitem =
> 'N'::bpchar AND closingstock >= 0::numeric)
>     NOT VALID;
> 
> We have only one bracket, in the final updated one.
> 
> Since there are AND and OR conditions, without brackets the whole
> conditions becomes useless. 

did you see
https://www.postgresql.org/docs/current/sql-syntax-lexical.html#SQL-PRECEDENCE
?


mvg,
Wim

On 8/22/22 13:09, Wim Bertels wrote:
> sivapostgres@yahoo.com schreef op ma 22-08-2022 om 07:29 [+0000]:
>> ALTER TABLE public.tx_barcode_stock
>> ADD CONSTRAINT "tx_barcode_stock_CK1" CHECK
>> ( (branchcode = '1'::bpchar  and barcodeitem = 'Y'::bpchar and
>> closingstock >= 0::numeric)  Or (branchcode = '1' and barcodeitem =
>> 'N'::bpchar and closingstock >= 0::numeric )  Or (branchcode =
>> '2'::bpchar  and barcodeitem = 'Y'::bpchar and closingstock >=
>> 0::numeric)  Or (branchcode = '2' and  barcodeitem = 'N'::bpchar and
>> closingstock >= 0::numeric ) ) NOT VALID;
>>
>> After creation, when we check what we find is  [ in PgAdmin ]
>> ALTER TABLE public.tx_barcode_stock
>>      ADD CONSTRAINT "tx_barcode_stock_CK1" CHECK (branchcode::bpchar =
>> '1'::bpchar AND barcodeitem = 'Y'::bpchar AND closingstock >=
>> 0::numeric OR branchcode::text = '1'::text AND barcodeitem =
>> 'N'::bpchar AND closingstock >= 0::numeric OR branchcode::bpchar =
>> '2'::bpchar AND barcodeitem = 'Y'::bpchar AND closingstock >=
>> 0::numeric OR branchcode::text = '2'::text AND barcodeitem =
>> 'N'::bpchar AND closingstock >= 0::numeric)
>>      NOT VALID;
>>
>> We have only one bracket, in the final updated one.
>>
>> Since there are AND and OR conditions, without brackets the whole
>> conditions becomes useless.
> did you see
> https://www.postgresql.org/docs/current/sql-syntax-lexical.html#SQL-PRECEDENCE
> ?
>
>
> mvg,
> Wim

You can use plpgsql function
alter table tale_name add constraint xyz  check (func(table_name))

On Mon, 22 Aug 2022 at 09:29, sivapostgres@yahoo.com
<sivapostgres@yahoo.com> wrote:
> We populate constraint string dynamically and add it to the table with alter table command.  It gets added, but
withoutthe required brackets.
 

Brackets are NOT required in your example. You are doing a classic
form, OR of ANDs, sometime called sum of products. And has greater
precedence than OR ( amongst other things because this form of
conditions is used a lot ), as it has been pointed, so brackets (
round brackets are usually called parentheses abbreviated to parens,
in programming brackets is normaly nor used for those ) are not
necessary. What the system is doing is parsing and normalizing your
condition ( note how it changes bpchar to text and add types in some
places ) and reconstructing the condition, without brackets because
they are not required.

> Since there are AND and OR conditions, without brackets the whole conditions becomes useless.

Before stating these things, test, try to insert a condition violating
row to see if the condition is really useless, it is very rare to find
a bug ( changing the semantics of the condition would be one ) like
these.

> How to create a constraint like the above one, with braces in tact ?   Or any other way that we can implement a check
constraintas above?
 

The condition IS correctly implemented, but nothing guarantees you the
text returned by pgadmin will be the exact same condition you sent. In
fact, AFAIK, nothing guarantees you can recover a condition set on a
column. It is transformed to an equivalent. I'm not sure how it is
exactly done, but it probably also does whitespace normalization and
constant folding.

Francisco Olarte.