Re: [HACKERS] current_time?

Vince Vielhaber <vev(at)michvhf(dot)com> writes:
> Now I thought this was discussed recently and this:
> create table foo(
> x int,
> y datetime default current_time);
> would put the current date and time into y whenever a new record was
> inserted. It appears to give the date and time the stupid table was
> created. Is it me or is something broke?

The behavior for this was changed very recently. Since current sources
refuse the above:

regression=> create table foo(
regression-> x int,
regression-> y datetime default current_time);
ERROR: Attribute 'y' is of type 'datetime' but default expression is of type 'time'
You will need to rewrite or cast the expression

I am guessing you are trying it with 6.5.*, where indeed you will likely
get the time of table creation. Recommended approach is
y datetime default now()
which works the way you want in all Postgres versions AFAIK.

Next question is whether current sources are broken to refuse the above.
Since I get

regression=> create table zz (x datetime);
CREATE
regression=> insert into zz values(current_time);
ERROR: Attribute 'x' is of type 'datetime' but expression is of type 'time'
You will need to rewrite or cast the expression

it seems I managed to make default-expression handling consistent
with the rest of the system, but that doesn't necessarily mean this
behavior is desirable... Thomas, what say you?

regards, tom lane