Re: [PATCHES] commit TO_CHAR()

On Wed, 26 Jan 2000, Peter Eisentraut wrote:

> Couldn't this be reduced to functions for timestamp, int4, int8, float8,
> and numeric?

With datetime/timestamp I not sure, exapmle date_part() exist for both.
And how is it with year in timestamp?

test=> select '15000-01-26 21:01:57+01'::timestamp;
?column?
----------
invalid
(1 row)

test=> select '15000-01-26 21:01:57+01'::datetime;
?column?
---------------------------
Sun Jan 26 20:01:57 15000
(1 row)

Now is to_char's timestamp implementation very easy/terrible:

text *timestamp_to_char(time_t dt, text *fmt)
{
return datetime_to_char( timestamp_datetime(dt), fmt);
}

:-) it call my classic datetime to_char(). I want rewrite it (and delete
datetime version?), if I will certain with timestamp and timestamp to
'tm' struct conversion.

Use float8's to_char only:

The to_char() convert it to string itself and must check float number
range (FLT_DIG ..etc) and it is different for float4 and float8, but
don't worry it is only 50 row in C. Yes, float4->float8 cast can make
PG parser, but it is effective if you not need float8? IMHO this solution
reduce function/code, but not reduce performance/memory.

(I re-send this mail to hacker list)
Karel