Обсуждение: time interval behaviour seems odd

Hi,

while trying to create a query this afternoon dealing with intervals, I 
noticed the following behaviour and I wondered, if it is intended. (It was 
non-intuitive to me and if there was a warning in the docs i missed it.)
The lines marked   <---   make me wonder ...

tschwarz=> select version();                               version
------------------------------------------------------------------------PostgreSQL 7.3.2 on i386-pc-linux-gnu, compiled
byGCC gcc (GCC) 3.2.3
 
(1 row)

tschwarz=> select '1'::interval;interval
----------00:00:01
(1 row)

tschwarz=> select -'1'::interval;?column?
------------00:00:01
(1 row)

tschwarz=> select '-1'::interval;interval
-----------01:00        <---
(1 row)

tschwarz=> select -'-1'::interval;?column?
----------01:00        <---
(1 row)


Now with 1.0 everything looks fine:


tschwarz=> select '1.0'::interval;interval
----------00:00:01
(1 row)

tschwarz=> select -'1.0'::interval;?column?
------------00:00:01
(1 row)

tschwarz=> select '-1.0'::interval;interval
------------00:00:01
(1 row)

tschwarz=> select -'-1.0'::interval;?column?
----------00:00:01
(1 row)


But look at 0.1:


tschwarz=> select '0.1'::interval; interval
-------------00:00:00.10
(1 row)

tschwarz=> select -'0.1'::interval;  ?column?
---------------00:00:00.10
(1 row)

tschwarz=> select '-0.1'::interval; interval
-------------00:00:00.10   <---
(1 row)

tschwarz=> select -'-0.1'::interval;  ?column?
---------------00:00:00.10   <---
(1 row)

Tilo Schwarz <mail@tilo-schwarz.de> writes:
> while trying to create a query this afternoon dealing with intervals, I 
> noticed the following behaviour and I wondered, if it is intended.

> tschwarz=> select '-1'::interval;
>  interval
> ----------
>  -01:00        <---
> (1 row)

This appears to be intentional --- the comment in DecodeInterval says
                       /*                        * Only a signed integer? Then must assume a                        *
timezone-likeusage                        */                       type = DTK_HOUR;
 

whereas nearby code selects DTK_SECOND scaling for the cases of
fractional or unsigned numbers.  I'm not sure *why* it's intentional,
and am hesitant to change it without knowing what the rationale was.
There may be cases involving multiple fields that need the existing
behavior...

> tschwarz=> select '-0.1'::interval;
>   interval
> -------------
>  00:00:00.10   <---
> (1 row)

This is incontestably a bug.  Will fix for 7.4.1.
        regards, tom lane

Tom Lane <tgl@sss.pgh.pa.us> writes:
> Tilo Schwarz <mail@tilo-schwarz.de> writes:
>> tschwarz=> select '-0.1'::interval;
>> interval
>> -------------
>> 00:00:00.10   <---
>> (1 row)

> This is incontestably a bug.  Will fix for 7.4.1.

I've applied the attached patch to 7.4 and HEAD.  It'd probably work in
7.3 as well, but I didn't look.
        regards, tom lane

*** src/backend/utils/adt/datetime.c.orig    Sat Nov 29 14:51:58 2003
--- src/backend/utils/adt/datetime.c    Wed Dec 17 16:33:25 2003
***************
*** 3005,3011 ****                     if (*cp != '\0')                         return DTERR_BAD_FORMAT; 
!                     if (val < 0)                         fval = -(fval);                 }                 else if
(*cp== '\0')
 
--- 3005,3011 ----                     if (*cp != '\0')                         return DTERR_BAD_FORMAT; 
!                     if (*field[i] == '-')                         fval = -(fval);                 }
elseif (*cp == '\0')