SQL timestamp duration can't be trusted?

[Vern Hamberg <vhamberg@xxxxxxxxxxxxxxxxxxxxxxxxx>]

I want to get the duration in seconds between 2 timestamps in SQL. I have 
looked at the duration from subtracting 2 timestamps. But this does not 
seem a reliable way to do this, when leap years Feb 29 are involved. Or, 
actually, when relatively long periods are involved - 2 months can be 
anything from 59 -62 days.

The result of subtracting 2 dates is a DEC(8,0) formatted as yyyymmdd.

SELECT date('2004-02-29') - date('2003-02-28') FROM qsqptabl
   gives 00010001 (1 year + 1 day)

SELECT date('2005-02-28') - date('2004-02-28') FROM qsqptabl
   gives 00010000 (1 year)

SELECT date('2005-02-28') - date('2003-02-28') FROM qsqptabl
   gives 00020000 (2 years, no accounting for the extra day)

SELECT date('2005-02-28') - date('2004-02-29') FROM qsqptabl
   gives 00001128 (11 months + 28 days)

Makes me wonder whether durations in RPG suffer from the same problem.

Is there a way (without writing a UDF) to get the exact seconds between 
timestamps, in SQL? In an HLL I could use the CEESECS API.

Thanks

Vern