Re: [office-formula] day count basis tests

[robert_weir@us.ibm.com]

"Andreas J. Guelzow" <andreas.guelzow@concordia.ab.ca> wrote on 10/03/2010 
02:08:07 PM:

> 
> Re: [office-formula] day count basis tests
> 
> I think there is a small problem in procedure A:
> 
> 2. If both date1 and date2 are the last day of February, change both
> dates to the 30th of the month.
> 3. If date1 is the last day of February, change it to the 30th of the
> month.
> 
> Since we are changing date1 in (2) we may be creating a non-existing
> date, so 3 might not make sense. I suggest:
> 
> 2. If both date1 and date2 are the last day of February, change date2 to
> the 30th of the month.
> 3. If date1 is the last day of February, change it to the 30th of the
> month.
> 
> This change has no net effect since we are changing date1 in 3 if we
> would have changed it in 2.
>

That makes sense to me.  I've updated the proposal in JIRA.

 
> 
> The file basis-tests.ods uses equality to test whether an evaluator
> calculates the right value. Gnumeric fails twice with a difference of
> 10^{-16}. That should not be flagged as failing.
> 

Anyone who compares floating point numbers is, of course, in a state of 
sin.  So I better do something about this.

I'll adjust the sheet to look for the absolute value of the difference 
between the expected and calculated values and compare that to a 
user-settable epsilon value.

The only exception I think are the cases where date1 == date2, where the 
spec requires them to be exactly 0.

-Rob