Intended behavior of automatic decimal point (bug 120940)

[John Ralls]

> On Jul 27, 2017, at 6:27 PM, Eric Siegerman <pub08-gnc at davor.org> wrote:
> 
> On Thu, Jul 27, 2017 at 08:20:50AM +0000, David T. via gnucash-devel wrote:
>> I think of the decimal placement as applying to the final number in the field
>> (as a sort of edit mask, if you will), rather than a preprocessing function
>> that would apply to every element in an equation.
> 
> I'm not sure that would quite work either.
> 
> Currently, for simple numbers with no arithmetic, "1000" gets
> auto-decimal-pointed ("scaled" hereafter), but "4.50" doesn't,
> which are both just what one wants.  The same should apply in
> formulas (I think! -- but more about that at the end).  Assuming
> two auto-decimal places, consider:
>    1000 + 4.50
> 
> I (think I) want the first term to get scaled, but not the
> second, giving a result of 14.50.
> 
> OK, so how about we scale each term separately, so that:
>    1000 * 3 + 450 -> 34.50
> but also:
>    1000 * 3 + 4.50 -> 34.50
> ("->" meaning "yields a result of", since "=" just looks wrong
> under the circumstances :-) ).
> 
> But then:
>    10.00 * 3 + 4.50 -> 34.50
> We didn't want to scale the first term after all.
> 
> I've thought of a couple of different approaches:
>  - scale each term's resulting value if the term only contains
>    integers:
>        1000*3 + 4000   -> 30 + 40      = 70.00
>        1000*3 + 4000.  -> 30 + 4000    = 4030.00
>        1000*3. + 4000  -> 3000 + 40    = 3040.00
>        1000*3. + 4000. -> 3000 + 4000  = 7000.00
> 
>  - scale each term's *first* number if it's an integer,
>    but never second or subsequent numbers:
>        1000 * 3   -> 30
>        1000 * 3.  -> 30
>        1000. * 3  -> 3000
>        1000. * 3. -> 1000
>    This is based on the thought that ($20 * $3) is meaningless;
>    it only makes sense to multiply money by something that isn't
>    money
> 
> But neither of those works in all situations.
> 
> 
> The easiest way out, I think, is to never scale formulas at all,
> only simple numbers.  So:
>    4000   -> 40.00     # as currently happens
>    40.    -> 40.00     # likewise
> But:
>    4000+1 -> 4001.00
> 
> That's how my truly ancient copy of Excel behaves.  (I don't
> have access to a modern one.)
> 
> 
> Or perhaps: for formulas, scale the final result (as you say),
> but only if *all* of the numeric values the user typed are
> integers:
>    1000*3 + 4000   -> 70.00
>    1000*3 + 4000.  -> 7000.00
>    1000*3. + 4000  -> 7000.00
>    1000.*3 + 4000  -> 7000.00
> 
> That could boil down to:
>    Scale the final result unless the original input string
>    contains any "."s (or ","s depending on locale)
> (without even any need to worry whether it's a number or
> a formula).
> 
> But given that it's not entirely clear how even a simple:
>    1000 + 4.50
> should behave, anything with any subtlety at all is going to want
> a fair amount of testing to see whether people actually find it
> usable.  So an unsubtle approach like "never scale formulas" is
> probably the safest place to start.

I agree that the only sane way to have auto-decimal is to disable it if the input is a formula. The other sane approach is to remove it completely.

Regards,
John Ralls