https://bugs.documentfoundation.org/show_bug.cgi?id=146367

Eike Rathke <[email protected]> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Version|7.2.4.1 release             |7.2.0.0.beta1+
                 CC|                            |[email protected]

--- Comment #8 from Eike Rathke <[email protected]> ---
This is what you get when repeatedly adding binary floating point values.
See also https://erack.de/bookmarks/D.html#010203

Note also what happens to the displayed values when column E is widened.
Fwiw, Gnumeric quite agrees (with marginal different values) in that results
are not 0 as it does not try to eliminate accuracy problems.

Note that only changing the display format to a rounding one just hides the
proliferation of accuracy errors. In fact you have to round all intermediate
values in column E starting from the bottom to make this work, like in E72 
=ROUND(SUM(C72;E73;A72);2)  and copy&paste that up. Or instead of SUM() use the
+ operator that is implemented to try to tie values to 0 for marginal minimal
accuracy differences, like  =C72+E73+A72  and copy&paste that up. 

I also see no change in 7.2.4, I get the same display values in 7.2.0; however,
I do see the difference compared with 7.1.8 and earlier.

I assume this is due to the Kahan summation algorithm being implemented for
SUM() and some other functions taking cell ranges, see
https://wiki.documentfoundation.org/ReleaseNotes/7.2#Calc and bug 137679, where
previously SUM() for the first and one subsequent value did something similar
as the + operator.

Maybe we can add that back as a special case for single values like they are
used here.

-- 
You are receiving this mail because:
You are the assignee for the bug.

Reply via email to