[Bug 154792] Calc Round Down is Rounding Up for some values
https://bugs.documentfoundation.org/show_bug.cgi?id=154792 --- Comment #27 from b. --- @erAck, you are sure about 8.23456789013 in the roundup unit-tests? Didn't run the tests, just looked into the code and consider them questionable, think 8.23456789014 fits better. And about the concept, while you accept / provoke contradictions to 'math' in 'digits behind 12' you'll - IMHO - always have cases which fail 'left of that' as the devias propagate left in chains of '000...' or '999...' . And pls. don't answer it has to be like that, it's not a consequence of math, only partly a consequence of IEEE, and mostly resulting from Calc's combination of decisions and algorithms. -- You are receiving this mail because: You are the assignee for the bug.
[Bug 154792] Calc Round Down is Rounding Up for some values
https://bugs.documentfoundation.org/show_bug.cgi?id=154792 --- Comment #26 from b. --- I seem to remember that a while ago someone tried to fix errors in rounddown and roundup by preceding them with a 'roundsig' to 12 significant digits. Looks as if Eike kept that in his patches, then you may be a victim of the mathematically questionable quality of such measures. Feeding known inaccurate results of subtractions - cancellation - into other formulas and then bending these to ( try to ) produce wanted results for some cases has - IMHO - similar quality. -- You are receiving this mail because: You are the assignee for the bug.
[Libreoffice-bugs] [Bug 154792] Calc Round Down is Rounding Up for some values
https://bugs.documentfoundation.org/show_bug.cgi?id=154792 BogdanB changed: What|Removed |Added CC||buzea.bog...@libreoffice.or ||g Blocks||108252 Summary|Calc Round Down is Rounding |Calc Round Down is Rounding |Up for some values. |Up for some values Referenced Bugs: https://bugs.documentfoundation.org/show_bug.cgi?id=108252 [Bug 108252] [META] Cell-related bugs and enhancements (including formatting) -- You are receiving this mail because: You are the assignee for the bug.
[Libreoffice-bugs] [Bug 154792] Calc Round Down is Rounding Up for some values.
https://bugs.documentfoundation.org/show_bug.cgi?id=154792 --- Comment #25 from Eike Rathke --- *** Bug 157082 has been marked as a duplicate of this bug. *** -- You are receiving this mail because: You are the assignee for the bug.
[Libreoffice-bugs] [Bug 154792] Calc Round Down is Rounding Up for some values.
https://bugs.documentfoundation.org/show_bug.cgi?id=154792 Eike Rathke changed: What|Removed |Added Assignee|er...@redhat.com|libreoffice-b...@lists.free ||desktop.org --- Comment #24 from Eike Rathke --- We do not state a precision higher than 12 significant decimal digits. In fact this is on purpose here, see the change's source code comment: // A quite aggressive approach with 12 significant digits. // However, using 14 or some other doesn't work because other // values may fail, like =ROUNDDOWN(2-5E-015;13) would produce // 2 (another example in tdf#124286). If Excel does not exhibit a round-off _in this case_ then fine for Excel. Otherwise see https://en.wikipedia.org/wiki/Numeric_precision_in_Microsoft_Excel Your screen captures tell nothing by the way, they're different numbers and no indication what actually "fails". If you insist on reopening this bug again, I'm deassigning myself. If you want someone to entirely rewrite the numeric overall behaviour to implement arbitrary-precision arithmetic then this bug is not the place for that. -- You are receiving this mail because: You are the assignee for the bug.
[Libreoffice-bugs] [Bug 154792] Calc Round Down is Rounding Up for some values.
https://bugs.documentfoundation.org/show_bug.cgi?id=154792 Eike Rathke changed: What|Removed |Added Assignee|libreoffice-b...@lists.free |er...@redhat.com |desktop.org | Status|NEW |ASSIGNED Summary|Calc Round Down is Rounding |Calc Round Down is Rounding |Up |Up for some values. --- Comment #9 from Eike Rathke --- There is no decimal representation in binary floating point hence any decimal "nth digit" does not exist for values other than multiples of powers of two. Period. (In reply to NOYB from comment #8) > If the user want's it rounded down to 2 digits then 0.93 > is the correct answer/result for a value of 0.9395. It is not if that 0.9395 is a result of the calculation of 8.94-8 because the exact number 0.94 is not representable in binary floating point and the nearest value is 0.939946709294817992486059665679931640625 and the user expects a ROUNDDOWN(8.94-8;2) to be ROUNDDOWN(0.94;2) to equal 0.94 (which it never is). I don't say we can't have a better implementation, I just state the complexities. -- You are receiving this mail because: You are the assignee for the bug.
