sc/source/core/tool/interpr2.cxx | 49 +++++++++++++++++++++++++++++++++------
1 file changed, 42 insertions(+), 7 deletions(-)
New commits:
commit f0f2ceaf978a0eee8e3c1fcfe6dec546ab8f07d6
Author: Eike Rathke <er...@redhat.com>
AuthorDate: Sat Apr 15 01:38:39 2023 +0200
Commit: Adolfo Jayme Barrientos <fit...@ubuntu.com>
CommitDate: Thu Apr 20 09:51:17 2023 +0200
Resolves: tdf#138220 tdf#154792 Avoid double rounding; tdf#124286 follow-up
Change-Id: Ie4028b20f2d3087a54bbfafd35d59fa06ec7a061
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/150439
Reviewed-by: Eike Rathke <er...@redhat.com>
Tested-by: Jenkins
(cherry picked from commit 8186a01f2a26f05645a2a3c9c93b453bd35b796f)
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/150505
Reviewed-by: Mike Kaganski <mike.kagan...@collabora.com>
Reviewed-by: Adolfo Jayme Barrientos <fit...@ubuntu.com>
diff --git a/sc/source/core/tool/interpr2.cxx b/sc/source/core/tool/interpr2.cxx
index 7826522a89bf..a0b857defc76 100644
--- a/sc/source/core/tool/interpr2.cxx
+++ b/sc/source/core/tool/interpr2.cxx
@@ -962,20 +962,55 @@ void ScInterpreter::RoundNumber( rtl_math_RoundingMode
eMode )
fVal = ::rtl::math::round( GetDouble(), 0, eMode );
else
{
- sal_Int16 nDec = GetInt16();
- double fX = GetDouble();
+ const sal_Int16 nDec = GetInt16();
+ const double fX = GetDouble();
if (nGlobalError == FormulaError::NONE)
{
+ // 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).
+ constexpr sal_Int16 kSigDig = 12;
+
if ( ( eMode == rtl_math_RoundingMode_Down ||
eMode == rtl_math_RoundingMode_Up ) &&
- nDec < 12 && fmod( fX, 1.0 ) != 0.0 )
+ nDec < kSigDig && fmod( fX, 1.0 ) != 0.0 )
+
{
- // tdf124286 : round to 12 significant digits before rounding
+ // tdf124286 : round to significant digits before rounding
// down or up to avoid unexpected rounding errors
// caused by decimal -> binary -> decimal
conversion
- double fRes;
- RoundSignificant( fX, 12, fRes );
- fVal = ::rtl::math::round( fRes, nDec, eMode );
+
+ double fRes = fX;
+ // Similar to RoundSignificant() but omitting the back-scaling
+ // and interim integer rounding before the final rounding,
+ // which would result in double rounding. Instead, adjust the
+ // decimals and round into integer part before scaling back.
+ const double fTemp = floor( log10( std::abs(fRes))) + 1.0 -
kSigDig;
+ // Avoid inaccuracy of negative powers of 10.
+ if (fTemp < 0.0)
+ fRes *= pow(10.0, -fTemp);
+ else
+ fRes /= pow(10.0, fTemp);
+ if (std::isfinite(fRes))
+ {
+ // fRes is now at a decimal normalized scale.
+ // Truncate up-rounding to opposite direction for values
+ // like 0.0600000000000005 =ROUNDUP(8.06-8;2) that here now
+ // is 600000000000.005 and otherwise would yield 0.07
+ if (eMode == rtl_math_RoundingMode_Up)
+ fRes = ::rtl::math::approxFloor(fRes);
+ fVal = ::rtl::math::round( fRes, nDec + fTemp, eMode );
+ if (fTemp < 0.0)
+ fVal /= pow(10.0, -fTemp);
+ else
+ fVal *= pow(10.0, fTemp);
+ }
+ else
+ {
+ // Overflow. Let our round() decide if and how to round.
+ fVal = ::rtl::math::round( fX, nDec, eMode );
+ }
}
else
fVal = ::rtl::math::round( fX, nDec, eMode );
