Could the failing MSVC tinderbox be due to implementation-defined /

...for negative integers in < C++11?  Rather unlikely, but lets see...

Change-Id: I9abfcbf2c0e409fab4c77b62e5f734d3c2cc2719

1 files changed, 22 insertions, 2 deletions

diff --git a/sal/rtl/strtmpl.cxx b/sal/rtl/strtmpl.cxx
index 6859af15b31e..455a6aebd540 100644
--- a/sal/rtl/strtmpl.cxx
+++ b/sal/rtl/strtmpl.cxx
@@ -941,8 +941,28 @@ namespace {
             bNeg = sal_False;
         }
 
-        const T nDiv = bNeg ? -(std::numeric_limits<T>::min()/nRadix) : (std::numeric_limits<T>::max()/nRadix);
-        const sal_Int16 nMod = bNeg ? -(std::numeric_limits<T>::min()%nRadix) : (std::numeric_limits<T>::max()%nRadix);
+        T nDiv;
+        sal_Int16 nMod;
+        if ( bNeg )
+        {
+            nDiv = std::numeric_limits<T>::min() / nRadix;
+            nMod = std::numeric_limits<T>::min() % nRadix;
+            // Cater for C++03 implementations that round the quotient down
+            // instead of truncating towards zero as mandated by C++11:
+            if ( nMod > 0 )
+            {
+                --nDiv;
+                nMod -= nRadix;
+            }
+            nDiv = -nDiv;
+            nMod = -nMod;
+        }
+        else
+        {
+            nDiv = std::numeric_limits<T>::max() / nRadix;
+            nMod = std::numeric_limits<T>::max() % nRadix;
+        }
+
         while ( *pStr )
         {
             nDigit = rtl_ImplGetDigit( IMPL_RTL_USTRCODE( *pStr ), nRadix );